# Rehearsal record

Repository: https://github.com/yongquantan/rsvp-handoff-demo
Baseline commit: 8421a29

## Verified baseline
Independent browser reproduction: cancel Alex -> 1 / 2 seats filled; Jo stays waitlisted. This is the intentionally defective starting point.

## Verified local run
Devin CLI model: SWE-2 High (swe-2-high).
Session: zinc-swordfish.
The exported session shows 3 failures before the fix and 6 passing checks after it. Codex independently reran the checks: 6/6 pass. Reviewed diff: 2 files, 10 added lines (5 implementation lines and 5 regression-test lines).

Local browser verification also passed: cancelling Alex promotes Jo, and 2 / 2 seats remain filled.

## Native handoff verified
Actual cloud session: https://app.devin.ai/sessions/6548b19d884c4305bb59de00bcc2dec8
Native /handoff succeeded from the same resumed CLI conversation. API inspection confirmed repository, branch, recent prompt, and the attached changes_from_local_clone.diff. Cloud reported applying the carried diff and proceeding with tests, browser verification, and a PR.

## Cloud completed
PR: https://github.com/yongquantan/rsvp-handoff-demo/pull/1 — open, not merged.
Cloud final report: npm test 6/6 passed. Browser verified Alex cancellation promotes Jo, then Jo cancellation promotes Lee; cancelling waitlisted Lee promotes nobody. Repeated API cancellation three times left state unchanged; unknown registration ID returned 404.
Codex fetched the PR and compared the changed files against the independently tested local files: exact match, no diff.

Durable browser screenshots: [gallery](evidence/). Screenshots 1 and 2 were visually inspected independently and confirmed the reported states.

Cloud session created 07:01:25 UTC; final report 07:03:49 UTC on 2026-09-14 — 2 minutes 24 seconds. This is the observed rehearsal duration, not a timing guarantee.

This is a synthetic demo repository. Checks ran locally and in the cloud session; no GitHub CI result is claimed.

## Permission observation
The launch command requested accept-edits, but the local session export recorded Bypass, apparently after an interactive choice. The resumed handoff session was explicitly switched to accept-edits and its UI confirmed that mode. The full local run is not claimed to have remained in accept-edits mode.

The completed rehearsal is available as a clearly labelled fallback if a live demo is still running.