# Local socket recurrence — 24 October 2026

At11:05UTC the local Desktop engine pipe was absent and Desktop/backend processes were stopped.
Starting the local engine reproduced the exact `initializing Inference manager ... dockerInference`
startup error at11:07:41UTC. This was a local workstation failure, separate from the unsafe hosted
runner configuration and the application tests. No production host was accessed.

The [previous socket recovery](../../../database/mariadb-standardisation/DOCKER_LOCAL_RECOVERY.md)
was rechecked. Both named directories contained only zero-byte reparse-point socket entries:
three in `%LOCALAPPDATA%/Docker/run`, one in `%LOCALAPPDATA%/docker-secrets-engine`.
Only Desktop/backend processes started by this attempt were stopped after identity/start-time checks.
The directories were moved to verified sibling archive paths and fresh empty originals created:

- `C:/Users/Connor/AppData/Local/Docker/run.stale-20261024-p9-requal-cccb40b8`
- `C:/Users/Connor/AppData/Local/docker-secrets-engine.stale-20261024-p9-requal-660ce455`

Archive metadata records11:11:31UTC. Subsequent Desktop startup exited0 and the engine reported
29.7.2. The task-owned disposable MariaDB service alone was started and became healthy on loopback33308.
The persistent development service was left stopped; its container/volume were not removed.
There was no factory reset, image/data/volume deletion, WSL disk operation or production restart.

Raw scoped recovery script and metadata remain in `.artifacts/p9-requal-01/`. Initial recovery
guard attempts stopped on a timezone comparison and a changed process identity; neither moved
directories. The process set and UTC comparison were corrected before the successful archive.
No application qualification is inferred from recovery alone. Socket recurrence and workstation
lifecycle remain local tooling limitations; preserve the archives and diagnose another recurrence
before repeating any action. Do not archive broader Docker data directories.