> For the complete documentation index, see [llms.txt](https://series-1.gitbook.io/rundot-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://series-1.gitbook.io/rundot-docs/v5.28.0/readme/syncplay/syncplay-module-certification.md).

# Syncplay: Module Certification (PROVISIONAL)

> **PROVISIONAL:** The module certification API may change without deprecation until the two-genre promotion gate.

`certifyModule` checks a Syncplay component definition before it enters a game composition:

```ts
import {
  certifyModule,
  type SyncplayModuleCertificationFixture,
} from '@series-inc/rundot-syncplay/tools'

const result = await certifyModule(characterComponent, fixture)
if (!result.ok) {
  throw new Error(result.failures.map(({ code }) => code).join(','))
}
```

The fixture supplies four things:

* module source paths;
* the existing `runSyncplaySynctest` runtime factory and policy;
* pairs of equivalent inputs constructed in different iteration orders;
* warmup, sample count, and strict step-time/retained-heap limits.

Certification delegates straight, rollback, replay, hydration, and input fuzzing to `runSyncplaySynctest`. It does not implement a second replay engine.

## Module source policy

Source below `src/modules/` cannot use ambient `Math.*` or `Date`, import React, React Native, Three.js, R3F, or another render module, own a random stream, or iterate unordered object keys. Use deterministic math from the module context, explicit authority inputs for random outcomes, and stable collection helpers.

The policy follows imports within `src/modules/`. Engine helpers outside that directory keep the generic Syncplay determinism policy and are not accidentally reclassified as module source.

## Resource protocol

Run certification under Node with `--expose-gc`:

```sh
node --expose-gc --import tsx certify-game-modules.ts
```

Step time uses five warmed p99 batches. Certification runs the full declared warm-up before each batch. It calculates p99 from the declared sample count. It compares the median batch p99 with the declared limit. One or two slow batches cannot change the median. A slowdown in three batches fails the unchanged limit. Retained heap bytes per step are measured after forced GC while every sampled result is kept reachable. Missing GC support fails closed. The result never reports a zero-byte pass when GC support is missing.

## Stable failure codes

* `SYNCPLAY_MODULE_CERT_STATIC_MATH`
* `SYNCPLAY_MODULE_CERT_STATIC_DATE`
* `SYNCPLAY_MODULE_CERT_STATIC_RENDER_IMPORT`
* `SYNCPLAY_MODULE_CERT_STATIC_RANDOM`
* `SYNCPLAY_MODULE_CERT_STATIC_ITERATION`
* `SYNCPLAY_MODULE_CERT_STATIC_DETERMINISM`
* `SYNCPLAY_MODULE_CERT_ITERATION_ORDER`
* `SYNCPLAY_MODULE_CERT_SYNCTEST_DIVERGENCE`
* `SYNCPLAY_MODULE_CERT_TIME_BUDGET`
* `SYNCPLAY_MODULE_CERT_ALLOCATION_BUDGET`
* `SYNCPLAY_MODULE_CERT_EXECUTION_ERROR`

Malformed fixture declarations throw `SYNCPLAY_MODULE_CERT_FIXTURE_INVALID` before module code executes. Component, runtime, and fixture callback exceptions become `SYNCPLAY_MODULE_CERT_EXECUTION_ERROR`; their arbitrary error text is not copied into the result.