> For AI agents: the complete documentation index is available at https://open-spaced-repetition.github.io/ts-fsrs/llms.txt, the full documentation bundle is available at https://open-spaced-repetition.github.io/ts-fsrs/llms-full.txt.

# Lifecycle

Handlers run synchronously around the scheduler core in the order the middleware was registered. `.use(a, b)` and `.use(a).use(b)` register the same order, and each `use` returns a new definition without changing the original. The scheduler core sits in the middle of the chain, so code before `next()` prepares the operation and code after it sees the result.

## Order around the core

For `.use(a, b)`, the call order is:

```text
a: before next()
  b: before next()
    scheduler core
  b: after next()
a: after next()
```

Call `next()` once to continue the normal scheduling flow. Set inputs such as `ctx.candidate.desiredRetention[ctx.input.grade]` before it; inspect or change the computed result after it. In review, `ctx.scheduledDays` is the interval in days. The scheduler applies the configured chrono after the handlers finish to calculate the due time, then validates the result.

Order matters when policies interact: register `schedulerScheduledDaysMiddleware` before the interval policies so it records their final result on unwind; keep learning steps before the maximum- and monotonic-interval policies so explicit steps retain their delay. Register leech before handlers that set `scheduleStatus` after `next()`, so suspension is applied last. The [Built-in policies](https://open-spaced-repetition.github.io/ts-fsrs/guide/middleware/built-in.md) page lists each policy and how `DefaultScheduler` orders them.

## Review and rollback

Rollback uses its own handlers in the same registration order. A rollback handler can read `ctx.config`, `ctx.input.card`, and `ctx.input.revlog`, and writes only `ctx.result.card`. Schema declarations alone do not restore custom fields: save the required old values during review and restore them in the rollback handler according to the configured policy, as the streak example on the [Overview](https://open-spaced-repetition.github.io/ts-fsrs/guide/middleware/index.md) does.

`preview()` runs the review handlers for the requested candidate ratings. Keep handlers free of database writes, notifications, and other external effects; persist the selected review result in your application. `forget()` resets a card with its defaults rather than running a separate handler, so the streak example resets the counter to zero.