[digitalerdude]

Supersedes #833, which is why this is a new branch rather than a rebase. Addresses both review points from that PR.

The chip has moved since July — #859 made its state resolvable without an Application, #1472 changed the age token, #1208 its height — so the old diff no longer described the code it patched. Rebuilding on current main was cheaper than replaying it, and the surrounding code had changed enough that the result is not the same patch.

1. Plurals — the substantive half

"1 pages behind" was reported, but the accessibility line carries TWO counts, and Android <plurals> inflects one quantity per resource. So each count gets its own <plurals> and a template joins them. The joiner lives inside the template, never on a resource edge — aapt2 strips edge whitespace, which is the trap #2041 had to clean up across 59 sites.

That also fixes "1 chunks", which predates this change and was wrong in every locale.

Apple needed less than expected: "Syncing strap history, %lld chunks" already had plural variations in 9 of 10 catalogue locales. Only pl was still flat, and it is now inflected like the rest. The template's wording is derived from each locale's existing chunk translation plus its joiner, so nothing already reviewed is re-translated.

Seven Android locales, not the five the old branch knew: pl and ru landed since July.

2. "at connect", per the second review point

The figure is sampled once from the connect-time GET_DATA_RANGE reply and never re-polled (#377), so a static number under a "syncing" label otherwise reads as a stalled live one. The copy now says which it is.

A zero backlog is dropped, not rendered

"0 pages behind at connect" beside a running sync contradicts itself, and a zero sample gives a reader nothing to act on. The rule lives in resolve on both platforms so neither view layer decides it alone, and it is pinned by a test on each side.

Where it surfaces differs by platform, deliberately

Android's chip is a capsule with room for a second text run, so the figure is visible there (maxLines = 1 + ellipsis so a long localisation cannot push the header cluster off a narrow screen). Apple's is a spinner with the count inside it and no room for a phrase, so the figure lands in the accessibility read-out that already spoke the chunk count. Feature-level parity, not pixel parity.

Verification

• ./gradlew compileFullDebugKotlin — BUILD SUCCESSFUL
• ./gradlew testFullDebugUnitTest --tests "com.noop.ui.SyncChipStateTest" — BUILD SUCCESSFUL (4 new cases)
• xcodebuild … -scheme Strand -destination 'platform=macOS' build — BUILD SUCCEEDED. Worth noting the catalogue is genuinely checked by this: an earlier attempt failed here with Missing required key 'stringUnit' when a variations block was malformed, so a green build is real evidence about the strings and not only about the Swift.
• Tools/doc_comment_lint.py — OK

Not established, and I would rather say so than imply otherwise:

• The four new Swift tests compile and were launched under xcodebuild … test, but my invocation filtered its own output and I did not capture a pass/fail line. Treat them as unrun until CI or someone else confirms them. The Kotlin twins did pass.
• pl and ru plural forms follow CLDR and the existing entries in each file. They are NOT native-speaker checked; that is the whole claim.
• Chip width on a narrow display is unverified. The detail run is new visible text in the Today header cluster, which Android Today: on a narrow display the date and the third hero ring clip at the screen edge #1835 already reports as tight at 455 px. Ellipsis bounds it, but whether it looks right there has not been seen. A shorter visible form (dropping "at connect" from the chip while keeping it in the accessibility line) is an easy alternative if you prefer.
• No hardware run. The decode path this reads from is unchanged and already confirmed on 4.0 and 5.0/MG (test(tools): collect test_whoop_spot_hrv too — 8 more tests that never ran #815); what is new is display, the zero rule, and wording.

[ryanbr]

Thanks @digitalerdude. Reviewed on a checkout, and ran the two gates that CI does not.

Your Swift tests DID run, and passed

You flagged them as unrun because your invocation swallowed the result. They ran in CI: app-build.yml has a Test Strand step, xcodebuild -scheme Strand -destination 'platform=macOS' test, gated to the macOS leg because StrandTests is hosted in the macOS app. On this PR that step reports success:

JOB: build (Strand, generic/platform=macOS, ...) [success]
   success   Build Strand
   success   Test Strand

So StrandTests/SyncChipStateTests.swift executed. The NOOPiOS leg skips it, which is expected.

Ran here, because CI does not

• Tools/i18n_audit.py --ci main: OK. No new gaps, no new English-only translations, both ratchets moving down.
• ./gradlew lintVitalFullRelease -PstagingRelease: BUILD SUCCESSFUL. This is the one that matters for a strings change, since fatal ExtraTranslation is not caught by a compile or by the audit, and a plural resource added to seven locales is exactly its territory.
• SyncChipStateTest: 11 tests, 0 failures, forced with --rerun-tasks.
• Merging onto current main leaves the parity authority reproducing, so this will not need a re-derivation behind it.

The plural work holds up

The joiner is inside the template (%1$s, %2$s) and never on a resource edge, so the aapt2 edge-strip that #2041 had to clean up across 59 sites cannot bite. That was the right call and it is the part most easily got wrong.

All eight files carry both plurals. zh correctly has other alone.

On pl and ru, which you called out as not native-checked: they follow CLDR, including the detail people most often miss. other in both languages is the FRACTIONAL form, and you used the genitive singular for it, 1,5 strony and 1,5 страницы, rather than repeating the few form. one/few/many are right too: strona/strony/stron and страница/страницы/страниц. I am not a native speaker either, but that is a correct plural table rather than a plausible-looking one.

Fixing "1 chunks" while here is worth more than it sounds, given it was wrong in every locale.

Two notes, neither blocking

The base is stale. This branch is on faabe478d; main is now several commits ahead, including a parity authority re-derivation. Worth an update before merge, and note that the gate now also runs on pushes to main, so a stale base is likelier to show up as a surprise than it used to be.

"at connect" is the right fix for the second review point. The figure comes from the connect-time GET_DATA_RANGE reply and is never re-polled (#377), so a bare number under a "syncing" label reads as live. Saying which it is costs one phrase and removes the whole class of confusion.

Dropping a zero rather than rendering it, with the rule in resolve on both platforms and a test each side, is also the right shape: neither view layer gets to decide it alone.