Download examples/behavioral-intervention.json from Gautam-R-Patil/sova-format: direct link, hf CLI and curl.
- Browser
- Download file 1.62 kB
-
https://huggingface.co/datasets/Gautam-R-Patil/sova-format/resolve/main/examples/behavioral-intervention.json
- Command line
-
hf download hf://datasets/Gautam-R-Patil/sova-format/examples/behavioral-intervention.json
-
curl -L -o behavioral-intervention.json https://huggingface.co/datasets/Gautam-R-Patil/sova-format/resolve/main/examples/behavioral-intervention.json
1.62 kB
| { | |
| "artifactType": "sova.scenario", | |
| "schemaVersion": "0.1.0", | |
| "id": "sova:scenario:019fae00-0000-7000-8000-000000000004", | |
| "version": "0.1.0", | |
| "title": "Observable context intervention", | |
| "purpose": "Compare outputs before and after removing one declared context item.", | |
| "parameters": {"contextItem": "The synthetic preference is concise answers."}, | |
| "preconditions": [{"kind": "model", "instrumentation": "observable-io-only"}], | |
| "sequences": [], | |
| "procedure": { | |
| "steps": [ | |
| { | |
| "id": "baseline", | |
| "action": "model.prompt-with-context", | |
| "inputs": {"prompt": "Describe the fixture.", "contextFromParameter": "contextItem"}, | |
| "onFailure": "inconclusive", | |
| "requires": ["model.prompt/0.1"] | |
| }, | |
| { | |
| "id": "intervention", | |
| "action": "model.prompt-with-context", | |
| "inputs": {"prompt": "Describe the fixture.", "omitContext": true}, | |
| "onFailure": "inconclusive", | |
| "requires": ["model.prompt/0.1"] | |
| } | |
| ] | |
| }, | |
| "triggers": [], | |
| "mutations": [{"factor": "contextItem", "operation": "omit"}], | |
| "expectedEffects": [{"kind": "response-difference", "dimension": "length"}], | |
| "oracles": [{"kind": "record-comparison", "deterministic": false}], | |
| "evidenceRequirements": ["prompt.sent", "model.response", "evaluation.completed"], | |
| "safety": { | |
| "budgets": {"maxSteps": 2}, | |
| "forbiddenEffects": ["tool.call", "network.egress"], | |
| "stopConditions": [] | |
| }, | |
| "cleanup": [], | |
| "limitations": [ | |
| "This observes input/output behavior and does not claim access to hidden model reasoning." | |
| ], | |
| "extensions": {} | |
| } | |