@acentric/replay

Runs an agent's pure half with no runner. It is what runners and the CLI use to replay and fork sessions; for tests, use @acentric/testing, which re-exports it.

• simulate(agent, config) creates a session in memory and delivers messages to step one at a time, matching effects as a runner does. It runs no handlers: the caller reports their outcomes with ok, fail and interrupt, and plays the clock with advance. upgrade(next) moves the session to another version through its migrate.
• replay(journal, agents) delivers a session's journal again. Faithful replay, the default, runs each entry on the version that ran it and checks that it gives the answer and digest the journal recorded. What-if replay, { mode: 'what-if' }, runs every entry on one agent from init, to see what that code would have done. Either way the result says which entries differ.
• journalDigest(snapshot, config, appended) computes a journal entry's digest: SHA-256 of the RFC 8785 canonical JSON of the three, as every runner computes it.

import { forReplay, replay } from '@acentric/replay';

const result = replay(journal, [forReplay(v1), forReplay(v2)]);
if (result.differs !== null) console.log(`entry ${result.differs} differs`);

forReplay erases an agent's types, so versions with different snapshots replay together. The journal format and digest are in the runner specification; the design is in docs/reference/replay.md.