Type Alias AdaptiveTokenEstimator

estimator

corrections

• corrections(): AdaptiveModelCorrection[]

  Snapshot of per-model calibration state, in insertion order.

  Returns AdaptiveModelCorrection[]

observe

• observe(request: LLMRequest, usage: TokenUsage): void

  Record one completed call: the request as estimated, and the provider-reported actual usage. Wire this to the bulkhead's release event: on("release", (e) => { if (e.usage) observe(e.request, e.usage) }). Observations with a zero base input estimate are ignored (no ratio can be formed). Throws if usage fields are not non-negative integers.

  Parameters

  • request: LLMRequest
  • usage: TokenUsage

  Returns void

reset

• reset(model?: string): void

  Clear calibration for one model, or for all when omitted.

  Parameters

  • Optionalmodel: string

  Returns void