[ryanbr]

Follow-up to #1531. @bhelm corrected the Swift side's false claim that the two completeness guards are kept aligned by a parity test. The same falsehood was mirrored on the Android side, which that PR did not touch — so the two files were left openly contradicting each other.

What was wrong

ReportCompleteness.kt claimed its token map was "byte-identical to the Swift twin", and named a Swift type — ReportCompleteness — that does not exist. Swift's is CaptureCompleteness.

Measured against the ten tokens the test pins:

	Count	Examples
Match Swift exactly	4	sleep, workouts, recovery, hrv
Differ by a trailing fragment	3	universal, connection, display
No counterpart in Swift at all	3	import parser= vs import stage=, stepsEst vs stepsRaw, battery series= vs bank soc=

The divergence is not a bug

Each side must key on the tokens its own emitters write, and the two platforms word several of those lines differently. Aligning the maps would break the guard on whichever platform lost. The header now says so explicitly, with the three cases named, so nobody "corrects" them into alignment later.

The test was the worse half

ReportCompletenessParityTest asserted "Byte-identical to the Swift ReportCompleteness.killerTokens" — and then pinned literals held in the test file itself. It never read the Swift side, so it could not fail when the platforms drifted, and it stayed green throughout the divergence above.

A test that names a guarantee it cannot check is worse than no test, because it invites a reader to stop looking. That is exactly what happened here: the claim survived long enough for three tokens to lose their counterparts entirely.

Renamed to ReportCompletenessContractTest, with its doc rewritten to state what it actually guarantees — which is still worth having: Android's own map, section ordering and present/MISSING wording cannot change by accident. The cross-platform contract that does matter (the section a maintainer reads being identically shaped) is now called out as the real one.

Verification

• Comments and the class rename only — no assertion, token or logic changed. Verified mechanically: the sole non-comment diff line is class ReportCompletenessParityTest → class ReportCompletenessContractTest.
• All 17 tests pass under the new name; no stale references to the old class remain.
• Android 4,241 tests, 0 failures, matching main. Doc lint clean.
• No Swift touched, so app-build does not apply.