ReasoningSpec
Field |
Type |
Repeated |
Description |
state_json |
|||
prompt |
|||
explanation |
|||
events |
✅ |
||
local_analysis |
✅ |
||
include_frame |
|||
gate |
Optional gate on the evaluation LLM call itself (task 74 G2): a call is only made while the gate holds; analysisIntervalSecs rate-limits on top |
||
frame_policy |
|||
subtitle_window_secs |
Time window of recent subtitle/caption cues carried into the prompt and aggregated analysis (the node can be subscribed to any subtitle stream, e.g. a captioning node’s output). 0 = default (30s). |
||
include_audio_secs |
When > 0, attach the last N seconds of programme audio (16kHz mono WAV) to each generative request. Requires an audio subscription; supported by Gemini and OpenAI audio-capable models — not Claude. |
||
state_schema |
✅ |
Typed schema of the state record, one entry per state field — designed by the Plan phase (or supplied by hand). Drives strict, schema-enforced _update_state tool calls on providers that support them; when absent, field types are inferred from the current state values. |
|
requirements |
✅ |
Item 13: artifact requirements declared by a flow-designing planner — things it cannot invent (face/voice galleries, customer model files). The user attaches them before evaluation, or starts empty and uses the runtime gallery CRUD. Informational at evaluation time. |