The existing test suite is being reviewed progressively to make sure no test crystallizes a bug (that is, asserts as "correct" a behavior of the untested code that is actually wrong). To tell reviewed tests from not-yet-reviewed ones at a glance, we use a filename suffix:
*_test.go(no marker) = legacy, not yet reviewed.*_audited_test.go= reviewed, in one of two cases:- it was written after the baseline (with the code under test already known good), or
- it is a legacy file that was reread and approved (no bug crystallized).
Go only requires the file to end in _test.go, so *_audited_test.go is collected by
go test like any other test: the convention has no runtime cost.
- When you write new tests, name them
<something>_audited_test.godirectly. - When you finish reviewing a legacy file, rename it with
git mv:git mv foo_test.go foo_audited_test.go(history is preserved; the rename is itself the review log, so make one atomic commit per file or area). - For huge files reviewed in pieces, use per-function granularity with a dated tag
comment above the
func Test...:// audited: 2026-06-09 verified it does not crystallize a bug - Do not bulk-rename the legacy files: a rename happens only once a review is actually complete.
# how many are left to review
find . -name '*_test.go' -not -name '*_audited_test.go' -not -path './vendor/*' | wc -l
# how many are already reviewed
find . -name '*_audited_test.go' -not -path './vendor/*' | wc -l
# next ones to do in a package
find internal/orchestrator -name '*_test.go' -not -name '*_audited_test.go'As of this writing there are 347 legacy files still to review and 30 audited files.
The tag tests-audit-baseline (on 3222a30, 2026-06-09) marks the state of the
suite at the start of the review. To see what changed in the tests relative to that
point:
git diff --name-only tests-audit-baseline -- '*_test.go'The _audited_ suffix answers "reviewed / to review"; the tag answers "created
before/after the baseline". Together they cover both questions.
internal/uitest holds test-only helpers for the Charm/bubbletea driver tests. It is
imported only from _test.go files and is never linked into the production binary.
The driver tests poll a render buffer until a screen or line appears. Under the race
detector the bubbletea event loop runs roughly an order of magnitude slower, so a fixed
wall-clock deadline can fire spuriously even when the logic is correct.
uitest.Deadline(base) scales a base render-poll timeout by a race-aware factor
(raceScale = 8 under -race, 1 otherwise, selected by build tag). Because those polls
return as soon as the condition is met, a wider deadline is free on the success path: it
only adds headroom before a genuine hang is reported. Use it only for UI-render
polling deadlines, never for a test that asserts an operation's own timeout behavior.
The driver tests are also sensitive to CPU saturation: run them one package at a
time rather than fanning out go test (or -race) across packages in parallel, or the
Charm driver can still hit spurious render-poll timeouts under load. Always read the real
exit code rather than piping the output through tail or similar.
The review grew out of the pre-test coverage and correctness audit of 2026-06-09
(diagnostics/coverage-audit-2026-06-09.md, dev @ 3222a30, the same commit as the
tests-audit-baseline tag), which surveyed how much of the code we were about to test
was still unvetted. Writing tests over unvetted code risks freezing its defects in
place, so the rule is: verify the code is correct first, then write the test, which is
born _audited.