Published October 28 UTC on GitHub and the signed pilot feed. An offered or downloaded update is not an installed host; Live Lobbies shows observed room versions.

The two-hour empty-room recreation interval is retained. Provider readiness and a usable starting map are checked before closing the old room. Unavailable readiness defers recreation with backoff. Updates, maintenance, managed settings and bookings share lifecycle admission.

A failed replacement fences its affected alias while healthy siblings keep playing on a qualified shared connection and account authority. Unknown creation is never replayed or silently cleared. Genuine shared transport or authority failures retain guarded shutdown/recovery. Safe numeric rejection codes, selected map/ruleset context and any returned creation candidate are preserved; credentials and raw upstream error bodies are excluded.

The exact recreation repair guide and pinned repair launcher cover stopped existing hosts. The repair verifies the installation and dependency environment, saves helper preferences, takes a private consistent backup, upgrades in place and reviews ownership. Settings, credentials, enrollment, catalogs, ratings/statistics, Discord assignments and deliberately stopped slots are retained. Recovery pins change only through verified installation lineage.

At most one startup call is admitted and recorded durably before dispatch. A repeat after verified success observes the existing rooms without restarting them. A never-started installation with an explicit empty native journal proceeds to normal first-start admission; an existing unknown creation remains blocked. Do not reinstall, delete receipts, edit journals or repeat an uncertain start.

ownership-inventory makes two complete stable authenticated traversals of non-deleted owned realtime history, including ended rooms. It can discover the actual replacement ID but cannot prove absence of soft-deleted upstream rooms. Empty history or a generic rejected reply never authorizes ownership clearing. The previously closed room is not a replacement candidate. If discovery cannot identify it, upstream must confirm creation outcome including deleted rows and referee associations, or supply the actual replacement ID for direct owner/end verification. Publication does not resolve existing holds.

54 current repair tests passed. The exact rebuilt package upgraded a full-catalog v7.3.4 donor once and started healthy Loved and fixed Standard rooms with nondefault difficulty ranges, countdown, room ceiling, seeded ratings/enrollment and two reserved slots. Repeat made zero starts with unchanged workers/room IDs; native shutdown and temporary infrastructure cleanup passed. Expected first-start room files and empty locks were independently checked against native rendered settings and the pre-upgrade backup after an initial observer rejected their addition; no extra startup was made.

Earlier frozen-build evidence covers 411 affected cases, native stopped v7.3.4/6/7 upgrades, rejected/partial creation, occupied sibling gameplay and commands, candidate discovery, active-clear refusal, interrupted prepared recovery and explicit completion, repeated recovery, all six full catalogs and an automatic signed upgrade preserving a queued settings change. The rebuild changes only the stopped repair workflow; 363 other bundle files remain byte-identical. The full repository suite and full 411-case suite were not rerun after that localized correction. Original failed QA attempts remain retained. No production credentials or real-account test rooms were used. Actual installation, gameplay, website/Discord, ratings/statistics and updater/recovery health are verified separately per host.

Runtime SHA-256: c466ece586689b15d105fbf90b3b72fb04b5b3a3774c9b2bc89fc124ecc2499b.