| { |
| "title": "Dissei Financial Judgment", |
| "evaluated_question_count": 7, |
| "case_count": 1, |
| "full_environment_task_count": 70, |
| "protocol": { |
| "harbor": "0.23.0", |
| "agent": "Terminus-2 JSON", |
| "agent_turns": 40, |
| "task_timeout_seconds": 1800, |
| "max_output_tokens_per_request": 16384, |
| "agent_reasoning": "high", |
| "judge_model": "gemini-3.1-pro-preview", |
| "judge_reasoning": "low", |
| "judge_votes": 1, |
| "selected_outcomes_per_task": 1, |
| "infrastructure_recoveries_disclosed": true, |
| "dataroom_calls": 20 |
| }, |
| "results": [ |
| { |
| "model": "Claude Fable 5.1", |
| "model_id": "claude-fable-5-1", |
| "provider": "Anthropic", |
| "reward_per_100": 60.73, |
| "selected_outcomes": 7, |
| "submitted": 7, |
| "turn_limit_hits": 0, |
| "parser_errors": 0, |
| "unresolved_errors": 0, |
| "recovered_tasks": 0, |
| "per_family_reward_per_100": { |
| "diagnostic": 78.82, |
| "predictive": 70.94, |
| "explanatory": 45.7, |
| "quantitative": 43.33, |
| "counterfactual": 66.88, |
| "comparative": 74.55, |
| "strategic": 44.9 |
| }, |
| "status": "7 submitted; no infrastructure recovery" |
| }, |
| { |
| "model": "GPT-6 Astra", |
| "model_id": "gpt-6-astra", |
| "provider": "OpenAI", |
| "reward_per_100": 52.45, |
| "selected_outcomes": 7, |
| "submitted": 7, |
| "turn_limit_hits": 0, |
| "parser_errors": 0, |
| "unresolved_errors": 0, |
| "recovered_tasks": 0, |
| "per_family_reward_per_100": { |
| "diagnostic": 73.38, |
| "predictive": 73.38, |
| "explanatory": 22.85, |
| "quantitative": 40.0, |
| "counterfactual": 53.2, |
| "comparative": 74.55, |
| "strategic": 29.82 |
| }, |
| "status": "7 submitted; no infrastructure recovery" |
| }, |
| { |
| "model": "Claude Opus 5.5", |
| "model_id": "claude-opus-5-5", |
| "provider": "Anthropic", |
| "reward_per_100": 52.17, |
| "selected_outcomes": 7, |
| "submitted": 7, |
| "turn_limit_hits": 0, |
| "parser_errors": 0, |
| "unresolved_errors": 0, |
| "recovered_tasks": 0, |
| "per_family_reward_per_100": { |
| "diagnostic": 70.94, |
| "predictive": 70.94, |
| "explanatory": 22.85, |
| "quantitative": 43.33, |
| "counterfactual": 59.64, |
| "comparative": 74.55, |
| "strategic": 22.94 |
| }, |
| "status": "7 submitted; no infrastructure recovery" |
| }, |
| { |
| "model": "GLM-5.3", |
| "model_id": "glm-5.3", |
| "provider": "Z.ai", |
| "reward_per_100": 44.91, |
| "selected_outcomes": 7, |
| "submitted": 7, |
| "turn_limit_hits": 0, |
| "parser_errors": 1, |
| "unresolved_errors": 0, |
| "recovered_tasks": 1, |
| "per_family_reward_per_100": { |
| "diagnostic": 48.92, |
| "predictive": 57.08, |
| "explanatory": 22.85, |
| "quantitative": 50.0, |
| "counterfactual": 60.8, |
| "comparative": 52.18, |
| "strategic": 22.51 |
| }, |
| "status": "7 submitted; 1 task recovered; earlier recovery interrupted by the VM limit" |
| }, |
| { |
| "model": "Kimi K3", |
| "model_id": "kimi-k3", |
| "provider": "Moonshot AI", |
| "reward_per_100": 35.79, |
| "selected_outcomes": 7, |
| "submitted": 7, |
| "turn_limit_hits": 0, |
| "parser_errors": 0, |
| "unresolved_errors": 0, |
| "recovered_tasks": 0, |
| "per_family_reward_per_100": { |
| "diagnostic": 0.0, |
| "predictive": 55.18, |
| "explanatory": 23.4, |
| "quantitative": 34.67, |
| "counterfactual": 52.5, |
| "comparative": 61.71, |
| "strategic": 23.07 |
| }, |
| "status": "7 submitted; no infrastructure recovery" |
| }, |
| { |
| "model": "Gemini 3.8 Flash", |
| "model_id": "gemini-3.8-flash", |
| "provider": "Google", |
| "reward_per_100": 9.01, |
| "selected_outcomes": 7, |
| "submitted": 1, |
| "turn_limit_hits": 6, |
| "parser_errors": 50, |
| "unresolved_errors": 0, |
| "recovered_tasks": 5, |
| "per_family_reward_per_100": { |
| "diagnostic": 63.06, |
| "predictive": 0.0, |
| "explanatory": 0.0, |
| "quantitative": 0.0, |
| "counterfactual": 0.0, |
| "comparative": 0.0, |
| "strategic": 0.0 |
| }, |
| "status": "Protocol-limited: 1 submitted, 6 turn limits; 5 tasks recovered after infrastructure failures" |
| } |
| ], |
| "limitations": [ |
| "Single case and seven questions", |
| "No repeated-run variance estimate", |
| "Judge and harness dependence", |
| "Provider framing can differ", |
| "Independent reproduction requires approved access", |
| "Not an external certification" |
| ], |
| "harbor_dataset_url": "https://hub.harborframework.com/datasets/dissei/financial-judgment", |
| "scope": "Public task previews and recorded aggregate results only. No runnable benchmark, grading materials or private run records." |
| } |
|
|
