| { |
| "$schema": "https://json-schema.org/draft/2020-12/schema", |
| "$id": "https://paintedwolf.ai/schemas/pw-decide-row/1", |
| "title": "pw-decide-row/1", |
| "description": "One turn decision as a training row, written by `lycaon-debug decide export`: the state the engine read, the candidates the turn offered, what the engine answered, and what the session then did. Producers may add `meta` and judged scores; consumers read option sets from `offered` and option texts from the corpus.", |
| "type": "object", |
| "required": [ |
| "schema", |
| "receipt", |
| "session", |
| "root_session", |
| "opening_message_id", |
| "host", |
| "surface", |
| "catalog_revision", |
| "project", |
| "model", |
| "partial", |
| "lang", |
| "engine", |
| "offered", |
| "state", |
| "labels" |
| ], |
| "properties": { |
| "schema": { |
| "const": "pw-decide-row/1" |
| }, |
| "receipt": { |
| "type": "integer", |
| "description": "The turn-load receipt the row was read from." |
| }, |
| "session": { |
| "type": "string" |
| }, |
| "root_session": { |
| "type": "string", |
| "description": "The session itself, or the coordinator session a worker leg belongs to." |
| }, |
| "opening_message_id": { |
| "type": "string" |
| }, |
| "host": { |
| "enum": [ |
| "coordinator", |
| "worker" |
| ] |
| }, |
| "surface": { |
| "type": "string", |
| "description": "The coordinator surface, or the worker's tool profile." |
| }, |
| "catalog_revision": { |
| "type": "string" |
| }, |
| "project": { |
| "type": "string" |
| }, |
| "model": { |
| "type": "string", |
| "description": "The model that made most of the session's calls." |
| }, |
| "partial": { |
| "type": "boolean", |
| "description": "The turn did not complete; only its request_tools needs are labels." |
| }, |
| "lang": { |
| "enum": [ |
| "en", |
| "other" |
| ] |
| }, |
| "engine": { |
| "type": "object", |
| "required": [ |
| "state", |
| "reason", |
| "preloaded", |
| "omitted" |
| ], |
| "properties": { |
| "state": { |
| "enum": [ |
| "answered", |
| "abstained" |
| ], |
| "description": "An answered turn carried what the engine chose, so its tool labels are not independent of the engine." |
| }, |
| "reason": { |
| "type": "string" |
| }, |
| "preloaded": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| }, |
| "omitted": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| } |
| } |
| }, |
| "offered": { |
| "type": "object", |
| "required": [ |
| "floor", |
| "loadable", |
| "guides" |
| ], |
| "properties": { |
| "floor": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| }, |
| "loadable": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| }, |
| "guides": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| } |
| } |
| }, |
| "state": { |
| "type": "object", |
| "required": [ |
| "host", |
| "user" |
| ], |
| "description": "The turnload.State the engine read, verbatim." |
| }, |
| "labels": { |
| "type": "object", |
| "required": [ |
| "tools", |
| "requests", |
| "requested_names", |
| "requested_groups", |
| "skills", |
| "kind", |
| "guides" |
| ], |
| "properties": { |
| "tools": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| }, |
| "description": "Offered loadable tools the turn called." |
| }, |
| "requests": { |
| "type": "array", |
| "items": { |
| "type": "object", |
| "required": [ |
| "need", |
| "exact", |
| "after" |
| ], |
| "properties": { |
| "need": { |
| "type": "string" |
| }, |
| "exact": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| }, |
| "after": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| }, |
| "scores": { |
| "type": "object", |
| "additionalProperties": { |
| "type": "integer", |
| "minimum": 0, |
| "maximum": 4 |
| }, |
| "description": "Judged relevance of each ranked tool card to the need." |
| } |
| } |
| } |
| }, |
| "requested_names": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| }, |
| "requested_groups": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| }, |
| "skills": { |
| "type": "array", |
| "items": { |
| "type": "string" |
| } |
| }, |
| "kind": { |
| "enum": [ |
| "answer_only", |
| "inspect", |
| "change", |
| "run", |
| "delegate", |
| null |
| ], |
| "description": "The kind the turn's tool calls imply; null for a turn that did not complete." |
| }, |
| "guides": { |
| "type": "object", |
| "additionalProperties": { |
| "type": [ |
| "boolean", |
| "null" |
| ] |
| }, |
| "description": "Offered instruction units: needed, unneeded, or unknown (null)." |
| }, |
| "tool_scores": { |
| "type": "object", |
| "additionalProperties": { |
| "type": "integer", |
| "minimum": 0, |
| "maximum": 4 |
| }, |
| "description": "Judged relevance of each offered loadable tool card to the request." |
| }, |
| "skill_scores": { |
| "type": "object", |
| "additionalProperties": { |
| "type": "integer", |
| "minimum": 0, |
| "maximum": 4 |
| }, |
| "description": "Judged relevance of each skill card to the request." |
| }, |
| "second_scores": { |
| "type": "object", |
| "description": "A second judge's scores on the same cards: its model, skills and tools maps, and one map per request_tools need (null where it scored none).", |
| "properties": { |
| "model": {"type": "string"}, |
| "skills": {"type": "object", "additionalProperties": {"type": "integer", "minimum": 0, "maximum": 4}}, |
| "tools": {"type": "object", "additionalProperties": {"type": "integer", "minimum": 0, "maximum": 4}}, |
| "requests": {"type": "array", "items": {"type": ["object", "null"], "additionalProperties": {"type": "integer", "minimum": 0, "maximum": 4}}} |
| }, |
| "required": ["model"] |
| } |
| } |
| }, |
| "judge": { |
| "type": "object", |
| "description": "Which model judged the row's scores, and how." |
| }, |
| "meta": { |
| "type": "object", |
| "description": "Producer metadata, such as the task, repository, and prompt group; consumers may group by it but never train on it." |
| } |
| } |
| } |
|
|
