{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Problems Following System Update\", \"message\": \"Multiple products faced outages potentially due to recent system updates. An initial rollback was performed, and monitoring is currently in progress.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Problems Following System Update\", \"message\": \"Multiple products faced outages potentially due to recent system updates. An initial rollback was performed, and monitoring is currently in progress.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Returns and Exchanges\": \"Returns and Exchanges\", \"Beauty & Fitness/Fitness Training\": \"Beauty & Fitness/Fitness Training\", \"Service Outages and Maintenance\": \"Service Outages and Maintenance\", \"Shopping/E-commerce\": \"Shopping/E-commerce\", \"Books & Literature/Non-Fiction\": \"Books & Literature/Non-Fiction\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Request\": \"Request\", \"Change\": \"Change\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Request for Digital Strategies to Enhance Brand Expansion\", \"message\": \"Could you share insights on how digital strategies can promote brand development among clients? I am eager to understand your methodology.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Request for Digital Strategies to Enhance Brand Expansion\", \"message\": \"Could you share insights on how digital strategies can promote brand development among clients? I am eager to understand your methodology.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Online Communities/Forums\": \"Online Communities/Forums\", \"Customer Service\": \"Customer Service\", \"Science/Research\": \"Science/Research\", \"Hobbies & Leisure/Crafts\": \"Hobbies & Leisure/Crafts\", \"Shopping/E-commerce\": \"Shopping/E-commerce\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nOne of these vids should end with [NAME]."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"One of these vids should end with [NAME].\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Software Crash Issue\", \"message\": \"Our data analytics software is experiencing crashes. Incompatibility due to recent hardware or software updates could be the reason.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Software Crash Issue\", \"message\": \"Our data analytics software is experiencing crashes. Incompatibility due to recent hardware or software updates could be the reason.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Technical Support\": \"Technical Support\", \"Pets & Animals/Veterinary Care\": \"Pets & Animals/Veterinary Care\", \"Shopping/E-commerce\": \"Shopping/E-commerce\", \"Real Estate\": \"Real Estate\", \"Games\": \"Games\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhen will I get my new card?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"When will I get my new card?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"verify_top_up\": \"verify top up\", \"getting_virtual_card\": \"getting virtual card\", \"receiving_money\": \"receiving money\", \"card_arrival\": \"card arrival\", \"top_up_by_card_charge\": \"top up by card charge\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Technical Glitch on Our Data Analysis Platform Today\", \"message\": \"Our data analytics platform encountered a technical glitch. It is suspected that this might be due to incompatible software updates. The system was restarted and updates were verified.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Technical Glitch on Our Data Analysis Platform Today\", \"message\": \"Our data analytics platform encountered a technical glitch. It is suspected that this might be due to incompatible software updates. The system was restarted and updates were verified.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Sales and Pre-Sales\": \"Sales and Pre-Sales\", \"Finance/Personal Finance\": \"Finance/Personal Finance\", \"Product Support\": \"Product Support\", \"Food & Drink/Groceries\": \"Food & Drink/Groceries\", \"Real Estate\": \"Real Estate\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHow's the exchange rate?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"How's the exchange rate?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"declined_transfer\": \"declined transfer\", \"exchange_rate\": \"exchange rate\", \"lost_or_stolen_card\": \"lost or stolen card\", \"balance_not_updated_after_cheque_or_cash_deposit\": \"balance not updated after cheque or cash deposit\", \"top_up_failed\": \"top up failed\", \"visa_or_mastercard\": \"visa or mastercard\", \"pending_card_payment\": \"pending card payment\", \"request_refund\": \"request refund\", \"top_up_reverted\": \"top up reverted\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThis needs to go viral! Please RT! #BuildThatWall"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"This needs to go viral! Please RT! #BuildThatWall\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nsome caps fans *really* love jumping to conclusions without any basis whatsoever"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"some caps fans *really* love jumping to conclusions without any basis whatsoever\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Crash Issue with Uploads\", \"message\": \"The application is crashing during project uploads, which is likely due to compatibility issues with Django 3.2. Despite attempts to update libraries and reset configurations, the problem continues to occur.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Crash Issue with Uploads\", \"message\": \"The application is crashing during project uploads, which is likely due to compatibility issues with Django 3.2. Despite attempts to update libraries and reset configurations, the problem continues to occur.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Science/Research\": \"Science/Research\", \"Technical Support\": \"Technical Support\", \"Human Resources\": \"Human Resources\", \"Pets & Animals/Pet Services\": \"Pets & Animals/Pet Services\", \"Real Estate\": \"Real Estate\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Problem\": \"Problem\", \"Change\": \"Change\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWell stop drinking bovine growth hormone then."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Well stop drinking bovine growth hormone then.\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Support Needed for Aruba Switch Integration\", \"message\": \"I am attempting to incorporate TensorFlow and Scikit-learn with an Aruba 2530 Switch for analytics purposes. Could you offer guidance on how to set up this integration? Any documentation or tutorials to help initiate the process would be highly beneficial. Since this is a new type of integration for me, I want to ensure everything is configured correctly. Thank you for your support.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Support Needed for Aruba Switch Integration\", \"message\": \"I am attempting to incorporate TensorFlow and Scikit-learn with an Aruba 2530 Switch for analytics purposes. Could you offer guidance on how to set up this integration? Any documentation or tutorials to help initiate the process would be highly beneficial. Since this is a new type of integration for me, I want to ensure everything is configured correctly. Thank you for your support.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Billing and Payments\": \"Billing and Payments\", \"Product Support\": \"Product Support\", \"Games\": \"Games\", \"Food & Drink/Groceries\": \"Food & Drink/Groceries\", \"Beauty & Fitness/Cosmetics\": \"Beauty & Fitness/Cosmetics\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI might need a new card because it's not working at any of the ATMs. I don't think ive used all my money. Why is it doing this?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I might need a new card because it's not working at any of the ATMs. I don't think ive used all my money. Why is it doing this?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"balance_not_updated_after_bank_transfer\": \"balance not updated after bank transfer\", \"cancel_transfer\": \"cancel transfer\", \"getting_spare_card\": \"getting spare card\", \"unable_to_verify_identity\": \"unable to verify identity\", \"declined_cash_withdrawal\": \"declined cash withdrawal\", \"fiat_currency_support\": \"fiat currency support\", \"verify_top_up\": \"verify top up\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nCan I use this card at any ATM?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Can I use this card at any ATM?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"get_disposable_virtual_card\": \"get disposable virtual card\", \"transfer_into_account\": \"transfer into account\", \"failed_transfer\": \"failed transfer\", \"virtual_card_not_working\": \"virtual card not working\", \"atm_support\": \"atm support\", \"verify_my_identity\": \"verify my identity\", \"card_not_working\": \"card not working\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI keep thinking its gonna be a green day tomorrow but then i remember that it will be a white day so i cry"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I keep thinking its gonna be a green day tomorrow but then i remember that it will be a white day so i cry\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nPro-Mass Immigration &amp; Open Boarders Billionaire Koch Brothers Threaten to Support Democrats #StopTheInvasion v"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Pro-Mass Immigration &amp; Open Boarders Billionaire Koch Brothers Threaten to Support Democrats #StopTheInvasion v\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI did not do this payment"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I did not do this payment\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"card_linking\": \"card linking\", \"country_support\": \"country support\", \"card_payment_not_recognised\": \"card payment not recognised\", \"Refund_not_showing_up\": \"Refund not showing up\", \"passcode_forgotten\": \"passcode forgotten\", \"atm_support\": \"atm support\", \"change_pin\": \"change pin\", \"getting_spare_card\": \"getting spare card\", \"virtual_card_not_working\": \"virtual card not working\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAnd yet, it’s bad to manipulate people. Which is what the in laws are doing."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"And yet, it’s bad to manipulate people. Which is what the in laws are doing.\", \"questions\": {\"emotion_disgust\": {\"type\": \"noul\", \"instructions\": \"Does the author express disgust?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n>I *am* fascinated that some people call themselves incels while acknowledging that they aren't virgins. oh, then you should check out femcels."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \">I *am* fascinated that some people call themselves incels while acknowledging that they aren't virgins. oh, then you should check out femcels.\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nalways pull the gay card cause some exgirlfriends are fucking nuts 4th one in a week"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"always pull the gay card cause some exgirlfriends are fucking nuts 4th one in a week\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n\"My dad is going to see Ed Sheeran tomorrow in concert, I don't try to understand my parents anymore\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"\\\"My dad is going to see Ed Sheeran tomorrow in concert, I don't try to understand my parents anymore\\\"\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHow come I can see an extra charge for my transfer?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"How come I can see an extra charge for my transfer?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"get_disposable_virtual_card\": \"get disposable virtual card\", \"beneficiary_not_allowed\": \"beneficiary not allowed\", \"transfer_fee_charged\": \"transfer fee charged\", \"lost_or_stolen_phone\": \"lost or stolen phone\", \"exchange_rate\": \"exchange rate\", \"card_not_working\": \"card not working\", \"pending_top_up\": \"pending top up\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThese animals should be removed from our country immediately! #NODACA #BUILDTHEWALL!!!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"These animals should be removed from our country immediately! #NODACA #BUILDTHEWALL!!!\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAs high as ~72% in black families. Truly heartbreaking what the expansion of the welfare state has done to the black family."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"As high as ~72% in black families. Truly heartbreaking what the expansion of the welfare state has done to the black family.\", \"questions\": {\"emotion_disappointment\": {\"type\": \"noul\", \"instructions\": \"Does the author express disappointment?\"}, \"emotion_sadness\": {\"type\": \"noul\", \"instructions\": \"Does the author express sadness?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIn regards to my card payment, the exchange rate is incorrect."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"In regards to my card payment, the exchange rate is incorrect.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"compromised_card\": \"compromised card\", \"card_about_to_expire\": \"card about to expire\", \"country_support\": \"country support\", \"extra_charge_on_statement\": \"extra charge on statement\", \"topping_up_by_card\": \"topping up by card\", \"card_payment_wrong_exchange_rate\": \"card payment wrong exchange rate\", \"direct_debit_payment_not_recognised\": \"direct debit payment not recognised\", \"transfer_not_received_by_recipient\": \"transfer not received by recipient\", \"pin_blocked\": \"pin blocked\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI lost my phone. Can I disable to app?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I lost my phone. Can I disable to app?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"transfer_not_received_by_recipient\": \"transfer not received by recipient\", \"cash_withdrawal_not_recognised\": \"cash withdrawal not recognised\", \"lost_or_stolen_phone\": \"lost or stolen phone\", \"topping_up_by_card\": \"topping up by card\", \"request_refund\": \"request refund\", \"receiving_money\": \"receiving money\", \"unable_to_verify_identity\": \"unable to verify identity\", \"card_arrival\": \"card arrival\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI feel like I'm going through withdrawals but again I'm jus being a whiny ass bitch ass skank ass brat"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I feel like I'm going through withdrawals but again I'm jus being a whiny ass bitch ass skank ass brat\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhy is this my fucking friend [NAME] 😂"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Why is this my fucking friend [NAME] 😂\", \"questions\": {\"emotion_curiosity\": {\"type\": \"noul\", \"instructions\": \"Does the author express curiosity?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user @user Fucking bitch ! Shes doing that since 2 years when she give money to STUDENTS, STUPID HOE ! Now laugh in silence and poverty ! U want attention and like bcse its trendy ton talk shit about her ??"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user @user Fucking bitch ! Shes doing that since 2 years when she give money to STUDENTS, STUPID HOE ! Now laugh in silence and poverty ! U want attention and like bcse its trendy ton talk shit about her ??\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Assistance Required for Data Integration Delays\", \"message\": \"Dear Customer Support Team,<br><br>I am writing to report a problem with data integration delays faced by a financial organization. These delays are hindering analytical processes and decision-making, thereby impacting overall business functions.<br><br>The issue became evident as data pipelines failed to update on schedule. We are currently investigating and suspect that the problem may be linked to interruptions in AWS services. We have attempted to restart the data pipelines and conducted virus scans using ESET NOD32, but the issue remains unresolved.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Assistance Required for Data Integration Delays\", \"message\": \"Dear Customer Support Team,<br><br>I am writing to report a problem with data integration delays faced by a financial organization. These delays are hindering analytical processes and decision-making, thereby impacting overall business functions.<br><br>The issue became evident as data pipelines failed to update on schedule. We are currently investigating and suspect that the problem may be linked to interruptions in AWS services. We have attempted to restart the data pipelines and conducted virus scans using ESET NOD32, but the issue remains unresolved.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Home & Garden/Home Improvement\": \"Home & Garden/Home Improvement\", \"Travel & Transportation/Land Travel\": \"Travel & Transportation/Land Travel\", \"Travel & Transportation/Air Travel\": \"Travel & Transportation/Air Travel\", \"Health/Mental Health\": \"Health/Mental Health\", \"IT Support\": \"IT Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Application Crash Issue\", \"message\": \"Dear Customer Support,\\\\n\\\\nI am reporting frequent crashes of the application during video calls. The potential cause might be conflicts arising from multiple hardware devices integrated with the system. I have already attempted to resolve the issue by restarting the application, updating device drivers, and clearing cache data, but the problem continues. I would appreciate it if you could investigate this matter and offer a solution at your earliest convenience. The repeated crashes are disrupting my ability to conduct seamless video calls. I have verified that my system meets the necessary requirements.\\\\n\\\\nThank you for your assistance.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Application Crash Issue\", \"message\": \"Dear Customer Support,\\\\n\\\\nI am reporting frequent crashes of the application during video calls. The potential cause might be conflicts arising from multiple hardware devices integrated with the system. I have already attempted to resolve the issue by restarting the application, updating device drivers, and clearing cache data, but the problem continues. I would appreciate it if you could investigate this matter and offer a solution at your earliest convenience. The repeated crashes are disrupting my ability to conduct seamless video calls. I have verified that my system meets the necessary requirements.\\\\n\\\\nThank you for your assistance.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Food & Drink/Groceries\": \"Food & Drink/Groceries\", \"Customer Service\": \"Customer Service\", \"Law & Government/Government Services\": \"Law & Government/Government Services\", \"News\": \"News\", \"Technical Support\": \"Technical Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Incident\": \"Incident\", \"Request\": \"Request\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nYes dude, exactly them who didn't call the pi that would have ended the game."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Yes dude, exactly them who didn't call the pi that would have ended the game.\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThis is true, but he's perceived as white by most liberals."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"This is true, but he's perceived as white by most liberals.\", \"questions\": {\"emotion_approval\": {\"type\": \"noul\", \"instructions\": \"Does the author express approval?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAussie Directioners. Do you want to win a prize of a lifetime? Then listen to @user on Monday at 7pm."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Aussie Directioners. Do you want to win a prize of a lifetime? Then listen to @user on Monday at 7pm.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nCurtis Painter was asked in interviews today if as the lead singer of Nirvana was getting the band back together."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Curtis Painter was asked in interviews today if as the lead singer of Nirvana was getting the band back together.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nOMG I looked at this they're coming I can hear the boots."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"OMG I looked at this they're coming I can hear the boots.\", \"questions\": {\"emotion_surprise\": {\"type\": \"noul\", \"instructions\": \"Does the author express surprise?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nYou are the parent your grandson deserves. Your daughter will regret the pain she has caused."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"You are the parent your grandson deserves. Your daughter will regret the pain she has caused.\", \"questions\": {\"emotion_remorse\": {\"type\": \"noul\", \"instructions\": \"Does the author express remorse?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI forgot Plies gone be at the moon tomorrow"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I forgot Plies gone be at the moon tomorrow\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI blocked my card by mistake, how do I unblock it?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I blocked my card by mistake, how do I unblock it?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"compromised_card\": \"compromised card\", \"contactless_not_working\": \"contactless not working\", \"top_up_by_cash_or_cheque\": \"top up by cash or cheque\", \"wrong_amount_of_cash_received\": \"wrong amount of cash received\", \"passcode_forgotten\": \"passcode forgotten\", \"pin_blocked\": \"pin blocked\", \"transaction_charged_twice\": \"transaction charged twice\", \"card_payment_fee_charged\": \"card payment fee charged\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI bought an ecig through the sun paper yet not allowed to use it while watching football at Newcastle Utd"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I bought an ecig through the sun paper yet not allowed to use it while watching football at Newcastle Utd\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nGiving You A Sneak Peak Of Her Hot Little Cunt Hole"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Giving You A Sneak Peak Of Her Hot Little Cunt Hole\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhat was the extra charge for using my card?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"What was the extra charge for using my card?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"card_arrival\": \"card arrival\", \"card_payment_fee_charged\": \"card payment fee charged\", \"pin_blocked\": \"pin blocked\", \"transfer_not_received_by_recipient\": \"transfer not received by recipient\", \"transfer_fee_charged\": \"transfer fee charged\", \"extra_charge_on_statement\": \"extra charge on statement\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHe has never had success"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"He has never had success\", \"questions\": {\"emotion_disappointment\": {\"type\": \"noul\", \"instructions\": \"Does the author express disappointment?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nCan't fall asleep someone come choke me till I pass out you can do whatever you want to my body after I'm passed out #cumslut #slut #horny #wet #pussy #ass #anal #fuck #fuckme #naked #nudes #whore #bitch #naughty #cum #cumtribute #jailbait #blowjob #girls #sex #submissive"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Can't fall asleep someone come choke me till I pass out you can do whatever you want to my body after I'm passed out #cumslut #slut #horny #wet #pussy #ass #anal #fuck #fuckme #naked #nudes #whore #bitch #naughty #cum #cumtribute #jailbait #blowjob #girls #sex #submissive\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThis is as brutal of a loss as it gets. Up 6-0 then up 2 with 2 outs in the 9th only to blow it. It\\u2019s going to be a very long winter in WSH."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"This is as brutal of a loss as it gets. Up 6-0 then up 2 with 2 outs in the 9th only to blow it. It\\\\u2019s going to be a very long winter in WSH.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI know! I would absolutely do this as well. I’d love to to a AMA with him to find out more."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I know! I would absolutely do this as well. I’d love to to a AMA with him to find out more.\", \"questions\": {\"emotion_love\": {\"type\": \"noul\", \"instructions\": \"Does the author express love?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nso you want things to get tied up for months???? just relax... nothing we can do but give it a few days"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"so you want things to get tied up for months???? just relax... nothing we can do but give it a few days\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user Pres. Trump, Hurricane Harvey is Man Made to keep your wall from being built and to open the borders to illegal migration."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user Pres. Trump, Hurricane Harvey is Man Made to keep your wall from being built and to open the borders to illegal migration.\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nATM withdrawal not working."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"ATM withdrawal not working.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"direct_debit_payment_not_recognised\": \"direct debit payment not recognised\", \"declined_cash_withdrawal\": \"declined cash withdrawal\", \"transfer_timing\": \"transfer timing\", \"pin_blocked\": \"pin blocked\", \"disposable_card_limits\": \"disposable card limits\", \"verify_my_identity\": \"verify my identity\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Improving Digital Strategies for Brand Expansion Online\", \"message\": \"Customer support, seeking guidance on digital strategies to enhance brand growth. Would appreciate detailed information on methods that can be used to boost brand awareness and engagement. Looking forward to learning more about the applied approach and how it can benefit my business. Thank you for your time and look forward to hearing back soon.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Improving Digital Strategies for Brand Expansion Online\", \"message\": \"Customer support, seeking guidance on digital strategies to enhance brand growth. Would appreciate detailed information on methods that can be used to boost brand awareness and engagement. Looking forward to learning more about the applied approach and how it can benefit my business. Thank you for your time and look forward to hearing back soon.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Arts & Entertainment/Music\": \"Arts & Entertainment/Music\", \"People & Society/Culture & Society\": \"People & Society/Culture & Society\", \"IT & Technology/Software Development\": \"IT & Technology/Software Development\", \"Customer Service\": \"Customer Service\", \"Pets & Animals/Veterinary Care\": \"Pets & Animals/Veterinary Care\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Change\": \"Change\", \"Problem\": \"Problem\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Reported Data Breach and Wi-Fi Vulnerability\", \"message\": \"I am encountering difficulties with my Wi-Fi network and suspect a data breach. Despite restarting my router, the problem continues. I am worried about my personal information and need your help to address this urgently. Could you please outline the appropriate actions to take to secure my network and data?\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Reported Data Breach and Wi-Fi Vulnerability\", \"message\": \"I am encountering difficulties with my Wi-Fi network and suspect a data breach. Despite restarting my router, the problem continues. I am worried about my personal information and need your help to address this urgently. Could you please outline the appropriate actions to take to secure my network and data?\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Arts & Entertainment/Music\": \"Arts & Entertainment/Music\", \"Technical Support\": \"Technical Support\", \"Jobs & Education/Recruitment\": \"Jobs & Education/Recruitment\", \"Food & Drink/Groceries\": \"Food & Drink/Groceries\", \"Sports\": \"Sports\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Request\": \"Request\", \"Problem\": \"Problem\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHappy 20th Birthday @user God Bless You! Hope God Makes aoo your dreams come true! Love you! Happy Swift Day!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Happy 20th Birthday @user God Bless You! Hope God Makes aoo your dreams come true! Love you! Happy Swift Day!\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Docker Integration Guide\", \"message\": \"I would like to request information on the documentation for integrating Docker with your project management SaaS platform, specifically on an Ubuntu 20.04 system. Could you please provide me with more detailed instructions or direct me to the appropriate resources? Your assistance would be greatly appreciated.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Docker Integration Guide\", \"message\": \"I would like to request information on the documentation for integrating Docker with your project management SaaS platform, specifically on an Ubuntu 20.04 system. Could you please provide me with more detailed instructions or direct me to the appropriate resources? Your assistance would be greatly appreciated.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"IT Support\": \"IT Support\", \"Jobs & Education/Online Courses\": \"Jobs & Education/Online Courses\", \"Customer Service\": \"Customer Service\", \"Hobbies & Leisure/Collectibles\": \"Hobbies & Leisure/Collectibles\", \"Autos & Vehicles/Maintenance\": \"Autos & Vehicles/Maintenance\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nReaders respond: Kudos on immigrant features"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Readers respond: Kudos on immigrant features\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhat is your refund policy for products I have purchased?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"What is your refund policy for products I have purchased?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"virtual_card_not_working\": \"virtual card not working\", \"wrong_amount_of_cash_received\": \"wrong amount of cash received\", \"request_refund\": \"request refund\", \"fiat_currency_support\": \"fiat currency support\", \"compromised_card\": \"compromised card\", \"country_support\": \"country support\", \"lost_or_stolen_card\": \"lost or stolen card\", \"balance_not_updated_after_cheque_or_cash_deposit\": \"balance not updated after cheque or cash deposit\", \"declined_card_payment\": \"declined card payment\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Urgent: Several Hardware and Software Failures\", \"message\": \"Dear Customer Support Team,\\\\n\\\\nI am submitting a report regarding a series of critical hardware and software failures that have significantly disrupted our analytics investment operations. Over the past few days, we've encountered sporadic system outages that seem connected to potential power supply issues or overheating within our computing environment.\\\\n\\\\nInitially, we observed unusual system sluggishness and occasional application crashes during peak usage times. Following these symptoms, we performed initial troubleshooting steps, such as restarting the affected systems.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Urgent: Several Hardware and Software Failures\", \"message\": \"Dear Customer Support Team,\\\\n\\\\nI am submitting a report regarding a series of critical hardware and software failures that have significantly disrupted our analytics investment operations. Over the past few days, we've encountered sporadic system outages that seem connected to potential power supply issues or overheating within our computing environment.\\\\n\\\\nInitially, we observed unusual system sluggishness and occasional application crashes during peak usage times. Following these symptoms, we performed initial troubleshooting steps, such as restarting the affected systems.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Business & Industrial/Manufacturing\": \"Business & Industrial/Manufacturing\", \"IT & Technology/Software Development\": \"IT & Technology/Software Development\", \"Technical Support\": \"Technical Support\", \"Hobbies & Leisure/Collectibles\": \"Hobbies & Leisure/Collectibles\", \"General Inquiry\": \"General Inquiry\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Request\": \"Request\", \"Problem\": \"Problem\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user @user @user @user Nono, you are right: Not all men say not all men...; -P"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user @user @user @user Nono, you are right: Not all men say not all men...; -P\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"QuickBooks Integration Problem\", \"message\": \"Facing issues with integrating QuickBooks Online and Smartsheet. The problem may have arisen due to incompatible software updates. Despite our efforts to sync the applications, which initially seemed to work, the issue still persists. I would greatly appreciate your guidance on how to resolve this. Please let me know if you need any additional information.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"QuickBooks Integration Problem\", \"message\": \"Facing issues with integrating QuickBooks Online and Smartsheet. The problem may have arisen due to incompatible software updates. Despite our efforts to sync the applications, which initially seemed to work, the issue still persists. I would greatly appreciate your guidance on how to resolve this. Please let me know if you need any additional information.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Health/Mental Health\": \"Health/Mental Health\", \"IT Support\": \"IT Support\", \"Beauty & Fitness/Fitness Training\": \"Beauty & Fitness/Fitness Training\", \"Technical Support\": \"Technical Support\", \"Shopping/Retail Stores\": \"Shopping/Retail Stores\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Request\": \"Request\", \"Problem\": \"Problem\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Support for Enhancing Brand Growth Through Digital Strategies\", \"message\": \"Looking for insights on digital strategies to enhance brand growth for clients. Would appreciate detailed information on specific methods and techniques that can be used to drive growth. Please include case studies and testimonials from previous clients to help better understand the services and how they align with our business goals.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Support for Enhancing Brand Growth Through Digital Strategies\", \"message\": \"Looking for insights on digital strategies to enhance brand growth for clients. Would appreciate detailed information on specific methods and techniques that can be used to drive growth. Please include case studies and testimonials from previous clients to help better understand the services and how they align with our business goals.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Customer Service\": \"Customer Service\", \"Food & Drink/Groceries\": \"Food & Drink/Groceries\", \"Law & Government/Government Services\": \"Law & Government/Government Services\", \"Product Support\": \"Product Support\", \"Law & Government/Legal Advice\": \"Law & Government/Legal Advice\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user U suspend acct of rape victim, but not some one who promises 'Hell&Fury the planet has never seen' = #NuclearWar. You are a #HYPOCRITE"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user U suspend acct of rape victim, but not some one who promises 'Hell&Fury the planet has never seen' = #NuclearWar. You are a #HYPOCRITE\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI want to reactivate my card, I thought I had lost it but found it again in my jacket."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I want to reactivate my card, I thought I had lost it but found it again in my jacket.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"country_support\": \"country support\", \"passcode_forgotten\": \"passcode forgotten\", \"card_payment_fee_charged\": \"card payment fee charged\", \"card_linking\": \"card linking\", \"supported_cards_and_currencies\": \"supported cards and currencies\", \"cancel_transfer\": \"cancel transfer\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user @user AND @user all on The Jonathan Ross show tonight = Terrif sat night in! Well done @user #Toofunny"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user @user AND @user all on The Jonathan Ross show tonight = Terrif sat night in! Well done @user #Toofunny\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThe no rape was my favorite part, so glad people agree"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"The no rape was my favorite part, so glad people agree\", \"questions\": {\"emotion_joy\": {\"type\": \"noul\", \"instructions\": \"Does the author express joy?\"}, \"emotion_love\": {\"type\": \"noul\", \"instructions\": \"Does the author express love?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n#jihad #ISIS #terrorism #hate #ISIL #shariaWhen President Obama says we are going to thoroughly vet refugees..."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"#jihad #ISIS #terrorism #hate #ISIL #shariaWhen President Obama says we are going to thoroughly vet refugees...\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThis is the most accurate representation of BO4 I have seen yet👍👍👍"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"This is the most accurate representation of BO4 I have seen yet👍👍👍\", \"questions\": {\"emotion_admiration\": {\"type\": \"noul\", \"instructions\": \"Does the author express admiration?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nFebruary 13th\\u002c the day before and best buddy to valentines day\\u002c Desperation Day"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"February 13th\\\\u002c the day before and best buddy to valentines day\\\\u002c Desperation Day\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nSure Sadiq Kunt, whatever you say. 🖕😒"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Sure Sadiq Kunt, whatever you say. 🖕😒\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nTomorrow is #worldrefugeeday"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Tomorrow is #worldrefugeeday\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user happy Saturday, anyonr buy the new adidas jersey, what is the sizing like to nike? Nike always fit me small"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user happy Saturday, anyonr buy the new adidas jersey, what is the sizing like to nike? Nike always fit me small\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nOrganizing events? Hiring employees? Need a community of diverse immigrant women leaders? Email us for a partnership request: contact@femigrants.org #partnership #collaboration #diverseimmigrantwomen #diversewomenleaders #empowerfemaleimmigrants #womenleaders #femaleimmigrants"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Organizing events? Hiring employees? Need a community of diverse immigrant women leaders? Email us for a partnership request: contact@femigrants.org #partnership #collaboration #diverseimmigrantwomen #diversewomenleaders #empowerfemaleimmigrants #womenleaders #femaleimmigrants\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIllegal Alien kills 2 Search Dogs during pursuit #BuildTheWall #EndSanctuaryCities #EndChainMigration #AmericansFirst @user @user @user @user"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Illegal Alien kills 2 Search Dogs during pursuit #BuildTheWall #EndSanctuaryCities #EndChainMigration #AmericansFirst @user @user @user @user\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n‘˜Explosive situation’: Pro-migrant group forced to end its crucial work in Paris due to violence"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"‘˜Explosive situation’: Pro-migrant group forced to end its crucial work in Paris due to violence\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThe state should kill them yes, you seem to have a very low IQ and can’t grasp [RELIGION]. No wonder you took up Reddit Atheism"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"The state should kill them yes, you seem to have a very low IQ and can’t grasp [RELIGION]. No wonder you took up Reddit Atheism\", \"questions\": {\"emotion_anger\": {\"type\": \"noul\", \"instructions\": \"Does the author express anger?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"An issue with data encryption was encountered today\", \"message\": \"Data encryption has failed, leading to the exposure of patient data due to inadequate firewall updates.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"An issue with data encryption was encountered today\", \"message\": \"Data encryption has failed, leading to the exposure of patient data due to inadequate firewall updates.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Technical Support\": \"Technical Support\", \"Games\": \"Games\", \"People & Society/Culture & Society\": \"People & Society/Culture & Society\", \"Law & Government/Government Services\": \"Law & Government/Government Services\", \"Food & Drink/Restaurants\": \"Food & Drink/Restaurants\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Incident\": \"Incident\", \"Change\": \"Change\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nLooks like a bread line in Africa.. Can #Spain afford all this? Hundreds of migrants break through border at Spanish exclave Ceuta via @user"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Looks like a bread line in Africa.. Can #Spain afford all this? Hundreds of migrants break through border at Spanish exclave Ceuta via @user\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n\"Curtis - Inde, KY would go after Davis despite the 1st Amendment (And 2 Muslim truck drivers get $240 000. See )\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"\\\"Curtis - Inde, KY would go after Davis despite the 1st Amendment (And 2 Muslim truck drivers get $240 000. See )\\\"\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user @user it's crazy what happing in Middle East, 1st it was only just Israel nowit's spread&amp;ppl killingtheonetheylived4ages\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user @user it's crazy what happing in Middle East, 1st it was only just Israel nowit's spread&amp;ppl killingtheonetheylived4ages\\\"\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWould be nice 2 c Pirlo score from one of those amazing FK and Balotelli get on the field 2 day for Milan that would be a good Saturday"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Would be nice 2 c Pirlo score from one of those amazing FK and Balotelli get on the field 2 day for Milan that would be a good Saturday\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nA girl known as daddy... well that’s a new twist on an old trick"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"A girl known as daddy... well that’s a new twist on an old trick\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Struggles with Digital Growth Strategies\", \"message\": \"Attempted to enhance ad content and increase the budget, but digital strategies failed to boost brand growth as expected due to ineffective targeting and poor campaign execution.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Struggles with Digital Growth Strategies\", \"message\": \"Attempted to enhance ad content and increase the budget, but digital strategies failed to boost brand growth as expected due to ineffective targeting and poor campaign execution.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Arts & Entertainment/Music\": \"Arts & Entertainment/Music\", \"Health/Medical Services\": \"Health/Medical Services\", \"Product Support\": \"Product Support\", \"Science/Environmental Science\": \"Science/Environmental Science\", \"Law & Government/Government Services\": \"Law & Government/Government Services\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHave you ever noticed that the Gay Activists never go after Muslims...?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Have you ever noticed that the Gay Activists never go after Muslims...?\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n#Refugees not welcome!!!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"#Refugees not welcome!!!\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhat are the policies of disposable cards?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"What are the policies of disposable cards?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"exchange_charge\": \"exchange charge\", \"disposable_card_limits\": \"disposable card limits\", \"transfer_into_account\": \"transfer into account\", \"virtual_card_not_working\": \"virtual card not working\", \"why_verify_identity\": \"why verify identity\", \"balance_not_updated_after_bank_transfer\": \"balance not updated after bank transfer\", \"wrong_exchange_rate_for_cash_withdrawal\": \"wrong exchange rate for cash withdrawal\", \"pending_transfer\": \"pending transfer\", \"edit_personal_details\": \"edit personal details\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nYoga can help your child concentrate at school. Research shows that the deep breathing exercises may improve children's focus."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Yoga can help your child concentrate at school. Research shows that the deep breathing exercises may improve children's focus.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n\"Ok, I remember where I was when Josh Hamilton had the massive 1st round, etc in '08---and hats off to @user bcuz that was Epic also!!\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"\\\"Ok, I remember where I was when Josh Hamilton had the massive 1st round, etc in '08---and hats off to @user bcuz that was Epic also!!\\\"\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user @user Give it a rest, you fugly skank."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user @user Give it a rest, you fugly skank.\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Enhance Digital Strategy Growth with Microsoft Azure and Zapier\", \"message\": \"I am seeking assistance in optimizing our digital strategies using Microsoft Azure and Zapier. Our goal is to enhance brand growth and market reach. We need help in integrating tools to streamline operations and improve efficiency. Could you please provide guidance on how to get started and recommend best practices for implementation? I look forward to hearing from you soon.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Enhance Digital Strategy Growth with Microsoft Azure and Zapier\", \"message\": \"I am seeking assistance in optimizing our digital strategies using Microsoft Azure and Zapier. Our goal is to enhance brand growth and market reach. We need help in integrating tools to streamline operations and improve efficiency. Could you please provide guidance on how to get started and recommend best practices for implementation? I look forward to hearing from you soon.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Technical Support\": \"Technical Support\", \"Law & Government/Legal Advice\": \"Law & Government/Legal Advice\", \"IT & Technology/Security Operations\": \"IT & Technology/Security Operations\", \"Shopping/E-commerce\": \"Shopping/E-commerce\", \"Food & Drink/Groceries\": \"Food & Drink/Groceries\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Alert for Data Security Breach\", \"message\": \"A healthcare provider faced a data security breach due to insufficient encryption and access control measures.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Alert for Data Security Breach\", \"message\": \"A healthcare provider faced a data security breach due to insufficient encryption and access control measures.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Product Support\": \"Product Support\", \"Law & Government/Legal Advice\": \"Law & Government/Legal Advice\", \"Billing and Payments\": \"Billing and Payments\", \"IT & Technology/Network Infrastructure\": \"IT & Technology/Network Infrastructure\", \"Law & Government/Government Services\": \"Law & Government/Government Services\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Problem\": \"Problem\", \"Request\": \"Request\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Details on Scalability Features Project\", \"message\": \"Request for information\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Details on Scalability Features Project\", \"message\": \"Request for information\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Travel & Transportation/Air Travel\": \"Travel & Transportation/Air Travel\", \"Finance/Personal Finance\": \"Finance/Personal Finance\", \"Beauty & Fitness/Fitness Training\": \"Beauty & Fitness/Fitness Training\", \"Product Support\": \"Product Support\", \"IT Support\": \"IT Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Request\": \"Request\", \"Change\": \"Change\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Technical Difficulties Encountered\", \"message\": \"A financial organization has encountered simultaneous outages across various services, potentially due to security breaches in the network. The initial steps to diagnose the problem involved rebooting systems and conducting security inspections, but the issues remain unresolved. The team is actively working to fix the problem and will provide updates shortly. In the meantime, please feel free to share any questions or concerns. We appreciate your patience and understanding as we work to restore full service.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Technical Difficulties Encountered\", \"message\": \"A financial organization has encountered simultaneous outages across various services, potentially due to security breaches in the network. The initial steps to diagnose the problem involved rebooting systems and conducting security inspections, but the issues remain unresolved. The team is actively working to fix the problem and will provide updates shortly. In the meantime, please feel free to share any questions or concerns. We appreciate your patience and understanding as we work to restore full service.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Pets & Animals/Pet Services\": \"Pets & Animals/Pet Services\", \"People & Society/Culture & Society\": \"People & Society/Culture & Society\", \"Billing and Payments\": \"Billing and Payments\", \"Service Outages and Maintenance\": \"Service Outages and Maintenance\", \"News\": \"News\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Request\": \"Request\", \"Change\": \"Change\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThat’s a bold move, [NAME]. Let’s see if it pays off."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"That’s a bold move, [NAME]. Let’s see if it pays off.\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIm not sure if a seller is giving me my money back! I asked them to refund my money for the item but It's been a week now and still nothing has been given back to me. Can you resolve this for me?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Im not sure if a seller is giving me my money back! I asked them to refund my money for the item but It's been a week now and still nothing has been given back to me. Can you resolve this for me?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"change_pin\": \"change pin\", \"passcode_forgotten\": \"passcode forgotten\", \"automatic_top_up\": \"automatic top up\", \"card_linking\": \"card linking\", \"Refund_not_showing_up\": \"Refund not showing up\", \"edit_personal_details\": \"edit personal details\", \"getting_spare_card\": \"getting spare card\", \"atm_support\": \"atm support\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"\", \"message\": \"Dear Customer Support, I am encountering an issue with the data analytics tools that we use for investment optimization. These tools have been giving us inaccurate insights, which may be due to recent software updates that are conflicting with our existing integrations. I have already tried restarting the systems, checking for updates, and reviewing the configuration settings, but the issue still persists. I would greatly appreciate it if you could look into this matter and provide a solution as soon as possible. The inaccurate insights are causing difficulties with our investment decisions, and we need this issue resolved promptly.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"\", \"message\": \"Dear Customer Support, I am encountering an issue with the data analytics tools that we use for investment optimization. These tools have been giving us inaccurate insights, which may be due to recent software updates that are conflicting with our existing integrations. I have already tried restarting the systems, checking for updates, and reviewing the configuration settings, but the issue still persists. I would greatly appreciate it if you could look into this matter and provide a solution as soon as possible. The inaccurate insights are causing difficulties with our investment decisions, and we need this issue resolved promptly.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Product Support\": \"Product Support\", \"Sales and Pre-Sales\": \"Sales and Pre-Sales\", \"Autos & Vehicles/Maintenance\": \"Autos & Vehicles/Maintenance\", \"IT & Technology/Hardware Support\": \"IT & Technology/Hardware Support\", \"Jobs & Education/Recruitment\": \"Jobs & Education/Recruitment\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIt's October 17th and I just saw a Christmas commercial..."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"It's October 17th and I just saw a Christmas commercial...\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n\"This Sat., I'll be at the Whole Earth Festival in Davis, CA. Chatting with Kim Stanley Robinson about SF and the environment, from 1-3pm.\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"\\\"This Sat., I'll be at the Whole Earth Festival in Davis, CA. Chatting with Kim Stanley Robinson about SF and the environment, from 1-3pm.\\\"\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nMany of them do love LEGAL ones."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Many of them do love LEGAL ones.\", \"questions\": {\"emotion_love\": {\"type\": \"noul\", \"instructions\": \"Does the author express love?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI'm not attacking you. Which part of our conversation gives the impression that I'm attacking you? Honest question."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I'm not attacking you. Which part of our conversation gives the impression that I'm attacking you? Honest question.\", \"questions\": {\"emotion_curiosity\": {\"type\": \"noul\", \"instructions\": \"Does the author express curiosity?\"}, \"emotion_disapproval\": {\"type\": \"noul\", \"instructions\": \"Does the author express disapproval?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user @user Shut the fuck up bitch how do you know what people have and don't have?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user @user Shut the fuck up bitch how do you know what people have and don't have?\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nnot surprised Frank Ocean dropped out of FYF. my inner-curiousty had me hoping he wouls drop that jawn before or during Saturday but nawwww."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"not surprised Frank Ocean dropped out of FYF. my inner-curiousty had me hoping he wouls drop that jawn before or during Saturday but nawwww.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAre there any restrictions for the disposable cards that I need to be aware of?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Are there any restrictions for the disposable cards that I need to be aware of?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"declined_cash_withdrawal\": \"declined cash withdrawal\", \"cash_withdrawal_not_recognised\": \"cash withdrawal not recognised\", \"card_payment_not_recognised\": \"card payment not recognised\", \"transaction_charged_twice\": \"transaction charged twice\", \"atm_support\": \"atm support\", \"cancel_transfer\": \"cancel transfer\", \"disposable_card_limits\": \"disposable card limits\", \"apple_pay_or_google_pay\": \"apple pay or google pay\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWe really need to go on some sort of winning streak"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"We really need to go on some sort of winning streak\", \"questions\": {\"emotion_desire\": {\"type\": \"noul\", \"instructions\": \"Does the author express desire?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Assistance with Financial Products\", \"message\": \"Is it possible to get detailed information on financial products by utilizing data analytics to optimize investments? Your assistance is important to me, and I appreciate your help. I look forward to hearing back from you soon.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Assistance with Financial Products\", \"message\": \"Is it possible to get detailed information on financial products by utilizing data analytics to optimize investments? Your assistance is important to me, and I appreciate your help. I look forward to hearing back from you soon.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Autos & Vehicles/Maintenance\": \"Autos & Vehicles/Maintenance\", \"Product Support\": \"Product Support\", \"Autos & Vehicles/Sales\": \"Autos & Vehicles/Sales\", \"News\": \"News\", \"Service Outages and Maintenance\": \"Service Outages and Maintenance\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nEven when it's bad, it's still pretty good"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Even when it's bad, it's still pretty good\", \"questions\": {\"emotion_admiration\": {\"type\": \"noul\", \"instructions\": \"Does the author express admiration?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI put on my Fleetwood Mac playlist in the shower and somehow drag me down got on there and I heard Harry and just sat there in the shower"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I put on my Fleetwood Mac playlist in the shower and somehow drag me down got on there and I heard Harry and just sat there in the shower\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nMy password doesn't work."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"My password doesn't work.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"age_limit\": \"age limit\", \"top_up_by_bank_transfer_charge\": \"top up by bank transfer charge\", \"topping_up_by_card\": \"topping up by card\", \"verify_top_up\": \"verify top up\", \"transfer_fee_charged\": \"transfer fee charged\", \"card_payment_wrong_exchange_rate\": \"card payment wrong exchange rate\", \"passcode_forgotten\": \"passcode forgotten\", \"pending_card_payment\": \"pending card payment\", \"automatic_top_up\": \"automatic top up\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"\", \"message\": \"Dear Customer Support,\\\\n\\\\nI am reporting an issue with data access delays at a financial firm that we are currently managing. These delays are impacting our analytics functionalities and may be linked to a RAID-Controller NAS configuration problem.\\\\n\\\\nWe have already taken initial troubleshooting measures, such as rebooting the devices and verifying the RAID setup's integrity. Unfortunately, the issue continues despite these efforts.\\\\n\\\\nThe problem manifests as slow data retrieval times, which impede our ability to generate reports.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"\", \"message\": \"Dear Customer Support,\\\\n\\\\nI am reporting an issue with data access delays at a financial firm that we are currently managing. These delays are impacting our analytics functionalities and may be linked to a RAID-Controller NAS configuration problem.\\\\n\\\\nWe have already taken initial troubleshooting measures, such as rebooting the devices and verifying the RAID setup's integrity. Unfortunately, the issue continues despite these efforts.\\\\n\\\\nThe problem manifests as slow data retrieval times, which impede our ability to generate reports.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Beauty & Fitness/Cosmetics\": \"Beauty & Fitness/Cosmetics\", \"Product Support\": \"Product Support\", \"Online Communities/Social Networks\": \"Online Communities/Social Networks\", \"Shopping/Retail Stores\": \"Shopping/Retail Stores\", \"Sales and Pre-Sales\": \"Sales and Pre-Sales\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Problem\": \"Problem\", \"Change\": \"Change\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIn our military at the time you could pay out at any point. They really only wanted people who wanted to be there."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"In our military at the time you could pay out at any point. They really only wanted people who wanted to be there.\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI dont think she knows that once your body runs out of fat it starts eating your muscles and organs."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I dont think she knows that once your body runs out of fat it starts eating your muscles and organs.\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user Cameron is a coward he attacks the poor yet again because of immigration the numbers don't add up we need to get out now"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user Cameron is a coward he attacks the poor yet again because of immigration the numbers don't add up we need to get out now\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nDepression is swiping on gay tinder in front of your immigrant parents bc you don't fear death."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Depression is swiping on gay tinder in front of your immigrant parents bc you don't fear death.\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Strengthen Medical Data Security Procedures\", \"message\": \"Customer Support team, I am submitting a request to improve the security measures for medical data within hospital systems across designated healthcare devices and software. While the existing security protocols sufficiently safeguard sensitive patient information, I believe it is crucial to establish more robust security practices to prevent potential data breaches and cyber threats. I would like to request a review and update of the current security protocols to ensure the confidentiality, integrity, and availability of medical data.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Strengthen Medical Data Security Procedures\", \"message\": \"Customer Support team, I am submitting a request to improve the security measures for medical data within hospital systems across designated healthcare devices and software. While the existing security protocols sufficiently safeguard sensitive patient information, I believe it is crucial to establish more robust security practices to prevent potential data breaches and cyber threats. I would like to request a review and update of the current security protocols to ensure the confidentiality, integrity, and availability of medical data.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Law & Government/Government Services\": \"Law & Government/Government Services\", \"Finance/Personal Finance\": \"Finance/Personal Finance\", \"Travel & Transportation/Air Travel\": \"Travel & Transportation/Air Travel\", \"Product Support\": \"Product Support\", \"Beauty & Fitness/Fitness Training\": \"Beauty & Fitness/Fitness Training\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Change\": \"Change\", \"Incident\": \"Incident\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIf your man has an android, you're allowed to cheat on him"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"If your man has an android, you're allowed to cheat on him\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nRon Paul Snr Advisor Doug Wead Interview with Frost &#8211; Mar 31 2012"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Ron Paul Snr Advisor Doug Wead Interview with Frost &#8211; Mar 31 2012\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"\", \"message\": \"Customer Support, seeking assistance with configuration adjustments across various products to enhance digital marketing efforts and foster brand growth. The current setup is producing satisfactory results, but we believe that proper adjustments could further improve our online visibility and expand our audience reach. We would like to request updates to our social media profiles, email marketing campaigns, and search engine optimization (SEO) configurations. Specifically, we are interested in updating our Facebook account.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"\", \"message\": \"Customer Support, seeking assistance with configuration adjustments across various products to enhance digital marketing efforts and foster brand growth. The current setup is producing satisfactory results, but we believe that proper adjustments could further improve our online visibility and expand our audience reach. We would like to request updates to our social media profiles, email marketing campaigns, and search engine optimization (SEO) configurations. Specifically, we are interested in updating our Facebook account.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Pets & Animals/Veterinary Care\": \"Pets & Animals/Veterinary Care\", \"News\": \"News\", \"Home & Garden/Home Improvement\": \"Home & Garden/Home Improvement\", \"Games\": \"Games\", \"Technical Support\": \"Technical Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Request\": \"Request\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAgree. Keep trying, then if your rejected every time. I'm sorry your done."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Agree. Keep trying, then if your rejected every time. I'm sorry your done.\", \"questions\": {\"emotion_disappointment\": {\"type\": \"noul\", \"instructions\": \"Does the author express disappointment?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nRed Sox have runners at 1st and second in the bottom of the 3rd."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Red Sox have runners at 1st and second in the bottom of the 3rd.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhich part do you not understand?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Which part do you not understand?\", \"questions\": {\"emotion_confusion\": {\"type\": \"noul\", \"instructions\": \"Does the author express confusion?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nOh, you're completely right. I'm an idiot. I'll be here eating my words."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Oh, you're completely right. I'm an idiot. I'll be here eating my words.\", \"questions\": {\"emotion_annoyance\": {\"type\": \"noul\", \"instructions\": \"Does the author express annoyance?\"}, \"emotion_approval\": {\"type\": \"noul\", \"instructions\": \"Does the author express approval?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nPolish MP: Our country is safe because we didn’t accept illegal Muslim migrants"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Polish MP: Our country is safe because we didn’t accept illegal Muslim migrants\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nDo you know why my card payment has been declined?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Do you know why my card payment has been declined?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"direct_debit_payment_not_recognised\": \"direct debit payment not recognised\", \"declined_card_payment\": \"declined card payment\", \"visa_or_mastercard\": \"visa or mastercard\", \"transfer_not_received_by_recipient\": \"transfer not received by recipient\", \"exchange_via_app\": \"exchange via app\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nReverse racism is just racism"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Reverse racism is just racism\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user @user Well duh I made the gc you a hole"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user @user Well duh I made the gc you a hole\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nJust hit Leichhardt! Well, the wind has, waiting for the cool now though!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Just hit Leichhardt! Well, the wind has, waiting for the cool now though!\", \"questions\": {\"emotion_optimism\": {\"type\": \"noul\", \"instructions\": \"Does the author express optimism?\"}, \"emotion_realization\": {\"type\": \"noul\", \"instructions\": \"Does the author express realization?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHow can I tell if my top-up has been cancelled?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"How can I tell if my top-up has been cancelled?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"exchange_charge\": \"exchange charge\", \"pending_card_payment\": \"pending card payment\", \"verify_my_identity\": \"verify my identity\", \"edit_personal_details\": \"edit personal details\", \"cash_withdrawal_not_recognised\": \"cash withdrawal not recognised\", \"declined_card_payment\": \"declined card payment\", \"top_up_by_cash_or_cheque\": \"top up by cash or cheque\", \"top_up_reverted\": \"top up reverted\", \"transfer_fee_charged\": \"transfer fee charged\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHysterical pant -wetting liberals in full flow Grow up the lot of you #FoxAche"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Hysterical pant -wetting liberals in full flow Grow up the lot of you #FoxAche\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Support Request for Addressing Data Leak\", \"message\": \"Confidential medical data was exposed through an insecure USB-Stick\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Support Request for Addressing Data Leak\", \"message\": \"Confidential medical data was exposed through an insecure USB-Stick\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Hobbies & Leisure/Collectibles\": \"Hobbies & Leisure/Collectibles\", \"Returns and Exchanges\": \"Returns and Exchanges\", \"Arts & Entertainment/Movies\": \"Arts & Entertainment/Movies\", \"Technical Support\": \"Technical Support\", \"Online Communities/Forums\": \"Online Communities/Forums\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Problem with Today's Software Update\", \"message\": \"A significant software malfunction has impacted the digital marketing campaign. The issue may be due to conflicts with recent updates affecting applications.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Problem with Today's Software Update\", \"message\": \"A significant software malfunction has impacted the digital marketing campaign. The issue may be due to conflicts with recent updates affecting applications.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Sales and Pre-Sales\": \"Sales and Pre-Sales\", \"Shopping/Retail Stores\": \"Shopping/Retail Stores\", \"Science/Research\": \"Science/Research\", \"Technical Support\": \"Technical Support\", \"Sports\": \"Sports\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Request\": \"Request\", \"Change\": \"Change\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Support Required for Data Encryption Malfunction\", \"message\": \"There has been an encryption failure due to software problems\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Support Required for Data Encryption Malfunction\", \"message\": \"There has been an encryption failure due to software problems\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Books & Literature/Fiction\": \"Books & Literature/Fiction\", \"Online Communities/Social Networks\": \"Online Communities/Social Networks\", \"Pets & Animals/Veterinary Care\": \"Pets & Animals/Veterinary Care\", \"Arts & Entertainment/Music\": \"Arts & Entertainment/Music\", \"Technical Support\": \"Technical Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Request\": \"Request\", \"Incident\": \"Incident\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n#migrants perhaps all countries that gave massive food aid to countries that could not feed their own people due to too many people as is (eg: the country is a desert) years ago should readily welcome the aftermath in the form of endless migration of refugees to their homes?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"#migrants perhaps all countries that gave massive food aid to countries that could not feed their own people due to too many people as is (eg: the country is a desert) years ago should readily welcome the aftermath in the form of endless migration of refugees to their homes?\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHow do I add a card on to the app?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"How do I add a card on to the app?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"activate_my_card\": \"activate my card\", \"card_linking\": \"card linking\", \"age_limit\": \"age limit\", \"automatic_top_up\": \"automatic top up\", \"extra_charge_on_statement\": \"extra charge on statement\", \"fiat_currency_support\": \"fiat currency support\", \"getting_spare_card\": \"getting spare card\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Seeking Guidance\", \"message\": \"In need of customer support to seek guidance on integrating TensorFlow with SAP ERP for investment optimization. Would appreciate resources and documentation to help get started on the project. Looking to leverage machine learning for investment optimization using TensorFlow and SAP ERP. Specifically, interested in knowing the existing APIs and tools that can facilitate this integration. Thank you for your time and assistance. Looking forward to hearing back soon. Regards, [Your\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Seeking Guidance\", \"message\": \"In need of customer support to seek guidance on integrating TensorFlow with SAP ERP for investment optimization. Would appreciate resources and documentation to help get started on the project. Looking to leverage machine learning for investment optimization using TensorFlow and SAP ERP. Specifically, interested in knowing the existing APIs and tools that can facilitate this integration. Thank you for your time and assistance. Looking forward to hearing back soon. Regards, [Your\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Shopping/E-commerce\": \"Shopping/E-commerce\", \"Home & Garden/Home Improvement\": \"Home & Garden/Home Improvement\", \"Sales and Pre-Sales\": \"Sales and Pre-Sales\", \"Customer Service\": \"Customer Service\", \"Law & Government/Government Services\": \"Law & Government/Government Services\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI wish the best of luck to the boro football players and hope they kick butt Saturday for shaun #WinItForShaun"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I wish the best of luck to the boro football players and hope they kick butt Saturday for shaun #WinItForShaun\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nLol I have my moments!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Lol I have my moments!\", \"questions\": {\"emotion_amusement\": {\"type\": \"noul\", \"instructions\": \"Does the author express amusement?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI used my card and there was a fee"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I used my card and there was a fee\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"pin_blocked\": \"pin blocked\", \"card_payment_fee_charged\": \"card payment fee charged\", \"passcode_forgotten\": \"passcode forgotten\", \"unable_to_verify_identity\": \"unable to verify identity\", \"compromised_card\": \"compromised card\", \"card_swallowed\": \"card swallowed\", \"top_up_limits\": \"top up limits\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Problem with the Investment Optimization Platform\", \"message\": \"The investment optimization platform is currently down and may be due to a server malfunction or software incompatibility. We have tried restarting the services but to no avail. We are seeking assistance.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Problem with the Investment Optimization Platform\", \"message\": \"The investment optimization platform is currently down and may be due to a server malfunction or software incompatibility. We have tried restarting the services but to no avail. We are seeking assistance.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Autos & Vehicles/Sales\": \"Autos & Vehicles/Sales\", \"Hobbies & Leisure/Crafts\": \"Hobbies & Leisure/Crafts\", \"Service Outages and Maintenance\": \"Service Outages and Maintenance\", \"Sports\": \"Sports\", \"General Inquiry\": \"General Inquiry\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Systems for Security Breach\", \"message\": \"A security breach has been detected in the medical data systems, which may have exposed patient information. The updated antivirus software was reviewed, and the access logs indicate that the risk still persists.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Systems for Security Breach\", \"message\": \"A security breach has been detected in the medical data systems, which may have exposed patient information. The updated antivirus software was reviewed, and the access logs indicate that the risk still persists.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Product Support\": \"Product Support\", \"Customer Service\": \"Customer Service\", \"Travel & Transportation/Land Travel\": \"Travel & Transportation/Land Travel\", \"IT & Technology/Security Operations\": \"IT & Technology/Security Operations\", \"Games\": \"Games\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Improvement in Digital Marketing Tools Integration\", \"message\": \"This is a request for the enhancement of digital marketing tools integration. It would significantly boost campaign efficiency and brand visibility. I believe such an update would be beneficial for all users.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Improvement in Digital Marketing Tools Integration\", \"message\": \"This is a request for the enhancement of digital marketing tools integration. It would significantly boost campaign efficiency and brand visibility. I believe such an update would be beneficial for all users.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Travel & Transportation/Air Travel\": \"Travel & Transportation/Air Travel\", \"Technical Support\": \"Technical Support\", \"Sports\": \"Sports\", \"Business & Industrial/Manufacturing\": \"Business & Industrial/Manufacturing\", \"Customer Service\": \"Customer Service\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n\"https://t.co/ozTi5f5eFe Top Gear to return in the spring, says Chris Evans #ChrisEvans\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"\\\"https://t.co/ozTi5f5eFe Top Gear to return in the spring, says Chris Evans #ChrisEvans\\\"\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nDACA - Wednesday afternoon, Judge Hanen will hear arguments over whether to issue nationwide injunction of his own, which would stop Trump admin from continuing implement DACA in any way. Immigration experts expect him to issue that injunction. It nullifies decision of DC judge"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"DACA - Wednesday afternoon, Judge Hanen will hear arguments over whether to issue nationwide injunction of his own, which would stop Trump admin from continuing implement DACA in any way. Immigration experts expect him to issue that injunction. It nullifies decision of DC judge\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI don't know what's wrong with my card, but it has not worked at multiple places today."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I don't know what's wrong with my card, but it has not worked at multiple places today.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"balance_not_updated_after_bank_transfer\": \"balance not updated after bank transfer\", \"card_not_working\": \"card not working\", \"pending_transfer\": \"pending transfer\", \"transfer_into_account\": \"transfer into account\", \"atm_support\": \"atm support\", \"getting_virtual_card\": \"getting virtual card\", \"transfer_not_received_by_recipient\": \"transfer not received by recipient\", \"terminate_account\": \"terminate account\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAre you going to Shawn's show on September 8th? Me: (Vine by literally shawn;)"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Are you going to Shawn's show on September 8th? Me: (Vine by literally shawn;)\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nOo! OO!! I know this one... ME!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Oo! OO!! I know this one... ME!\", \"questions\": {\"emotion_excitement\": {\"type\": \"noul\", \"instructions\": \"Does the author express excitement?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Problem with Subscription Fees\", \"message\": \"I am facing unanticipated charges related to my monthly subscription. I have reviewed my account configurations and the invoices, yet the issue remains unresolved. I believe there might be a discrepancy in the billing system. Could you assist me in addressing this situation?\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Problem with Subscription Fees\", \"message\": \"I am facing unanticipated charges related to my monthly subscription. I have reviewed my account configurations and the invoices, yet the issue remains unresolved. I believe there might be a discrepancy in the billing system. Could you assist me in addressing this situation?\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Business & Industrial/Manufacturing\": \"Business & Industrial/Manufacturing\", \"Food & Drink/Restaurants\": \"Food & Drink/Restaurants\", \"Billing and Payments\": \"Billing and Payments\", \"Law & Government/Government Services\": \"Law & Government/Government Services\", \"Finance/Investments\": \"Finance/Investments\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Change\": \"Change\", \"Problem\": \"Problem\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWho is going to the #UnitedSoundsOfAfrica concert tomorrow at Irving Plaza NYC? cc @user"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Who is going to the #UnitedSoundsOfAfrica concert tomorrow at Irving Plaza NYC? cc @user\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Cybersecurity Strategies\", \"message\": \"Hello, inquiring about digital strategies to promote cybersecurity products such as Bitdefender. Could you provide details on methods used to raise awareness and drive sales? I am particularly interested in learning about different marketing approaches and types of products. Specifically, I am keen to know about social media campaigns, email marketing efforts, and tactics employed to reach the target audience. Looking forward to your response with digital strategies.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Cybersecurity Strategies\", \"message\": \"Hello, inquiring about digital strategies to promote cybersecurity products such as Bitdefender. Could you provide details on methods used to raise awareness and drive sales? I am particularly interested in learning about different marketing approaches and types of products. Specifically, I am keen to know about social media campaigns, email marketing efforts, and tactics employed to reach the target audience. Looking forward to your response with digital strategies.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Hobbies & Leisure/Crafts\": \"Hobbies & Leisure/Crafts\", \"Returns and Exchanges\": \"Returns and Exchanges\", \"Games\": \"Games\", \"Autos & Vehicles/Maintenance\": \"Autos & Vehicles/Maintenance\", \"Finance/Investments\": \"Finance/Investments\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Incident\": \"Incident\", \"Request\": \"Request\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nSam and sat in a Dunkin donuts for an hour bc I thought I locked my keys in the car but they were actually in my pocket the whole time :///"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Sam and sat in a Dunkin donuts for an hour bc I thought I locked my keys in the car but they were actually in my pocket the whole time :///\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nGreat article on the #refugee crisis in Europe. \"The driving question for Europeans is: “How do we make it stop?” But it refers not to the humanitarian drama of the forcibly displaced, just to their ability to reach Europe.\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Great article on the #refugee crisis in Europe. \\\"The driving question for Europeans is: “How do we make it stop?” But it refers not to the humanitarian drama of the forcibly displaced, just to their ability to reach Europe.\\\"\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI need my card right away."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I need my card right away.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"lost_or_stolen_phone\": \"lost or stolen phone\", \"reverted_card_payment?\": \"reverted card payment?\", \"activate_my_card\": \"activate my card\", \"card_about_to_expire\": \"card about to expire\", \"verify_my_identity\": \"verify my identity\", \"terminate_account\": \"terminate account\", \"top_up_by_cash_or_cheque\": \"top up by cash or cheque\", \"card_delivery_estimate\": \"card delivery estimate\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhy was my transfer stopped ?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Why was my transfer stopped ?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"verify_top_up\": \"verify top up\", \"beneficiary_not_allowed\": \"beneficiary not allowed\", \"virtual_card_not_working\": \"virtual card not working\", \"lost_or_stolen_card\": \"lost or stolen card\", \"wrong_amount_of_cash_received\": \"wrong amount of cash received\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhere do I go to terminate my account?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Where do I go to terminate my account?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"change_pin\": \"change pin\", \"age_limit\": \"age limit\", \"atm_support\": \"atm support\", \"lost_or_stolen_card\": \"lost or stolen card\", \"disposable_card_limits\": \"disposable card limits\", \"pending_transfer\": \"pending transfer\", \"pending_cash_withdrawal\": \"pending cash withdrawal\", \"terminate_account\": \"terminate account\", \"apple_pay_or_google_pay\": \"apple pay or google pay\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nSomeone besides me might be using my card."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Someone besides me might be using my card.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"pending_card_payment\": \"pending card payment\", \"compromised_card\": \"compromised card\", \"getting_spare_card\": \"getting spare card\", \"terminate_account\": \"terminate account\", \"card_payment_fee_charged\": \"card payment fee charged\", \"disposable_card_limits\": \"disposable card limits\", \"visa_or_mastercard\": \"visa or mastercard\", \"transfer_fee_charged\": \"transfer fee charged\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHe'll be the difference maker in the playoffs. He can get any shot he wants because of his all time great ball handling"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"He'll be the difference maker in the playoffs. He can get any shot he wants because of his all time great ball handling\", \"questions\": {\"emotion_admiration\": {\"type\": \"noul\", \"instructions\": \"Does the author express admiration?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nleft my phone at hotel, cant use app"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"left my phone at hotel, cant use app\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"failed_transfer\": \"failed transfer\", \"verify_my_identity\": \"verify my identity\", \"pending_top_up\": \"pending top up\", \"transfer_into_account\": \"transfer into account\", \"lost_or_stolen_phone\": \"lost or stolen phone\", \"transaction_charged_twice\": \"transaction charged twice\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Identification of Unavailable Medical Information Through Various Systems\", \"message\": \"During standard procedures, it has been noted that medical information is not accessible. This could be due to recent software updates or changes in server configurations. After restarting the affected services and verifying user permissions, the access issues still remain. Please assist us in resolving this matter.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Identification of Unavailable Medical Information Through Various Systems\", \"message\": \"During standard procedures, it has been noted that medical information is not accessible. This could be due to recent software updates or changes in server configurations. After restarting the affected services and verifying user permissions, the access issues still remain. Please assist us in resolving this matter.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"IT & Technology/Hardware Support\": \"IT & Technology/Hardware Support\", \"IT Support\": \"IT Support\", \"Travel & Transportation/Land Travel\": \"Travel & Transportation/Land Travel\", \"Technical Support\": \"Technical Support\", \"Sports\": \"Sports\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Change\": \"Change\", \"Problem\": \"Problem\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Problem with Delayed Notification System\", \"message\": \"Dear Customer Support, <br> We are encountering issues with delayed notifications from the project management tool. The problem might have started after the recent server updates. Our team has already checked the network settings and restarted the application, but the issue still persists. We would greatly appreciate it if you could look into this matter and provide a solution as soon as possible. The delayed notifications are causing inconvenience for our users and affecting their productivity. Please let us know if we need to provide any additional information to resolve the issue.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Problem with Delayed Notification System\", \"message\": \"Dear Customer Support, <br> We are encountering issues with delayed notifications from the project management tool. The problem might have started after the recent server updates. Our team has already checked the network settings and restarted the application, but the issue still persists. We would greatly appreciate it if you could look into this matter and provide a solution as soon as possible. The delayed notifications are causing inconvenience for our users and affecting their productivity. Please let us know if we need to provide any additional information to resolve the issue.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"IT Support\": \"IT Support\", \"Health/Mental Health\": \"Health/Mental Health\", \"Science/Environmental Science\": \"Science/Environmental Science\", \"General Inquiry\": \"General Inquiry\", \"IT & Technology/Software Development\": \"IT & Technology/Software Development\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Request\": \"Request\", \"Incident\": \"Incident\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI would like to top up my account automatically. How do I do that?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I would like to top up my account automatically. How do I do that?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"visa_or_mastercard\": \"visa or mastercard\", \"pending_cash_withdrawal\": \"pending cash withdrawal\", \"automatic_top_up\": \"automatic top up\", \"reverted_card_payment?\": \"reverted card payment?\", \"get_disposable_virtual_card\": \"get disposable virtual card\", \"edit_personal_details\": \"edit personal details\", \"declined_cash_withdrawal\": \"declined cash withdrawal\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Enhancing Investment Portfolio Management with Analytics\", \"message\": \"Hello customer support, I am contacting you to seek advice on optimizing investment portfolios through the use of data analytics tools and services. Can you assist me in understanding how to use data analytics to make well-informed investment choices? I would be grateful for any suggestions or materials you could share. Thank you for your help.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Enhancing Investment Portfolio Management with Analytics\", \"message\": \"Hello customer support, I am contacting you to seek advice on optimizing investment portfolios through the use of data analytics tools and services. Can you assist me in understanding how to use data analytics to make well-informed investment choices? I would be grateful for any suggestions or materials you could share. Thank you for your help.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Online Communities/Social Networks\": \"Online Communities/Social Networks\", \"IT Support\": \"IT Support\", \"Technical Support\": \"Technical Support\", \"Pets & Animals/Veterinary Care\": \"Pets & Animals/Veterinary Care\", \"News\": \"News\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhy did I get a fee from the ATM?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Why did I get a fee from the ATM?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"cash_withdrawal_charge\": \"cash withdrawal charge\", \"fiat_currency_support\": \"fiat currency support\", \"reverted_card_payment?\": \"reverted card payment?\", \"verify_top_up\": \"verify top up\", \"atm_support\": \"atm support\", \"edit_personal_details\": \"edit personal details\", \"declined_transfer\": \"declined transfer\", \"card_payment_fee_charged\": \"card payment fee charged\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n1st movie of the night: The Vow"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"1st movie of the night: The Vow\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nexplain the virtual card"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"explain the virtual card\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"declined_transfer\": \"declined transfer\", \"cancel_transfer\": \"cancel transfer\", \"lost_or_stolen_phone\": \"lost or stolen phone\", \"pin_blocked\": \"pin blocked\", \"getting_virtual_card\": \"getting virtual card\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nim goin to Allderdice on friday!! seems like it be smackin."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"im goin to Allderdice on friday!! seems like it be smackin.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nCurtis 8th grader Justin Unger runs a PR to open his season with a 5:21 1600!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Curtis 8th grader Justin Unger runs a PR to open his season with a 5:21 1600!\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user Oh c\\u2019mon. Theres like two whole weeks between the SuperBowl and the Daytona 500. That\\u2019s all the quality time I have to offer."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user Oh c\\\\u2019mon. Theres like two whole weeks between the SuperBowl and the Daytona 500. That\\\\u2019s all the quality time I have to offer.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user This one will finish Dark Souls 2 while awaiting April 12th... Also, a lot of skooma to pass the time.\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user This one will finish Dark Souls 2 while awaiting April 12th... Also, a lot of skooma to pass the time.\\\"\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n\"Defeating BOKO Haram is no longer in two months, it is now by December (7 months). #ChangingCHANGI\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"\\\"Defeating BOKO Haram is no longer in two months, it is now by December (7 months). #ChangingCHANGI\\\"\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Detected Software Compatibility Problems\", \"message\": \"A financial organization is encountering software compatibility issues among its applications.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Detected Software Compatibility Problems\", \"message\": \"A financial organization is encountering software compatibility issues among its applications.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Food & Drink/Groceries\": \"Food & Drink/Groceries\", \"Product Support\": \"Product Support\", \"Science/Research\": \"Science/Research\", \"IT & Technology/Software Development\": \"IT & Technology/Software Development\", \"Shopping/E-commerce\": \"Shopping/E-commerce\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Boosting Digital Brand Growth\", \"message\": \"Could you please suggest effective digital strategies for small enterprises?\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Boosting Digital Brand Growth\", \"message\": \"Could you please suggest effective digital strategies for small enterprises?\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Pets & Animals/Veterinary Care\": \"Pets & Animals/Veterinary Care\", \"Shopping/Retail Stores\": \"Shopping/Retail Stores\", \"Customer Service\": \"Customer Service\", \"Hobbies & Leisure/Collectibles\": \"Hobbies & Leisure/Collectibles\", \"General Inquiry\": \"General Inquiry\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Incident\": \"Incident\", \"Change\": \"Change\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Integration Methods for Scalability Capabilities\", \"message\": \"Customer Support team, I am reaching out to inquire about the options available for integrating scalability features within your SaaS platform. As our business expands, it becomes crucial for our software tools to adapt accordingly. We would greatly appreciate if you could provide comprehensive documentation and guidance on how your product can scale to meet our growing needs. Specifically, we are interested in understanding the available integration methods, such as APIs, webhooks, and the protocols used to connect your platform with existing systems. Additionally,\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Integration Methods for Scalability Capabilities\", \"message\": \"Customer Support team, I am reaching out to inquire about the options available for integrating scalability features within your SaaS platform. As our business expands, it becomes crucial for our software tools to adapt accordingly. We would greatly appreciate if you could provide comprehensive documentation and guidance on how your product can scale to meet our growing needs. Specifically, we are interested in understanding the available integration methods, such as APIs, webhooks, and the protocols used to connect your platform with existing systems. Additionally,\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Autos & Vehicles/Sales\": \"Autos & Vehicles/Sales\", \"Shopping/E-commerce\": \"Shopping/E-commerce\", \"Online Communities/Forums\": \"Online Communities/Forums\", \"Product Support\": \"Product Support\", \"Technical Support\": \"Technical Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user I had lot of respect for u when u came to power.But in pursuit of power,Delhi have fast become a Slum,coz u r encouraging illegal migrants,encroachments.Ur naked Appeasement of Muslims is highly condemnable.URGE U TO VISIT NIZZM TILL INDIA"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user I had lot of respect for u when u came to power.But in pursuit of power,Delhi have fast become a Slum,coz u r encouraging illegal migrants,encroachments.Ur naked Appeasement of Muslims is highly condemnable.URGE U TO VISIT NIZZM TILL INDIA\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nanother day, another “mets interested in player who will take away [NAME] playing time!!”"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"another day, another “mets interested in player who will take away [NAME] playing time!!”\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user @user @user We don't want our countries to be over run with rapefugee parasite Moslem scum. You already have 57 countries, stay in those countries. Those countries suck because they are full of your type. #StopIslam #BanSh"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user @user @user We don't want our countries to be over run with rapefugee parasite Moslem scum. You already have 57 countries, stay in those countries. Those countries suck because they are full of your type. #StopIslam #BanSh\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Urgently Improve Hospital Data Security\", \"message\": \"Put in place strong encryption methods to safeguard medical data and bolster hospital security procedures. This involves securing electronic health records, defending against cyber threats, and adhering to compliance rules. Regular security evaluations and staff training are crucial. By taking these measures, hospitals can safeguard sensitive patient data and preserve trust.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Urgently Improve Hospital Data Security\", \"message\": \"Put in place strong encryption methods to safeguard medical data and bolster hospital security procedures. This involves securing electronic health records, defending against cyber threats, and adhering to compliance rules. Regular security evaluations and staff training are crucial. By taking these measures, hospitals can safeguard sensitive patient data and preserve trust.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Finance/Investments\": \"Finance/Investments\", \"Technical Support\": \"Technical Support\", \"Customer Service\": \"Customer Service\", \"Returns and Exchanges\": \"Returns and Exchanges\", \"Home & Garden/Home Improvement\": \"Home & Garden/Home Improvement\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nMy PIN number was incorrect, and I can't access it."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"My PIN number was incorrect, and I can't access it.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"direct_debit_payment_not_recognised\": \"direct debit payment not recognised\", \"transfer_fee_charged\": \"transfer fee charged\", \"apple_pay_or_google_pay\": \"apple pay or google pay\", \"balance_not_updated_after_cheque_or_cash_deposit\": \"balance not updated after cheque or cash deposit\", \"cash_withdrawal_charge\": \"cash withdrawal charge\", \"top_up_reverted\": \"top up reverted\", \"pin_blocked\": \"pin blocked\", \"declined_card_payment\": \"declined card payment\", \"verify_top_up\": \"verify top up\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Trouble with Video Lag in Virtual Meetings\", \"message\": \"The financial analytics team is facing video lag and audio issues during virtual meetings. We have already updated drivers, restarted devices, and adjusted settings, but the problem continues to affect performance. To assist further, could you provide details about the devices and software used for the meetings, and the type of internet connection you are using? It would also be helpful to schedule a call to discuss this issue.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Trouble with Video Lag in Virtual Meetings\", \"message\": \"The financial analytics team is facing video lag and audio issues during virtual meetings. We have already updated drivers, restarted devices, and adjusted settings, but the problem continues to affect performance. To assist further, could you provide details about the devices and software used for the meetings, and the type of internet connection you are using? It would also be helpful to schedule a call to discuss this issue.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"IT Support\": \"IT Support\", \"Technical Support\": \"Technical Support\", \"Online Communities/Social Networks\": \"Online Communities/Social Networks\", \"IT & Technology/Software Development\": \"IT & Technology/Software Development\", \"Travel & Transportation/Air Travel\": \"Travel & Transportation/Air Travel\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAnyone need a ticket for Sunday the 13th's White Sox game??? Potential K-Zone game. $15 if interested."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Anyone need a ticket for Sunday the 13th's White Sox game??? Potential K-Zone game. $15 if interested.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHaha okay. Much love brother!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Haha okay. Much love brother!\", \"questions\": {\"emotion_amusement\": {\"type\": \"noul\", \"instructions\": \"Does the author express amusement?\"}, \"emotion_love\": {\"type\": \"noul\", \"instructions\": \"Does the author express love?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhat other fiat currencies can I use?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"What other fiat currencies can I use?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"fiat_currency_support\": \"fiat currency support\", \"get_physical_card\": \"get physical card\", \"contactless_not_working\": \"contactless not working\", \"age_limit\": \"age limit\", \"get_disposable_virtual_card\": \"get disposable virtual card\", \"card_about_to_expire\": \"card about to expire\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nExactly. Nothing but a burden at first, but our boy knows how to fight now."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Exactly. Nothing but a burden at first, but our boy knows how to fight now.\", \"questions\": {\"emotion_approval\": {\"type\": \"noul\", \"instructions\": \"Does the author express approval?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nYet another PB today in race out by CIT. This time at 10 km distance. Finished in 51.27. Finished with races now until half marathon in Dec."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Yet another PB today in race out by CIT. This time at 10 km distance. Finished in 51.27. Finished with races now until half marathon in Dec.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nMmmm might go have one sometime!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Mmmm might go have one sometime!\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nBlum & Poe just sent out email announcement of Chuck Close's show opening on the 29th!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Blum & Poe just sent out email announcement of Chuck Close's show opening on the 29th!\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAgreed. If anyone was \"murdered by words\" it was all the posts, including the top posts that make [NAME] look like an idiot."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Agreed. If anyone was \\\"murdered by words\\\" it was all the posts, including the top posts that make [NAME] look like an idiot.\", \"questions\": {\"emotion_approval\": {\"type\": \"noul\", \"instructions\": \"Does the author express approval?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nRT @user #meninists #yesallmen"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"RT @user #meninists #yesallmen\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user We can\\u2019t wait to see what the cast members at Capt. EO do when they see me tomorrow in my costume."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user We can\\\\u2019t wait to see what the cast members at Capt. EO do when they see me tomorrow in my costume.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nLast I checked people weren't running around raping people wearing MAGA hats."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Last I checked people weren't running around raping people wearing MAGA hats.\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nDo you send more than one card?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Do you send more than one card?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"compromised_card\": \"compromised card\", \"getting_spare_card\": \"getting spare card\", \"extra_charge_on_statement\": \"extra charge on statement\", \"request_refund\": \"request refund\", \"lost_or_stolen_card\": \"lost or stolen card\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThat face screams \"Fucking gods, do I have to do *everything* around here?\""}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"That face screams \\\"Fucking gods, do I have to do *everything* around here?\\\"\", \"questions\": {\"emotion_annoyance\": {\"type\": \"noul\", \"instructions\": \"Does the author express annoyance?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhy is verifaction required for my Identity?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Why is verifaction required for my Identity?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"exchange_charge\": \"exchange charge\", \"card_payment_fee_charged\": \"card payment fee charged\", \"direct_debit_payment_not_recognised\": \"direct debit payment not recognised\", \"pending_card_payment\": \"pending card payment\", \"contactless_not_working\": \"contactless not working\", \"passcode_forgotten\": \"passcode forgotten\", \"why_verify_identity\": \"why verify identity\", \"wrong_amount_of_cash_received\": \"wrong amount of cash received\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIona is off tomorrow so like ... damn tus you fuckin up. #thisbetterbeajoke"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Iona is off tomorrow so like ... damn tus you fuckin up. #thisbetterbeajoke\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIs my new bank card on the way and is there a way to track it?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Is my new bank card on the way and is there a way to track it?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"card_linking\": \"card linking\", \"country_support\": \"country support\", \"exchange_charge\": \"exchange charge\", \"exchange_via_app\": \"exchange via app\", \"getting_spare_card\": \"getting spare card\", \"fiat_currency_support\": \"fiat currency support\", \"card_arrival\": \"card arrival\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n🇲🇦#Morocco : a desire to counselling African illegal #migrants not to risking their lives for reaching #Europe 🇪🇺https://t.co/oqmIcku2Lb"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"🇲🇦#Morocco : a desire to counselling African illegal #migrants not to risking their lives for reaching #Europe 🇪🇺https://t.co/oqmIcku2Lb\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThat gives you zero right or say over what someone wears. You are wrong."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"That gives you zero right or say over what someone wears. You are wrong.\", \"questions\": {\"emotion_disapproval\": {\"type\": \"noul\", \"instructions\": \"Does the author express disapproval?\"}, \"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Reported Issue of Software Crashes\", \"message\": \"Hello Customer Support, I am contacting you to report a recurring issue with software crashes that are happening simultaneously. It might be due to compatibility issues. Despite my efforts to resolve it by restarting, updating, and reinstalling the software, the problem still exists. I would greatly appreciate your help in investigating this and providing a resolution at the earliest. These crashes are significantly affecting my work and I am keen to resolve this issue. Please inform me if you need any additional information from me to assist with the troubleshooting. Thank you for your time and assistance in this matter.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Reported Issue of Software Crashes\", \"message\": \"Hello Customer Support, I am contacting you to report a recurring issue with software crashes that are happening simultaneously. It might be due to compatibility issues. Despite my efforts to resolve it by restarting, updating, and reinstalling the software, the problem still exists. I would greatly appreciate your help in investigating this and providing a resolution at the earliest. These crashes are significantly affecting my work and I am keen to resolve this issue. Please inform me if you need any additional information from me to assist with the troubleshooting. Thank you for your time and assistance in this matter.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Travel & Transportation/Land Travel\": \"Travel & Transportation/Land Travel\", \"Autos & Vehicles/Sales\": \"Autos & Vehicles/Sales\", \"General Inquiry\": \"General Inquiry\", \"News\": \"News\", \"Technical Support\": \"Technical Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Request\": \"Request\", \"Change\": \"Change\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI'll probably do the same thing next year that I did this year and that was targeting [NAME] pretty late"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I'll probably do the same thing next year that I did this year and that was targeting [NAME] pretty late\", \"questions\": {\"emotion_approval\": {\"type\": \"noul\", \"instructions\": \"Does the author express approval?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIt's cool outside and I've got AC/DC playing inside. This is how you adult on a Saturday night."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"It's cool outside and I've got AC/DC playing inside. This is how you adult on a Saturday night.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHow can I close my account?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"How can I close my account?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"transaction_charged_twice\": \"transaction charged twice\", \"declined_transfer\": \"declined transfer\", \"transfer_into_account\": \"transfer into account\", \"direct_debit_payment_not_recognised\": \"direct debit payment not recognised\", \"exchange_via_app\": \"exchange via app\", \"why_verify_identity\": \"why verify identity\", \"terminate_account\": \"terminate account\", \"failed_transfer\": \"failed transfer\", \"top_up_by_card_charge\": \"top up by card charge\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhy does my transfer keep generating an error message?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Why does my transfer keep generating an error message?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"card_not_working\": \"card not working\", \"visa_or_mastercard\": \"visa or mastercard\", \"beneficiary_not_allowed\": \"beneficiary not allowed\", \"verify_top_up\": \"verify top up\", \"direct_debit_payment_not_recognised\": \"direct debit payment not recognised\", \"lost_or_stolen_phone\": \"lost or stolen phone\", \"pending_top_up\": \"pending top up\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user love the Greenbrier team. Tipped two of them and possibly Harris English\\u2019s biggest fan. May yet add him to my staking plan."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user love the Greenbrier team. Tipped two of them and possibly Harris English\\\\u2019s biggest fan. May yet add him to my staking plan.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThis is a Harper worth looking forward to. Go Set a Watchman's 1st chapter stuns fans with bombshell plot twist"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"This is a Harper worth looking forward to. Go Set a Watchman's 1st chapter stuns fans with bombshell plot twist\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nSteve Emerson on SUN explains the riots and real motives"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Steve Emerson on SUN explains the riots and real motives\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user Hahahahaha so hysterical she the real living proof cher cherry picker we have only bitch read da real KSU SB Hit! I Alo!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user Hahahahaha so hysterical she the real living proof cher cherry picker we have only bitch read da real KSU SB Hit! I Alo!\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHello Everybody!!!How are you today?Have a Beautiful Green Day!!!!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Hello Everybody!!!How are you today?Have a Beautiful Green Day!!!!\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI just saw a 3rd grader talking on her I phone 6... Wtf is wrong with this world. I would be lucky to have a Nokia in 3rd grade..."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I just saw a 3rd grader talking on her I phone 6... Wtf is wrong with this world. I would be lucky to have a Nokia in 3rd grade...\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Issue with Unauthorized Access to Medical Records\", \"message\": \"The healthcare organization has flagged a possible breach involving access to medical information, likely due to obsolete firewall configurations. Our team has initiated password resets and conducted system scans in an effort to resolve the problem.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Issue with Unauthorized Access to Medical Records\", \"message\": \"The healthcare organization has flagged a possible breach involving access to medical information, likely due to obsolete firewall configurations. Our team has initiated password resets and conducted system scans in an effort to resolve the problem.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Customer Service\": \"Customer Service\", \"Games\": \"Games\", \"IT & Technology/Hardware Support\": \"IT & Technology/Hardware Support\", \"Hobbies & Leisure/Collectibles\": \"Hobbies & Leisure/Collectibles\", \"General Inquiry\": \"General Inquiry\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Request\": \"Request\", \"Incident\": \"Incident\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI can't see all the refunds in my statement"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I can't see all the refunds in my statement\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"transfer_timing\": \"transfer timing\", \"getting_virtual_card\": \"getting virtual card\", \"fiat_currency_support\": \"fiat currency support\", \"transfer_not_received_by_recipient\": \"transfer not received by recipient\", \"Refund_not_showing_up\": \"Refund not showing up\", \"balance_not_updated_after_cheque_or_cash_deposit\": \"balance not updated after cheque or cash deposit\", \"balance_not_updated_after_bank_transfer\": \"balance not updated after bank transfer\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user Thank you my love! And I have had a fab day! Distinct lack of Hannibal but I'll put that right tomorrow! xx"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user Thank you my love! And I have had a fab day! Distinct lack of Hannibal but I'll put that right tomorrow! xx\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Concern Over Security Breach in Medical Data\", \"message\": \"A potential security incident at the healthcare provider has led to outdated access controls, resulting in observed impacts. Initial measures such as network scans, review of access logs, and system patching have been taken, but the issues remain unresolved.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Concern Over Security Breach in Medical Data\", \"message\": \"A potential security incident at the healthcare provider has led to outdated access controls, resulting in observed impacts. Initial measures such as network scans, review of access logs, and system patching have been taken, but the issues remain unresolved.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Service Outages and Maintenance\": \"Service Outages and Maintenance\", \"Returns and Exchanges\": \"Returns and Exchanges\", \"IT & Technology/Network Infrastructure\": \"IT & Technology/Network Infrastructure\", \"Pets & Animals/Pet Services\": \"Pets & Animals/Pet Services\", \"Technical Support\": \"Technical Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Request\": \"Request\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Frequent Disconnections in Software During Peak Hours\", \"message\": \"Users encounter disconnections during peak traffic times. Despite optimizing database queries and increasing server capacity, the issues remain. A follow-up request for more information on disconnection error messages and their frequency is needed. Also, a call at a convenient time to discuss and resolve the issue is requested. Please provide a suitable time as <tel_num>.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Frequent Disconnections in Software During Peak Hours\", \"message\": \"Users encounter disconnections during peak traffic times. Despite optimizing database queries and increasing server capacity, the issues remain. A follow-up request for more information on disconnection error messages and their frequency is needed. Also, a call at a convenient time to discuss and resolve the issue is requested. Please provide a suitable time as <tel_num>.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Arts & Entertainment/Music\": \"Arts & Entertainment/Music\", \"Arts & Entertainment/Movies\": \"Arts & Entertainment/Movies\", \"Service Outages and Maintenance\": \"Service Outages and Maintenance\", \"Technical Support\": \"Technical Support\", \"Law & Government/Government Services\": \"Law & Government/Government Services\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIs it possible to change my PIN?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Is it possible to change my PIN?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"change_pin\": \"change pin\", \"get_disposable_virtual_card\": \"get disposable virtual card\", \"request_refund\": \"request refund\", \"fiat_currency_support\": \"fiat currency support\", \"compromised_card\": \"compromised card\", \"direct_debit_payment_not_recognised\": \"direct debit payment not recognised\", \"card_swallowed\": \"card swallowed\", \"atm_support\": \"atm support\", \"transfer_not_received_by_recipient\": \"transfer not received by recipient\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI think the exchange rate for the cash I withdrew was wrong."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I think the exchange rate for the cash I withdrew was wrong.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"automatic_top_up\": \"automatic top up\", \"card_about_to_expire\": \"card about to expire\", \"wrong_exchange_rate_for_cash_withdrawal\": \"wrong exchange rate for cash withdrawal\", \"pending_transfer\": \"pending transfer\", \"verify_source_of_funds\": \"verify source of funds\", \"exchange_charge\": \"exchange charge\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nSaudi Arabia to buy 30 An-178 planes for its air force That's a nice chunk of change for #Ukraine They can use it."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Saudi Arabia to buy 30 An-178 planes for its air force That's a nice chunk of change for #Ukraine They can use it.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nCA - well now. Then your votes wont count if you allow illegal aliens to vote. THAT IS AN AMERICAN PRIVILEGE!! #BuildThatWall"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"CA - well now. Then your votes wont count if you allow illegal aliens to vote. THAT IS AN AMERICAN PRIVILEGE!! #BuildThatWall\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n*Fires back* hoe ass bitch with a big ass forehead 😭 I knowwwwww 😭😭😭"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"*Fires back* hoe ass bitch with a big ass forehead 😭 I knowwwwww 😭😭😭\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhat is the status of my top up?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"What is the status of my top up?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"atm_support\": \"atm support\", \"exchange_rate\": \"exchange rate\", \"verify_source_of_funds\": \"verify source of funds\", \"top_up_reverted\": \"top up reverted\", \"lost_or_stolen_phone\": \"lost or stolen phone\", \"terminate_account\": \"terminate account\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhy did I get charged more than once?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Why did I get charged more than once?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"card_swallowed\": \"card swallowed\", \"declined_cash_withdrawal\": \"declined cash withdrawal\", \"passcode_forgotten\": \"passcode forgotten\", \"beneficiary_not_allowed\": \"beneficiary not allowed\", \"visa_or_mastercard\": \"visa or mastercard\", \"transaction_charged_twice\": \"transaction charged twice\", \"verify_top_up\": \"verify top up\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIt's the right sub but why did we even set-up so defensively in the first place? Utterly bizarre."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"It's the right sub but why did we even set-up so defensively in the first place? Utterly bizarre.\", \"questions\": {\"emotion_curiosity\": {\"type\": \"noul\", \"instructions\": \"Does the author express curiosity?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nMy bad! It looked so real XD thanks for the heads up"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"My bad! It looked so real XD thanks for the heads up\", \"questions\": {\"emotion_gratitude\": {\"type\": \"noul\", \"instructions\": \"Does the author express gratitude?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI got charged twice for the same payment. Please removed the pending payment. Thank you."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I got charged twice for the same payment. Please removed the pending payment. Thank you.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"top_up_by_bank_transfer_charge\": \"top up by bank transfer charge\", \"transaction_charged_twice\": \"transaction charged twice\", \"pending_top_up\": \"pending top up\", \"card_linking\": \"card linking\", \"activate_my_card\": \"activate my card\", \"transfer_timing\": \"transfer timing\", \"top_up_limits\": \"top up limits\", \"supported_cards_and_currencies\": \"supported cards and currencies\", \"exchange_rate\": \"exchange rate\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Enquiry about Data Analytics Services\", \"message\": \"I am interested in learning about your data analytics services to optimize my investments. Could you provide details on the types of services you offer to assist in making informed investment decisions? I would greatly appreciate any information you can provide regarding your services.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Enquiry about Data Analytics Services\", \"message\": \"I am interested in learning about your data analytics services to optimize my investments. Could you provide details on the types of services you offer to assist in making informed investment decisions? I would greatly appreciate any information you can provide regarding your services.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Billing and Payments\": \"Billing and Payments\", \"Beauty & Fitness/Cosmetics\": \"Beauty & Fitness/Cosmetics\", \"Home & Garden/Landscaping\": \"Home & Garden/Landscaping\", \"Law & Government/Legal Advice\": \"Law & Government/Legal Advice\", \"Sports\": \"Sports\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Change\": \"Change\", \"Request\": \"Request\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n\\u2019Saturday Kitchen\\u2019 is on and it\\u2019s Sunday. Mind = fucked."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"\\\\u2019Saturday Kitchen\\\\u2019 is on and it\\\\u2019s Sunday. Mind = fucked.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI want Gatekeeper to dj every event I go to. From here on out (@ 7th Street Entry for Teengirl Fantasy w/ @user"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I want Gatekeeper to dj every event I go to. From here on out (@ 7th Street Entry for Teengirl Fantasy w/ @user\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThis is a seriously good idea."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"This is a seriously good idea.\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user Hey illegal alien invader Hugger @user #StopTheInvasion#DeportThemAll#NoAmnesty#BuildTheWall"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user Hey illegal alien invader Hugger @user #StopTheInvasion#DeportThemAll#NoAmnesty#BuildTheWall\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user you're a pussy, stfu"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user you're a pussy, stfu\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhy am I being charged a fee on a card payment?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Why am I being charged a fee on a card payment?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"card_payment_fee_charged\": \"card payment fee charged\", \"getting_virtual_card\": \"getting virtual card\", \"pending_cash_withdrawal\": \"pending cash withdrawal\", \"pending_transfer\": \"pending transfer\", \"compromised_card\": \"compromised card\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Trouble with Integration\", \"message\": \"Facing issues with data discrepancies in investment reports. The problem seems to stem from integration issues between SAP ERP and RapidMiner, which might be due to recent software updates or configuration errors. After multiple attempts to recalibrate settings and rerun analyses, the issue still persists. I need your help to resolve it.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Trouble with Integration\", \"message\": \"Facing issues with data discrepancies in investment reports. The problem seems to stem from integration issues between SAP ERP and RapidMiner, which might be due to recent software updates or configuration errors. After multiple attempts to recalibrate settings and rerun analyses, the issue still persists. I need your help to resolve it.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Business & Industrial/Manufacturing\": \"Business & Industrial/Manufacturing\", \"Customer Service\": \"Customer Service\", \"Online Communities/Forums\": \"Online Communities/Forums\", \"People & Society/Culture & Society\": \"People & Society/Culture & Society\", \"Technical Support\": \"Technical Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Request\": \"Request\", \"Problem\": \"Problem\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Integrate Airtable with Discord\", \"message\": \"I am interested in setting up an integration between Airtable and Discord to receive project updates and notifications on my Dell UltraSharp U2720Q display. Could you please inform me about the steps involved and any potential limitations? It would be helpful if you could provide a step-by-step guide or direct me to a resource that can assist me in achieving this integration.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Integrate Airtable with Discord\", \"message\": \"I am interested in setting up an integration between Airtable and Discord to receive project updates and notifications on my Dell UltraSharp U2720Q display. Could you please inform me about the steps involved and any potential limitations? It would be helpful if you could provide a step-by-step guide or direct me to a resource that can assist me in achieving this integration.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Autos & Vehicles/Sales\": \"Autos & Vehicles/Sales\", \"Real Estate\": \"Real Estate\", \"Online Communities/Forums\": \"Online Communities/Forums\", \"Product Support\": \"Product Support\", \"Sports\": \"Sports\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Request\": \"Request\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Digital Marketing Campaign No Longer Functional\", \"message\": \"The agency's digital marketing campaign has unexpectedly stopped working, which might be due to a technical glitch or server issue. Initial troubleshooting steps, such as restarting platforms, clearing caches, and verifying connection setups, were taken but were not successful in resolving the issue. We require your assistance to identify the root cause of the problem and provide guidance on the next steps to take. We appreciate your prompt attention to this matter, as the continued disruption of the campaign could impact business operations.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Digital Marketing Campaign No Longer Functional\", \"message\": \"The agency's digital marketing campaign has unexpectedly stopped working, which might be due to a technical glitch or server issue. Initial troubleshooting steps, such as restarting platforms, clearing caches, and verifying connection setups, were taken but were not successful in resolving the issue. We require your assistance to identify the root cause of the problem and provide guidance on the next steps to take. We appreciate your prompt attention to this matter, as the continued disruption of the campaign could impact business operations.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Business & Industrial/Manufacturing\": \"Business & Industrial/Manufacturing\", \"Shopping/Retail Stores\": \"Shopping/Retail Stores\", \"IT Support\": \"IT Support\", \"Returns and Exchanges\": \"Returns and Exchanges\", \"Sports\": \"Sports\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIf you're real. Go to disneyland tomorrow"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"If you're real. Go to disneyland tomorrow\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHow can I receive my salary in another currency?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"How can I receive my salary in another currency?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"card_arrival\": \"card arrival\", \"top_up_by_cash_or_cheque\": \"top up by cash or cheque\", \"pin_blocked\": \"pin blocked\", \"top_up_reverted\": \"top up reverted\", \"request_refund\": \"request refund\", \"disposable_card_limits\": \"disposable card limits\", \"automatic_top_up\": \"automatic top up\", \"receiving_money\": \"receiving money\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI think I was double charged"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I think I was double charged\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"transaction_charged_twice\": \"transaction charged twice\", \"declined_card_payment\": \"declined card payment\", \"topping_up_by_card\": \"topping up by card\", \"passcode_forgotten\": \"passcode forgotten\", \"pending_card_payment\": \"pending card payment\", \"top_up_by_card_charge\": \"top up by card charge\", \"age_limit\": \"age limit\", \"getting_virtual_card\": \"getting virtual card\", \"wrong_amount_of_cash_received\": \"wrong amount of cash received\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThey are free to go back home!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"They are free to go back home!\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nOur Daily Planet: EPA ignoring air pollution, immigrant fire crews battle fires in Trump country and an orca mom just can't let go (get your tissues!):"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Our Daily Planet: EPA ignoring air pollution, immigrant fire crews battle fires in Trump country and an orca mom just can't let go (get your tissues!):\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Inquiry About Data Analytics Services and Investment Optimization\", \"message\": \"Would be grateful for a detailed overview of the data analytics services, particularly those related to investment optimization. This information would greatly assist us in formulating our investment plans. Thank you for your prompt assistance with this matter. We look forward to hearing from you soon.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Inquiry About Data Analytics Services and Investment Optimization\", \"message\": \"Would be grateful for a detailed overview of the data analytics services, particularly those related to investment optimization. This information would greatly assist us in formulating our investment plans. Thank you for your prompt assistance with this matter. We look forward to hearing from you soon.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Beauty & Fitness/Fitness Training\": \"Beauty & Fitness/Fitness Training\", \"Billing and Payments\": \"Billing and Payments\", \"Pets & Animals/Pet Services\": \"Pets & Animals/Pet Services\", \"Games\": \"Games\", \"Jobs & Education/Online Courses\": \"Jobs & Education/Online Courses\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nCRP doesn't only serve refugees living in Jordan. We also offer assistance and programming to Jordanians living in the area. Learn more about our diverse community in this blog post:"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"CRP doesn't only serve refugees living in Jordan. We also offer assistance and programming to Jordanians living in the area. Learn more about our diverse community in this blog post:\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n#BuildThatWall is all about homeland security and an open border clearly leaves Americans less safe. Makes sense to use our military to protect us by building a wall."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"#BuildThatWall is all about homeland security and an open border clearly leaves Americans less safe. Makes sense to use our military to protect us by building a wall.\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAlright\\u002c just woke up\\u002c about to go ball. Open Lobby tonight and tomorrow\\u002c going to be going for crazy 100+s! *cough* Desert Eagle *cough* ;)"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Alright\\\\u002c just woke up\\\\u002c about to go ball. Open Lobby tonight and tomorrow\\\\u002c going to be going for crazy 100+s! *cough* Desert Eagle *cough* ;)\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHow long does it take to activate my card?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"How long does it take to activate my card?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"activate_my_card\": \"activate my card\", \"why_verify_identity\": \"why verify identity\", \"change_pin\": \"change pin\", \"apple_pay_or_google_pay\": \"apple pay or google pay\", \"cash_withdrawal_not_recognised\": \"cash withdrawal not recognised\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user And the son of a bitch can't buy it for his little pee pee whore boy and that just rankles his ass. Shut u…"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user And the son of a bitch can't buy it for his little pee pee whore boy and that just rankles his ass. Shut u…\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nDo you have a return policy?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Do you have a return policy?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"reverted_card_payment?\": \"reverted card payment?\", \"wrong_amount_of_cash_received\": \"wrong amount of cash received\", \"balance_not_updated_after_cheque_or_cash_deposit\": \"balance not updated after cheque or cash deposit\", \"cash_withdrawal_charge\": \"cash withdrawal charge\", \"pending_top_up\": \"pending top up\", \"lost_or_stolen_card\": \"lost or stolen card\", \"getting_spare_card\": \"getting spare card\", \"request_refund\": \"request refund\", \"transfer_timing\": \"transfer timing\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user That may explain the problem we're having with Muslims fundamentalists."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user That may explain the problem we're having with Muslims fundamentalists.\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nOK OK WE GOT IT"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"OK OK WE GOT IT\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n\"10 Best Ideas of the Week: The ESPY Awards, 'Project Runway' 10th Anniversary Party, Calgary Stampede"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"\\\"10 Best Ideas of the Week: The ESPY Awards, 'Project Runway' 10th Anniversary Party, Calgary Stampede\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nGreat, so we know the spanish pronounciation is therefore racist and now I'm offended by it."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Great, so we know the spanish pronounciation is therefore racist and now I'm offended by it.\", \"questions\": {\"emotion_disappointment\": {\"type\": \"noul\", \"instructions\": \"Does the author express disappointment?\"}, \"emotion_realization\": {\"type\": \"noul\", \"instructions\": \"Does the author express realization?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nCan someone else use my account if they have my phone? I lost mine."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Can someone else use my account if they have my phone? I lost mine.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"exchange_charge\": \"exchange charge\", \"top_up_by_bank_transfer_charge\": \"top up by bank transfer charge\", \"get_disposable_virtual_card\": \"get disposable virtual card\", \"top_up_limits\": \"top up limits\", \"exchange_rate\": \"exchange rate\", \"beneficiary_not_allowed\": \"beneficiary not allowed\", \"cash_withdrawal_not_recognised\": \"cash withdrawal not recognised\", \"passcode_forgotten\": \"passcode forgotten\", \"lost_or_stolen_phone\": \"lost or stolen phone\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI make my cows super stronk."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I make my cows super stronk.\", \"questions\": {\"emotion_admiration\": {\"type\": \"noul\", \"instructions\": \"Does the author express admiration?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nFake rape bitch is still crying. LOL so upset he didn’t get the attention he wanted. Got your address btw little ra…"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Fake rape bitch is still crying. LOL so upset he didn’t get the attention he wanted. Got your address btw little ra…\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI'm gonna call out of work tomorrow because Kane turns 1 and it's a big deal #catmom #gotmypictureforinstagramready"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I'm gonna call out of work tomorrow because Kane turns 1 and it's a big deal #catmom #gotmypictureforinstagramready\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"IBM SPSS Assistance Required\", \"message\": \"Could you please offer comprehensive instructions for integrating the IBM SPSS Statistics 28 data analytics platform?\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"IBM SPSS Assistance Required\", \"message\": \"Could you please offer comprehensive instructions for integrating the IBM SPSS Statistics 28 data analytics platform?\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Finance/Personal Finance\": \"Finance/Personal Finance\", \"Online Communities/Social Networks\": \"Online Communities/Social Networks\", \"Customer Service\": \"Customer Service\", \"Autos & Vehicles/Maintenance\": \"Autos & Vehicles/Maintenance\", \"Law & Government/Government Services\": \"Law & Government/Government Services\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHeyyy that's the chair i'm sitting in right now at my home office!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Heyyy that's the chair i'm sitting in right now at my home office!\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWHO DAT!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"WHO DAT!\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\ni'm listening to nirvana and drinking wine from a mug how's your wednesday night going"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"i'm listening to nirvana and drinking wine from a mug how's your wednesday night going\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n‘They knew what they were doing,’ immigrant arrested in Bastrop says"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"‘They knew what they were doing,’ immigrant arrested in Bastrop says\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nMy card payment failed."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"My card payment failed.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"exchange_via_app\": \"exchange via app\", \"lost_or_stolen_phone\": \"lost or stolen phone\", \"declined_card_payment\": \"declined card payment\", \"card_payment_fee_charged\": \"card payment fee charged\", \"card_payment_not_recognised\": \"card payment not recognised\", \"disposable_card_limits\": \"disposable card limits\", \"top_up_by_cash_or_cheque\": \"top up by cash or cheque\", \"wrong_exchange_rate_for_cash_withdrawal\": \"wrong exchange rate for cash withdrawal\", \"wrong_amount_of_cash_received\": \"wrong amount of cash received\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI think my card is broken or something. I tried using it and it wouldn't go through, please help."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I think my card is broken or something. I tried using it and it wouldn't go through, please help.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"transaction_charged_twice\": \"transaction charged twice\", \"lost_or_stolen_card\": \"lost or stolen card\", \"beneficiary_not_allowed\": \"beneficiary not allowed\", \"exchange_via_app\": \"exchange via app\", \"card_arrival\": \"card arrival\", \"cash_withdrawal_charge\": \"cash withdrawal charge\", \"card_not_working\": \"card not working\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThe wrong account was used during a transaction."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"The wrong account was used during a transaction.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"fiat_currency_support\": \"fiat currency support\", \"transfer_fee_charged\": \"transfer fee charged\", \"cancel_transfer\": \"cancel transfer\", \"card_acceptance\": \"card acceptance\", \"top_up_by_cash_or_cheque\": \"top up by cash or cheque\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI agree, I love acdc, I just don't love the comments on their songs cause so many of them are dumb lmao"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I agree, I love acdc, I just don't love the comments on their songs cause so many of them are dumb lmao\", \"questions\": {\"emotion_approval\": {\"type\": \"noul\", \"instructions\": \"Does the author express approval?\"}, \"emotion_love\": {\"type\": \"noul\", \"instructions\": \"Does the author express love?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nBecause it seems like a first step on the same road that peope here have taken. I figured you have experience in the matter."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Because it seems like a first step on the same road that peope here have taken. I figured you have experience in the matter.\", \"questions\": {\"emotion_realization\": {\"type\": \"noul\", \"instructions\": \"Does the author express realization?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nCould you tell me what your exchange rates are?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Could you tell me what your exchange rates are?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"card_payment_not_recognised\": \"card payment not recognised\", \"cash_withdrawal_charge\": \"cash withdrawal charge\", \"receiving_money\": \"receiving money\", \"card_linking\": \"card linking\", \"country_support\": \"country support\", \"topping_up_by_card\": \"topping up by card\", \"exchange_rate\": \"exchange rate\", \"apple_pay_or_google_pay\": \"apple pay or google pay\", \"transaction_charged_twice\": \"transaction charged twice\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhile Sweden Wait for 300,000 new Muslim migrants, Islam Radicalism Spikes"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"While Sweden Wait for 300,000 new Muslim migrants, Islam Radicalism Spikes\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThere is a direct debit charge that I don't recognize, what can I do?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"There is a direct debit charge that I don't recognize, what can I do?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"get_physical_card\": \"get physical card\", \"exchange_via_app\": \"exchange via app\", \"compromised_card\": \"compromised card\", \"cash_withdrawal_not_recognised\": \"cash withdrawal not recognised\", \"card_acceptance\": \"card acceptance\", \"order_physical_card\": \"order physical card\", \"direct_debit_payment_not_recognised\": \"direct debit payment not recognised\", \"transfer_not_received_by_recipient\": \"transfer not received by recipient\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIllegal Alien With Long Criminal Record Murdered Man In North Carolina via @user #EnforceUSLaws#EndDaca #BanSanctuaryCities#KeepAmericansSafe"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Illegal Alien With Long Criminal Record Murdered Man In North Carolina via @user #EnforceUSLaws#EndDaca #BanSanctuaryCities#KeepAmericansSafe\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Required Assistance with Digital Marketing Strategies\", \"message\": \"Our agency has noticed a decrease in client engagement. After updating content and analyzing analytics, we observed some improvements.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Required Assistance with Digital Marketing Strategies\", \"message\": \"Our agency has noticed a decrease in client engagement. After updating content and analyzing analytics, we observed some improvements.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Travel & Transportation/Land Travel\": \"Travel & Transportation/Land Travel\", \"Returns and Exchanges\": \"Returns and Exchanges\", \"Hobbies & Leisure/Crafts\": \"Hobbies & Leisure/Crafts\", \"Sports\": \"Sports\", \"Customer Service\": \"Customer Service\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Problem\": \"Problem\", \"Incident\": \"Incident\", \"Request\": \"Request\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user Man: (harasses and stalks woman for months on end) Woman: WHY THE FUCK WONT U LEAVE ME ALONE Man: haha crazy bitch"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user Man: (harasses and stalks woman for months on end) Woman: WHY THE FUCK WONT U LEAVE ME ALONE Man: haha crazy bitch\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nNo because they’re your property on your property."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"No because they’re your property on your property.\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Strategies for Safeguarding Medical Data Systems\", \"message\": \"Inquiring about methods to secure medical data with Alteryx Oracle Database 19c in hospital settings. Could you provide the necessary steps to ensure the confidentiality, integrity, and availability of sensitive medical information? I would greatly appreciate guidance on data encryption, access controls, and auditing mechanisms. Additionally, could you recommend specific configuration settings for Alteryx Oracle Database 19c to meet healthcare regulations?\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Strategies for Safeguarding Medical Data Systems\", \"message\": \"Inquiring about methods to secure medical data with Alteryx Oracle Database 19c in hospital settings. Could you provide the necessary steps to ensure the confidentiality, integrity, and availability of sensitive medical information? I would greatly appreciate guidance on data encryption, access controls, and auditing mechanisms. Additionally, could you recommend specific configuration settings for Alteryx Oracle Database 19c to meet healthcare regulations?\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Arts & Entertainment/Music\": \"Arts & Entertainment/Music\", \"Online Communities/Social Networks\": \"Online Communities/Social Networks\", \"IT Support\": \"IT Support\", \"Hobbies & Leisure/Crafts\": \"Hobbies & Leisure/Crafts\", \"Travel & Transportation/Land Travel\": \"Travel & Transportation/Land Travel\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Change\": \"Change\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Boost Security of Medical Data\", \"message\": \"To the support team, I am inquiring about ways to improve encryption for the storage and transmission of medical data on hospital IT systems and related devices. Could you provide more details on the current security measures and suggest any enhancement options to protect sensitive patient information?\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Boost Security of Medical Data\", \"message\": \"To the support team, I am inquiring about ways to improve encryption for the storage and transmission of medical data on hospital IT systems and related devices. Could you provide more details on the current security measures and suggest any enhancement options to protect sensitive patient information?\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Finance/Investments\": \"Finance/Investments\", \"Arts & Entertainment/Music\": \"Arts & Entertainment/Music\", \"Sales and Pre-Sales\": \"Sales and Pre-Sales\", \"Home & Garden/Home Improvement\": \"Home & Garden/Home Improvement\", \"Customer Service\": \"Customer Service\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Request\": \"Request\", \"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Support for Security Incident Involving Medical Data\", \"message\": \"A security breach has been detected, which might have exposed medical data due to a software vulnerability in Express.js. We have applied patches and strengthened user authentication methods to prevent future breaches.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Support for Security Incident Involving Medical Data\", \"message\": \"A security breach has been detected, which might have exposed medical data due to a software vulnerability in Express.js. We have applied patches and strengthened user authentication methods to prevent future breaches.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Health/Mental Health\": \"Health/Mental Health\", \"Online Communities/Social Networks\": \"Online Communities/Social Networks\", \"Travel & Transportation/Land Travel\": \"Travel & Transportation/Land Travel\", \"Billing and Payments\": \"Billing and Payments\", \"IT Support\": \"IT Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI'm glad you managed to find your own fun and most probably made a big contribution for this great ending. The whole masked rogue idea was amazing"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I'm glad you managed to find your own fun and most probably made a big contribution for this great ending. The whole masked rogue idea was amazing\", \"questions\": {\"emotion_admiration\": {\"type\": \"noul\", \"instructions\": \"Does the author express admiration?\"}, \"emotion_relief\": {\"type\": \"noul\", \"instructions\": \"Does the author express relief?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nYou a childish ass bitch if u don't suck dick and u a gay ass nigga if u don't eat pussy 💯🤣 Idc let's argue"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"You a childish ass bitch if u don't suck dick and u a gay ass nigga if u don't eat pussy 💯🤣 Idc let's argue\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Ad Campaign Metrics Problem\", \"message\": \"I'm encountering difficulties with my ad campaign metrics not tracking correctly online, which I believe may stem from an incorrect API integration. After confirming my API keys and reviewing my campaign settings, the issue remains unresolved. I would be grateful for your support in addressing this.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Ad Campaign Metrics Problem\", \"message\": \"I'm encountering difficulties with my ad campaign metrics not tracking correctly online, which I believe may stem from an incorrect API integration. After confirming my API keys and reviewing my campaign settings, the issue remains unresolved. I would be grateful for your support in addressing this.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Beauty & Fitness/Fitness Training\": \"Beauty & Fitness/Fitness Training\", \"Food & Drink/Restaurants\": \"Food & Drink/Restaurants\", \"Books & Literature/Non-Fiction\": \"Books & Literature/Non-Fiction\", \"IT & Technology/Security Operations\": \"IT & Technology/Security Operations\", \"Technical Support\": \"Technical Support\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Request\": \"Request\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI received duplicate charges."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I received duplicate charges.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"wrong_amount_of_cash_received\": \"wrong amount of cash received\", \"get_physical_card\": \"get physical card\", \"terminate_account\": \"terminate account\", \"transaction_charged_twice\": \"transaction charged twice\", \"cash_withdrawal_not_recognised\": \"cash withdrawal not recognised\", \"exchange_via_app\": \"exchange via app\", \"pending_cash_withdrawal\": \"pending cash withdrawal\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhat did this mother think, that she was bringing her child to a birthday party. The U.S. has immigration laws and she, not anyone else broke immigration laws."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"What did this mother think, that she was bringing her child to a birthday party. The U.S. has immigration laws and she, not anyone else broke immigration laws.\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nI'm gonna put that SC butterfly filter around the head of my dick so y'all can see how stupid you look 🤗"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"I'm gonna put that SC butterfly filter around the head of my dick so y'all can see how stupid you look 🤗\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIf it makes you feel better im gonna be scrubbing out hard for a while so maybe ill see you round to give me another loss"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"If it makes you feel better im gonna be scrubbing out hard for a while so maybe ill see you round to give me another loss\", \"questions\": {\"emotion_sadness\": {\"type\": \"noul\", \"instructions\": \"Does the author express sadness?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nFor sure. I am infatuated with sharks, and much of that obsession comes from absolute fear haha."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"For sure. I am infatuated with sharks, and much of that obsession comes from absolute fear haha.\", \"questions\": {\"emotion_amusement\": {\"type\": \"noul\", \"instructions\": \"Does the author express amusement?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nOhio Gov. John Kasich up to 7th-place tie in new Iowa Poll; Ben Carson deadlocked with Donald Trump at the top"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Ohio Gov. John Kasich up to 7th-place tie in new Iowa Poll; Ben Carson deadlocked with Donald Trump at the top\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nMy money hasn't been deposited yet. When will it be?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"My money hasn't been deposited yet. When will it be?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"transfer_into_account\": \"transfer into account\", \"disposable_card_limits\": \"disposable card limits\", \"country_support\": \"country support\", \"pending_top_up\": \"pending top up\", \"beneficiary_not_allowed\": \"beneficiary not allowed\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nAnd I was cool with both guys 😂 shows you women are sneaky and ain't shit just like Niggas. Always knew my baby mama was a hoe 😭"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"And I was cool with both guys 😂 shows you women are sneaky and ain't shit just like Niggas. Always knew my baby mama was a hoe 😭\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user They won't be able to , all planes will be grounded, our ports will be shut down and in any case they will have to apply for visas, which Barnier will refuse , unless they are granted refugee status and that will take years."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user They won't be able to , all planes will be grounded, our ports will be shut down and in any case they will have to apply for visas, which Barnier will refuse , unless they are granted refugee status and that will take years.\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nMuslim refugee: In Greece they don't have enough money, therefore we want to go to Germany"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Muslim refugee: In Greece they don't have enough money, therefore we want to go to Germany\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user #NoDACA #TuesdayThoughtsObama Used Dreamer Illegal AliensTo Defraud The American Dream#NoDACAdeal #AmericaFirst Clint Eastwood was mocked forhis 2012 RNC speech, yet the new#ObamaPainting confirms ObamaIS nothing but an empty chair.CLINT"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user #NoDACA #TuesdayThoughtsObama Used Dreamer Illegal AliensTo Defraud The American Dream#NoDACAdeal #AmericaFirst Clint Eastwood was mocked forhis 2012 RNC speech, yet the new#ObamaPainting confirms ObamaIS nothing but an empty chair.CLINT\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n“Touch me again and I’ll kill you.” I’m low key blown away that The Last Boy Scout even *made* it to Australia."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"“Touch me again and I’ll kill you.” I’m low key blown away that The Last Boy Scout even *made* it to Australia.\", \"questions\": {\"emotion_anger\": {\"type\": \"noul\", \"instructions\": \"Does the author express anger?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nIm picking @user and @user is keeping the title till Royal Rumble on Sunday January 27 2013 #HelInACell"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Im picking @user and @user is keeping the title till Royal Rumble on Sunday January 27 2013 #HelInACell\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWhat are the disposable virtual card limits?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"What are the disposable virtual card limits?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"disposable_card_limits\": \"disposable card limits\", \"get_disposable_virtual_card\": \"get disposable virtual card\", \"age_limit\": \"age limit\", \"declined_card_payment\": \"declined card payment\", \"transfer_not_received_by_recipient\": \"transfer not received by recipient\", \"cash_withdrawal_charge\": \"cash withdrawal charge\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nHow do I find out my source of funds?"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"How do I find out my source of funds?\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"pending_cash_withdrawal\": \"pending cash withdrawal\", \"receiving_money\": \"receiving money\", \"card_linking\": \"card linking\", \"pin_blocked\": \"pin blocked\", \"top_up_by_card_charge\": \"top up by card charge\", \"cash_withdrawal_charge\": \"cash withdrawal charge\", \"verify_source_of_funds\": \"verify source of funds\", \"balance_not_updated_after_cheque_or_cash_deposit\": \"balance not updated after cheque or cash deposit\", \"unable_to_verify_identity\": \"unable to verify identity\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nExcept they call him. He doesn't go to their campuses and try to pick off the weak."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Except they call him. He doesn't go to their campuses and try to pick off the weak.\", \"questions\": {\"emotion_neutral\": {\"type\": \"noul\", \"instructions\": \"Does the author express neutral?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"\", \"message\": \"Hello customer support, I am encountering an issue with the analytics dashboard report for investment performance metrics. The displayed data seems to be incorrect, which is misleading for my decision-making process. I suspect this issue might have been caused by a recent software update. I have already tried restarting the application and clearing the cache, but the problem still persists. I would greatly appreciate it if you could look into this matter and provide a solution as soon as possible. Please let me know if there is any additional information you need to resolve this issue. Thank you for your time and assistance.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"\", \"message\": \"Hello customer support, I am encountering an issue with the analytics dashboard report for investment performance metrics. The displayed data seems to be incorrect, which is misleading for my decision-making process. I suspect this issue might have been caused by a recent software update. I have already tried restarting the application and clearing the cache, but the problem still persists. I would greatly appreciate it if you could look into this matter and provide a solution as soon as possible. Please let me know if there is any additional information you need to resolve this issue. Thank you for your time and assistance.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Real Estate\": \"Real Estate\", \"Shopping/E-commerce\": \"Shopping/E-commerce\", \"Health/Mental Health\": \"Health/Mental Health\", \"Technical Support\": \"Technical Support\", \"Travel & Transportation/Air Travel\": \"Travel & Transportation/Air Travel\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Request\": \"Request\", \"Problem\": \"Problem\", \"Incident\": \"Incident\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nLIBYA TELLS ITALY: \"We don't care how much you pay us, we don't want to take back the migrant wannabes that you reject because “they include terrorists, criminals, and human traffickers\" via @user"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"LIBYA TELLS ITALY: \\\"We don't care how much you pay us, we don't want to take back the migrant wannabes that you reject because “they include terrorists, criminals, and human traffickers\\\" via @user\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\n@user Joey, once again you have lost the plot. The tweet was about taxing the money migrants send home &amp; use it to encourage our own people to once again start having children. Sorry about the wages paid by Mcdonald's in Zagreb. If they can get"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"@user Joey, once again you have lost the plot. The tweet was about taxing the money migrants send home &amp; use it to encourage our own people to once again start having children. Sorry about the wages paid by Mcdonald's in Zagreb. If they can get\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nThat's the longest dimple I've ever seen>"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"That's the longest dimple I've ever seen>\", \"questions\": {\"emotion_admiration\": {\"type\": \"noul\", \"instructions\": \"Does the author express admiration?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"Security breach at healthcare provider due to obsolete antivirus program\", \"message\": \"Despite applying system scan updates, the problems remain unresolved.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"Security breach at healthcare provider due to obsolete antivirus program\", \"message\": \"Despite applying system scan updates, the problems remain unresolved.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Billing and Payments\": \"Billing and Payments\", \"General Inquiry\": \"General Inquiry\", \"Travel & Transportation/Land Travel\": \"Travel & Transportation/Land Travel\", \"IT Support\": \"IT Support\", \"News\": \"News\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Change\": \"Change\", \"Request\": \"Request\", \"Incident\": \"Incident\", \"Problem\": \"Problem\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nVery sad. People like this are ruining Aus."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Very sad. People like this are ruining Aus.\", \"questions\": {\"emotion_sadness\": {\"type\": \"noul\", \"instructions\": \"Does the author express sadness?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation (JSON state):\n{\"subject\": \"\", \"message\": \"Dear Customer Support, I am reporting an issue with the SaaS platform while writing a report. There seems to be an error when saving project data, which might be related to integration issues with external tools. Despite refreshing the application and clearing the cache, the problem still persists. I would greatly appreciate it if you could look into the matter and provide a solution as soon as possible. Please let me know if there is any additional information needed to resolve the issue. Thank you for your time and assistance. I look forward to hearing from you.\"}"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"support ticket triage\", \"request\": {\"model\": \"clef-omni\", \"state\": {\"subject\": \"\", \"message\": \"Dear Customer Support, I am reporting an issue with the SaaS platform while writing a report. There seems to be an error when saving project data, which might be related to integration issues with external tools. Despite refreshing the application and clearing the cache, the problem still persists. I would greatly appreciate it if you could look into the matter and provide a solution as soon as possible. Please let me know if there is any additional information needed to resolve the issue. Thank you for your time and assistance. I look forward to hearing from you.\"}, \"questions\": {\"route_to\": {\"type\": \"choice\", \"instructions\": \"Which support queue should handle this ticket?\", \"criteria\": {\"Human Resources\": \"Human Resources\", \"Product Support\": \"Product Support\", \"Pets & Animals/Pet Services\": \"Pets & Animals/Pet Services\", \"Service Outages and Maintenance\": \"Service Outages and Maintenance\", \"Beauty & Fitness/Fitness Training\": \"Beauty & Fitness/Fitness Training\"}}, \"ticket_type\": {\"type\": \"choice\", \"instructions\": \"What kind of ticket is this?\", \"criteria\": {\"Incident\": \"Incident\", \"Problem\": \"Problem\", \"Request\": \"Request\", \"Change\": \"Change\"}}, \"priority\": {\"type\": \"score\", \"instructions\": \"How urgent is this ticket?\", \"criteria\": [\"Low, can wait days\", \"Medium, handle this week\", \"High, handle today\", \"Critical, service blocked\"]}, \"needs_escalation\": {\"type\": \"noul\", \"instructions\": \"Should this be escalated beyond first-line support?\"}}}}"}], "source": "tickets"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nProblem verifying my account"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Problem verifying my account\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"unable_to_verify_identity\": \"unable to verify identity\", \"top_up_reverted\": \"top up reverted\", \"apple_pay_or_google_pay\": \"apple pay or google pay\", \"passcode_forgotten\": \"passcode forgotten\", \"automatic_top_up\": \"automatic top up\", \"top_up_by_card_charge\": \"top up by card charge\"}}}}}"}], "source": "banking77"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nField trip to the West End!RT @user Imelda Staunton May Star in London Revival of Gypsy"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Field trip to the West End!RT @user Imelda Staunton May Star in London Revival of Gypsy\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nCongratulations on scoring your first goal for Swansea yesterday. May I say that you look exactly like Jonas Gutierrez ! @user"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"sentiment rating\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Congratulations on scoring your first goal for Swansea yesterday. May I say that you look exactly like Jonas Gutierrez ! @user\", \"questions\": {\"sentiment\": {\"type\": \"score\", \"instructions\": \"How does the author feel in this message?\", \"criteria\": [\"Negative tone\", \"Neutral, no strong tone\", \"Positive tone\"]}, \"is_complaint\": {\"type\": \"noul\", \"instructions\": \"Is the author complaining about a product or service?\"}}}}"}], "source": "sentiment"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nWatch: Muslim refugee in Denmark say they have a right to target Jews - Free Speech Time - @user"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"hate speech screen\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Watch: Muslim refugee in Denmark say they have a right to target Jews - Free Speech Time - @user\", \"questions\": {\"is_hate_speech\": {\"type\": \"noul\", \"instructions\": \"Does this message contain hate speech against a group?\", \"criteria\": {\"true\": \"Targets a protected group with hate.\", \"false\": \"No hate toward a protected group.\"}}}}}"}], "source": "hate"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nYou love your joke!"}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"emotion tagging\", \"request\": {\"model\": \"clef-omni\", \"state\": \"You love your joke!\", \"questions\": {\"emotion_love\": {\"type\": \"noul\", \"instructions\": \"Does the author express love?\"}}}}"}], "source": "go_emotions"}
{"prompt": [{"role": "system", "content": "You are a Clef-Omni orchestrator. For every user situation you decide how to interrogate Clef-Omni, a decision model served at POST /v1/systemone. Clef-Omni judges a state against a schema of typed questions in a single forward pass and returns a probability for every allowed option of every question.\n\nQuestion types:\n- \"noul\": yes/no. Returns the probability of true. Use when one property, true or false, decides something. Optional criteria: {\"true\": \"...\", \"false\": \"...\"}.\n- \"choice\": exactly one of several labels fits. criteria is an object mapping option id to a short, concrete description. Use 2 to 10 options; descriptions matter more than ids.\n- \"score\": ordered levels from lowest to highest. criteria is a list of 2 to 6 level descriptions. Prefer score over choice when the levels have an order.\n\nRules:\n- 1 to 8 questions per request. All questions are judged against the same state.\n- Several labels can hold at once -> ask one noul question per label in the same request.\n- Question ids: lowercase snake_case, letters, digits, '_', '-', '.', max 100 chars.\n- The state may be a raw string or a JSON object. Pass through the user's situation, trimmed of noise.\n- When several labels can hold at once, or one property needs a yes/no, an ordered scale, and a routing label together, combine the question types in one request.\n- Respond with ONLY a JSON object, no markdown fences:\n{\"use_case\": \"<2-6 words naming the decision>\", \"request\": {\"model\": \"clef-omni\", \"state\": <state>, \"questions\": {<question id>: {\"type\": \"noul\"|\"choice\"|\"score\", \"instructions\": \"...\", \"criteria\": ...}, ...}}}"}, {"role": "user", "content": "Situation:\nReasons for top up cancellation."}], "completion": [{"role": "assistant", "content": "{\"use_case\": \"banking intent routing\", \"request\": {\"model\": \"clef-omni\", \"state\": \"Reasons for top up cancellation.\", \"questions\": {\"intent\": {\"type\": \"choice\", \"instructions\": \"What does the customer want? Which banking intent fits this message?\", \"criteria\": {\"disposable_card_limits\": \"disposable card limits\", \"lost_or_stolen_card\": \"lost or stolen card\", \"card_not_working\": \"card not working\", \"reverted_card_payment?\": \"reverted card payment?\", \"passcode_forgotten\": \"passcode forgotten\", \"exchange_charge\": \"exchange charge\", \"get_physical_card\": \"get physical card\", \"verify_top_up\": \"verify top up\", \"top_up_reverted\": \"top up reverted\"}}}}}"}], "source": "banking77"}