# Land queue reports failure on a completed land

outcome: a `ship.sh land` client that loses contact with the conductor never reports `reachability-failed` for a land that actually merged; agents stop being misled into duplicate direct-lands.
status: IDLE
source request: agent-found defect 2026-08-15 — ft-narrowing lane's `ship.sh land` client waited 56 minutes, reported `reachability-failed`, while the conductor had already merged the branch (`8e926172d`). The lane nearly performed a duplicate land and burned an hour of wall clock.

## Defect

The client-side wait path times out and reports failure without re-checking whether the ticket's branch actually reached origin/main. Failure report on a SUCCEEDED land is the worst shape: it invites a duplicate land or a direct-land bypass.

## Acceptance criteria

- On wait timeout/reachability loss, the client re-fetches and checks origin/main for the candidate commit before reporting; a merged land reports `landed (late)`, never a failure.
- A regression test covers the timeout-after-merge sequence.
- Note: the new `ship.sh submit` flow sidesteps the wait entirely — verify whether the legacy `land` path is worth fixing or should just refuse sooner in favor of `submit` (the phase-out already planned in CLAUDE.md).

## Receipt

2026-08-15: registered from the lane's Not-verified flag. Next executable action: owner of the lander/conductor lane picks it up (check overlap with the submit phase-out before coding).