ReasoningEvaluateSettings interface

Settings for a ReasoningEvaluateNode. The spec carries the prompt and state from the plan phase - evaluate just needs the spec, a provider, and video configuration.

Signature:

export interface ReasoningEvaluateSettings extends ProcessorNodeSettings<ReasoningEvaluateNode>

Extends: ProcessorNodeSettings<ReasoningEvaluateNode>

Properties

Property	Type	Description
analysisIntervalSecs?	number	(Optional)
contextDocuments?	string[]	(Optional) Reference documents (markdown) providing domain context for the evaluation LLM. These are included in the system instruction and cached across frames.
onAnalysis?	(analysis: AggregatedAnalysis) ⇒ void	(Optional)
onLLMResponse?	(response: LLMResponse) ⇒ void	(Optional)
onRetry?	(status: RetryStatus) ⇒ void	(Optional)
provider?	ReasoningProvider	(Optional)
spec	ReasoningSpec
stateFailure?	{ escalateAfter?: number; suppressEvents?: boolean; }	(Optional) Policy for _update_state failures. escalateAfter: raise onError once this many consecutive state updates have failed (default 3). suppressEvents: drop user events from a response whose state update failed — strict mode for pipelines where events trigger real actions (default false). Failures never poison the state: the previous value is kept and the model is corrected on the next request.
videoSettings	ReasoningVideoSettings

Property

Type

Description

analysisIntervalSecs?

number

(Optional)

contextDocuments?

string[]

(Optional) Reference documents (markdown) providing domain context for the evaluation LLM. These are included in the system instruction and cached across frames.

onAnalysis?

(analysis: AggregatedAnalysis) ⇒ void

(Optional)

onLLMResponse?

(response: LLMResponse) ⇒ void

(Optional)

onRetry?

(status: RetryStatus) ⇒ void

(Optional)

(Optional)

{ escalateAfter?: number; suppressEvents?: boolean; }

(Optional) Policy for _update_state failures. escalateAfter: raise onError once this many consecutive state updates have failed (default 3). suppressEvents: drop user events from a response whose state update failed — strict mode for pipelines where events trigger real actions (default false). Failures never poison the state: the previous value is kept and the model is corrected on the next request.

videoSettings

ReasoningVideoSettings