PR 28954 real-environment verification PR head: 6d36ce6f015183bf61376d17ebba9096f10f0d0e Baseline: github/main f6503f0e71 Environment: disposable Linux Boat VM, Chrome /usr/bin/google-chrome, Playwright 1.60.0 from repository dependencies, one dev server at a time. Both environments: create /home/user/settle-verification as a real git repo; add it through the app's Add project -> Local folder flow; configure an Action through the app with the settle trigger enabled: printf "settled %s\n" "$PWD" >> /tmp/settlement-actions.log Create New worktree threads through the composer and run real Codex gpt-6.1-sol turns that read README.md. No mocked provider or fixture events. BASELINE $ git worktree add ../main github/main $ vp i $ setsid nohup vp run dev > /tmp/main-dev.log 2>&1 & [dev-runner] serverPort=16733 webPort=8693 baseDir=/home/user/main/.t3 Real turn completed: "settlement verification ready." Click Settle thread -> action appends one line. Delete thread -> confirm -> confirm worktree deletion. Thread removed from client and worktree removed. $ curl http://127.0.0.1:9334 # local read-only Node inspector helper map after deletion: [["0408711c-4380-42d4-80a0-0dd983aff1d8",1793668638908]] RESULT: deleted thread's settle record is still retained on main. Stop baseline using captured process-group PID before PR startup. PR $ git switch --detach 6d36ce6f015183bf61376d17ebba9096f10f0d0e $ vp i $ T3CODE_HOME=/home/user/t3code/.t3 T3CODE_DEV_SEED=0 setsid nohup vp run dev > /tmp/dev.log 2>&1 & [dev-runner] serverPort=13773 webPort=5733 baseDir=/home/user/t3code/.t3 Two real Codex worktree turns completed: target 43881a03-0001-4c1e-8d61-8530a6f204b8: settlement verification ready. neighbor 92d186d8-08bc-49ce-9b24-d949b354ef05: live neighbor ready. Settle neighbor, then target via sidebar buttons. map: [["92d186d8-08bc-49ce-9b24-d949b354ef05",1793668875641], ["43881a03-0001-4c1e-8d61-8530a6f204b8",1793668895020]] Action log: one baseline run, one neighbor run, one target run. Un-settle target -> settle target again. map: [["92d186d8-08bc-49ce-9b24-d949b354ef05",1793668875641], ["43881a03-0001-4c1e-8d61-8530a6f204b8",1793668919345]] Action log: target runs a second time, four total lines. Delete target through sidebar context menu -> confirm -> confirm worktree deletion. map: [["92d186d8-08bc-49ce-9b24-d949b354ef05",1793668875641]] Action log stays four lines; neighbor's entry and timestamp are preserved. RESULT: target cleanup succeeds and neighboring settlement state survives. Observation method: enable Node's inspector on the actual dev server; a breakpoint immediately after settleActionRunAt.set captures a remote reference to the existing map, then resumes immediately. Subsequent reads use Array.from(this.entries()) through Runtime.callFunctionOn. No source edits, replacement services, event replay, or synthetic records. CHECKS (Blacksmith task-owned worker) $ blacksmith-check run --id 'cd apps/server && vp test run src/orchestration-v2/ThreadSettlementService.test.ts' 31 tests passed. $ blacksmith-check run --id 'git fetch --depth=5 https://github.com/pingdotgg/t3code.git refs/pull/28954/head && vp run check:changed fa31b66bbb2ae5234205a8628a78e038d917202f' pass format, lint, knip (2), graphql, typecheck (3 packages), test (2 files / 32 tests). History fetch was required because the worker snapshot initially lacked a merge base. Existing GitHub CI: all applicable reported checks pass, others skipped. NOT VERIFIED Native iOS/Android clients, Electron/macOS/Windows shells, other providers, production heap impact, repeated identical-timestamp event delivery in the real client (covered by the existing focused suite), retry/failure races. No user-visible performance claim is made from these small real workloads.