Every world has a single clock that starts at t0. Time-consuming processes (a run settling, a transcription completing) have named durations, and events fire when “now” crosses their due time.

​
Two modes

Wall-clock ticking

“Now” advances with real time, like a staging environment — events fire on their own as time passes. The right default for a long-lived raw sandbox. (Ships first.)

Freeze-in-place

“Now” moves only when you advance it. Every time-derived field is reproducible by construction, and due timers drain inside your advance() call. The eventual default for CI. (Roadmap.)

​
Advancing time

await sandbox.clock.advance("5s");
// → { now: "2025-01-01T00:00:05Z", fired: ["task_run.completed"] }
When you advance time explicitly, every due event fires in deterministic due-time order, each to completion, before the call returns. A POST finishes its own state change first, then its cascade fires through the ordered queue. There’s no “halfway” state you can accidentally observe.

​
Making CI deterministic today: collapse the durations

Until freeze-in-place ships, you get deterministic sequencing by setting named durations to 0. The timer queue then drains in order on the triggering call — no waiting on real time, no flake: 
await sandbox.clock.setDurations({ run_settle: "0s", transcription: "0s" });

const run = await deck.taskRuns.create({ taskId: "task_rent" });
await sandbox.clock.advance("0s");   // drains the cascade inline
// the run is already `completed` and its webhook already fired
Durations→0 gives you deterministic ordering and “did the cascade fire?” assertions today. Absolute timestamps still track wall-clock until freeze lands — so if you need to assert an exact created_at, that’s what freeze mode (and a pinned t0) are for. See Deterministic CI.

​
Reset rewinds to t0

reset() empties state and rewinds the clock to t0. Because t0 is a fixed constant by default, every fresh world is byte-identical. Override it per world to test date boundaries: 
await sandbox.reset({ t0: "2025-12-31T23:59:00Z" }); // test the year rollover