v7.1.19 — room refresh and connection recovery

Published October 19, 2026 UTC: release notes and downloads.

• Empty-room refresh waits for a fresh room snapshot and ordered player-roster delivery instead of closing a new room solely because that delivery is still catching up.
• Referee reconnection receives its full five-second connection budget after API pacing admission. Cooldowns no longer consume almost all of that connection time.
• Reconnection uses bounded backoff and reports one disconnect lifecycle event per outage.
• Included v7.1.18 work reconciles authoritative player rosters for counts and vote thresholds, repairs directory reporting during scoring, and finishes the initial catalog audit before startup readiness is evaluated.
• Unknown provider mutation outcomes are not blindly replayed.

Healthy opted-in exact v7.1.16 and owner-local v7.1.17 baselines receive the signed pilot update. All configured rooms on each host must be empty for the shared-controller transition; an empty room does not interrupt an occupied sibling, and late joins reset admission.

Settings, authorization, catalogs, ratings, statistics, titles, enrollment and saved operator choices remain. Full catalog contents, dependencies, storage formats and API rate limits are unchanged. Paused, deliberately stopped and ownership-uncertain hosts retain their checks.

143 affected tests passed, including delayed pacing admission followed by real TLS negotiation, ordered roster recovery and uncertain-write non-replay. Separate exact-artifact installed upgrade checks covered both supported baselines. No third-party account or actual room was used in donor qualification.

Existing healthy signed updaters fetch the runtime. Fresh installations use the release's installer ZIP and README; GitHub's source ZIP is not the installer.

Release history · Room updates · Hosting guide