API Exports¶

Everything exported from @peleke.s/interoception:

import {
  // Sensor
  createPreExecSensor,
  DEFAULT_METRICS,

  // Metric functions
  computeGoalDrift,
  goalDriftMetric,
  computeMemoryRetention,
  memoryRetentionMetric,
  computeContradictionPressure,
  contradictionPressureMetric,
  computeSemanticDiffusion,
  semanticDiffusionMetric,

  // Coherence index
  computeCoherenceIndex,
  classifyBand,

  // Constants
  DEFAULT_WEIGHTS,
  DEFAULT_THRESHOLDS,

  // Vector utilities
  cosineSimilarity,
  normalize,
  dot,
  magnitude,
  mean,
  clamp01,
} from "@peleke.s/interoception";

Functions¶

Function	Module	Description
`createPreExecSensor(options)`	sensor	Create a coherence sensor
`computeGoalDrift(goalEmbeddings, contextEmbeddings)`	metrics	Compute goal drift score
`computeMemoryRetention(goalEmbeddings, memoryEmbeddings)`	metrics	Compute memory retention score
`computeContradictionPressure(contextEmbeddings, threshold?)`	metrics	Compute contradiction pressure
`computeSemanticDiffusion(contextEmbeddings)`	metrics	Compute semantic diffusion
`computeCoherenceIndex(metrics, weights?, invertedMetrics?)`	metrics	Compute weighted coherence index
`classifyBand(coherenceIndex, thresholds?)`	metrics	Classify index into a band
`cosineSimilarity(a, b)`	util	Cosine similarity between vectors
`dot(a, b)`	util	Dot product of two vectors
`magnitude(v)`	util	L2 norm of a vector
`normalize(v)`	util	L2-normalize a vector
`mean(values)`	util	Arithmetic mean
`clamp01(v)`	util	Clamp value to [0, 1]

Constants¶

Constant	Type	Description
`DEFAULT_METRICS`	`MetricFn[]`	The four core metric functions
`DEFAULT_WEIGHTS`	`MetricWeights`	Equal weights (0.25 each)
`DEFAULT_THRESHOLDS`	`BandThresholds`	`{ green: 0.8, yellow: 0.6, orange: 0.4 }`

Metric Instances¶

Instance	Type	Inverted	Description
`goalDriftMetric`	`MetricFn`	Yes	Goal drift as a pluggable metric
`memoryRetentionMetric`	`MetricFn`	No	Memory retention as a pluggable metric
`contradictionPressureMetric`	`MetricFn`	Yes	Contradiction pressure as a pluggable metric
`semanticDiffusionMetric`	`MetricFn`	Yes	Semantic diffusion as a pluggable metric

Types¶

Type	Module	Description
`Embedder`	types	Pluggable embedding interface
`StateProvider`	types	Pluggable agent state access
`MetricFn`	types	Embedding-based metric strategy
`ScalarMetricFn`	types	Scalar metric strategy (no embeddings)
`MetricInput`	types	Input data for metric functions
`MetricSnapshot`	types	All metric values
`MetricWeights`	types	Weights for coherence index
`BandThresholds`	types	Band classification thresholds
`Band`	types	`"green" \\| "yellow" \\| "orange" \\| "red"`
`CoherenceReading`	types	Single coherence measurement
`PreExecSensor`	sensor	Sensor interface
`PreExecSensorOptions`	sensor	Sensor configuration
`Tick`	types (re-export)	Clock tick from `@peleke.s/cadence`