Pin the voyage event roll on the module docks actually calls (CI red on main)

[dmccoystephenson]

Summary

• main has been red for its last two runs (35554330178, 35556530705) on tests/location/test_docks.py::test_a_voyage_that_founders_ends_early, the flake that Pin the event roll in the foundering-voyage test (flake) #201 was meant to fix.
• Pin the event roll in the foundering-voyage test (flake) #201's pin was applied to src.business.adventures.rollEvent, but src/location/docks.py imports from business import adventures. Because pytest.ini puts both . and src on the path, src.business.adventures and business.adventures are two separate module objects, so the patch never reached the call site and the real random.choice roll kept running. (The random.randint / random.random patches were unaffected: random is one shared module either way.)
• The roll is now pinned with patch.object(docks.adventures, "rollEvent", ...) — the module docks actually resolves — and the event is taken from that module's own EVENTS.
• The day assertion is tightened from < startingDay + legs to == startingDay + 1: with 51% of hull and a 99-point leak the boat founders on the first leg exactly. An unpinned roll now fails five runs in seven rather than one in forty, so a regression of this kind would surface on the first CI run rather than the fortieth.

No tracking issue — #200 was closed by #201; this is the gap found at triage (CI red on main). No open issues were skipped: the backlog was empty at triage time.

Test-only change; no production code, schema, or persistence path is touched.

Test plan

• Adversarial probe: with the real random.choice forced to draw driftwood every leg, the test on main fails and the test on this branch passes (the stash-and-run check for the rubric's Tests-fix item).
• 200 consecutive plain runs of the test: 0 failures on this branch, 2 failures on main's version.
• Full suite under the dummy SDL drivers: 910 passed.
• black on the changed file: unchanged.
• CI on this PR head.

This PR description was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).

---

drafted by Claude on behalf of Daniel Stephenson

🤖 Generated with Claude Code

[dmccoystephenson]

Self-review rubric (posted as a comment; the formal Review API is not available to this session):

• Scope: PASS — one file, tests/location/test_docks.py; the only hunks are the failing test's pin, its explanatory comment, the tightened day assertion, and the plan local that assertion no longer needs.
• Tests-new: no signal — no new public method or function.
• Tests-fix: PASS — empirical. With the fix stashed, a probe that forces the real random.choice to draw driftwood every leg makes the test FAIL, and 200 plain runs produce 2 failures; with the fix restored the adversarial probe PASSES and 200 plain runs produce 0 failures.
• Sibling structure: PASS — patch.object(<module the code under test resolves>, ...) is the pattern tests/business/test_boats.py:58 already uses (patch.object(boats.fish, "rollFishType", ...)); the # prepare / # call / # check layout is unchanged.
• Sibling renames: no signal — nothing renamed.
• Docs: PASS — README, schemas/*.json, PLANNING.md, and version.txt are unaffected by a test-only change; none describes test internals.
• Issue resolution: no signal — no Closes reference; Flaky test: test_a_voyage_that_founders_ends_early fails about one run in forty #200 was already closed by Pin the event roll in the foundering-voyage test (flake) #201 and this PR fixes the gap Pin the event roll in the foundering-voyage test (flake) #201 left, as the body states.
• CI: PASS — run 35710298793 green on head cc52cd3 (test job, 1m05s).
• Schema-sync: no signal — no persisted field touched.
• Money-format: no signal — no money display touched.
• Deterministic-tests: PASS — every random draw the test reaches is now pinned at the module docks actually calls (rollEvent) or on the shared random module (randint, random); loseCrew's random.choice is not reached because the leak's first choice only damages.
• Headless-pygame: no signal — no pygame path touched.
• camelCase: PASS — the one new local is sailedAdventures.

One observation outside the diff, not acted on here: the same dual-module trap (src.x vs x both importable via pytest.ini) would silently neutralise any future patch("src.<pkg>.<module>.<name>") where the caller resolved <pkg>.<module>. A sweep of tests/ found no other instance today — the remaining src.-prefixed patches target either the shared random/shutil modules or a function the test itself calls through the same src. module — so it was left as a note rather than an issue.

tests/location/test_docks.py:1520 — the == startingDay + 1 assertion is deliberately exact rather than <; it is what turns an unpinned roll from a one-in-forty flake into a five-in-seven failure, so a regression of this kind would show on the first CI run.

This comment was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).

---

drafted by Claude on behalf of Daniel Stephenson