[ryanduguid]

Summary

These tests spent most of their time waiting out deadlines that fail by design. Test-only overrides shorten those waits, and only for the cases that wait them out. Production values don't change: the overrides exist only under cfg(test).

Wait	Where	Production and other cases	Cases that wait it out
Attachment budget floor	attachment_budget	45 s	5 s, the fixture's command budget
Watch after a refused close	3 close routines	up to 5 s	1 s

• The attachment override applies to the 4 upload faults that never settle. The successful upload keeps the 45 s floor, and wrong-receipt faults keep it too, so expiry can't stand in for the refusal they test.
• The close override starts after the close command returns, so the command keeps its full budget. The watch only reads the target list, so it needs no room for a durable write.
• The 5 s command budget is unchanged, including its warning about 1 s being too tight.

Evidence

Each test alone on Windows with --test-threads=1:

Test	Before	After
unconfirmed_upload_commands_or_cleanup_never_authorise_close_or_send	186 s	25 s
completed_close_retries_resume_only_the_exact_durable_marker	87 s	42 s
same_process_reconciliation_never_closes_running_or_uncertain_requests	69 s	26 s
owned_upload_must_reach_readiness_and_cleanup_before_send	49 s	10 s
Total	391 s	103 s

Another run of the same code with temporary logging took 80 s.

• Coverage: for all 25 cases in these tests, final status, recoveryRequired, error text, cleanup warning, close and send counts and the CDP request sequence (repeats collapsed) match main. This came from temporary logging, not committed.
• Before: library suite on main, 666 tests, 14 ignored.
  After: library suite, 666 passed, 0 failed, 14 ignored, on the commit before the final one.
• On the final head: cargo fmt --check and cargo clippy --all-targets --locked -- -D warnings pass; browser_target 5 passed, browser_retirement 10 passed.

Merge danger

Door: two-way. Blast radius: tests

A panicking test could leave an override set for later tests on the same thread, as the existing TEST_CANCELLED_SEND_SETTLE can. That only matters in a run that has already failed.

Unverified

• A clean before and after for the whole suite: both runs shared the machine with other work.
• The attachment budget as the cause of the upload wait is inferred from the code and the drop after changing only that floor; it wasn't timed separately.
• Not run locally on Linux; this PR's CI covers it.
• The full library suite was not rerun after the final commit, which moves the close override and is test-only.
• bridge_doctor_accepts_and_ignores_verbose_like_commander and local_browser_cli_and_mcp_admit_foreground_and_detached_rust_workers fail locally, as they do on main. No browser is installed here, which is the likely cause (not confirmed).

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues
0 New dependency risks

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud