Singleton policy fixture isolation¶
Six pure metadata and folder-policy cases now run in SingletonAssetPolicyTests, without asset creation, save, refresh or cleanup. The seven integration cases keep the original SingletonAssetPollutionPreventionTests setup and cleanup. Both fixtures share the same cached type registry. All thirteen case bodies remain covered; no production test hooks were added.
Three fresh-process pairs in Windows x64 Unity 6000.6.7f1 Editor Mono retain the same thirteen normalized case identities. The second pair reverses order. Every process passes all cases, returns raw exit zero and reaches both stream ends without stopping output drainage. Fixture ownership checks distinguish the baseline's zero policy-fixture cases from the candidate's six.
| Pair | Original summed fixture time | Split fixture time | After/before |
|---|---|---|---|
| 1 | 8.265951 s | 7.973383 s | 0.9646 |
| 2 | 8.322782 s | 7.902470 s | 0.9495 |
| 3 | 10.121433 s | 7.837843 s | 0.7744 |
The median decreases from 8.322782 s to 7.902470 s, about 5%. These observations measure the selected fixtures, including their setup and cleanup, and exclude editor startup. They establish no whole-suite speed claim. Shared-machine variation is visible in the third baseline process.
The replay archive contains exact source variants, all six complete NUnit trees, process receipts, credential-checked logs, the native controller and a SHA-256 inventory. The controller rejects changed sources, missing cases and unhealthy process completion. Its README gives replay instructions. A preparatory count guard incorrectly expected fourteen cases and rejected that capture; it contributes no accepted observation.