AuthoritativeAgent

An agent whose model decisions are taken in one place.

class AuthoritativeAgent

import { AuthoritativeAgent } from '@driftengine/ai';

In depth

Wraps the two sessions the package already has and picks between them. It adds no policy, no provider handling and no replay logic; all three exist and are reused.

Constructor

`new`

constructor(session: AgentSession<unknown>, replay: ReplaySession, options: AuthoritativeAgentOptions)

Parameter	Type	Description
`session`	`AgentSession<unknown>`
`replay`	`ReplaySession`
`options`	`AuthoritativeAgentOptions`

Accessors

Name	Type	Description
`publishedDecisions`get	`number`	Decisions this peer put on the wire. Zero on a participant, always.
`acceptedDecisions`get	`number`	Decisions taken off the wire. Zero on the authority, always.
`earliestLateTick`get	`number`	The oldest tick a decision arrived for after that tick had already run, or -1. More A consumer holding a `RewindLoop` reads this, rewinds to it and replays. Cleared by `clearLate` once they have.

Methods

`clearLate`

clearLate(): void

`tick`

tick(tickNumber: number, nowMs: number): Intent

Advance one tick and return what the agent is doing.

Parameter	Type	Description
`tickNumber`	`number`
`nowMs`	`number`

Arriving decisions are drained before the tick runs, so a command for this tick is in the log by the time the replay reads it. A command for an earlier tick is recorded anyway — the log is keyed by acceptance tick, so it lands where a later rewind will find it — and noted in earliestLateTick.