Skip to main content

Class: EventDirector

Defined in: apps/paracosm/src/runtime/director.ts:246

Constructors​

Constructor​

new EventDirector(): EventDirector

Returns​

EventDirector

Methods​

generateEventBatch()​

generateEventBatch(ctx, maxEvents?, provider?, model?, instructions?, onUsage?, onProviderError?, onSchemaAttempt?): Promise<DirectorEventBatch>

Defined in: apps/paracosm/src/runtime/director.ts:258

Generate 1 to maxEvents events for a turn via the schema-validated generateObject wrapper. Falls back to canned events when the LLM call or validation fails after maxRetries.

Parameters​

ctx​

DirectorContext

maxEvents?​

number = 3

provider?​

LlmProvider = 'openai'

model?​

string = 'gpt-5.4'

instructions?​

string

onUsage?​

(result) => void

Optional callback invoked with the director's LLM response so the orchestrator can account for director spend in its cost telemetry. Director runs once per turn on the flagship model; without this hook the call was invisible to the per-run cost object returned by runSimulation().

onProviderError?​

(err) => void

Called with the raw caught error when the director's LLM call throws. Lets the orchestrator classify the error (quota / auth / etc.) and emit a provider_error SSE event before the existing fallback path quietly substitutes canned events.

onSchemaAttempt?​

(attempts, fellBack) => void

Called with the attempts count + fallback flag so the orchestrator can feed per-schema retry telemetry into its cost tracker.

Returns​

Promise<DirectorEventBatch>

getMilestoneCrisis()​

getMilestoneCrisis(turn, maxTurns): DirectorCrisis | null

Defined in: apps/paracosm/src/runtime/director.ts:333

Get a milestone crisis (Turn 1 or final turn). These are fixed for narrative anchoring.

Parameters​

turn​

number

maxTurns​

number

Returns​

DirectorCrisis | null