The work-row shimmer now uses a timer-driven 12-step progress cycle with a pause and cleanup. The mobile component uses a native animated value and starts the cycle only when its existing width, reduced-motion, app-state, and screen-focus conditions pass.

Changes: Mobile work-row shimmer

Layer / File(s)	Summary
Timed shimmer progress apps/mobile/src/features/threads/work-shimmer-steps.ts, apps/mobile/src/features/threads/work-shimmer-steps.test.ts	A new helper emits 12 progress steps, pauses between sweeps, and returns a function that clears the pending timer. A fake-timer test checks the step values, pause, repeated sweep, and cleanup.
Native shimmer integration apps/mobile/src/features/threads/thread-work-log.tsx	The component uses a native animated value and starts the helper only when its existing conditions pass. The sweep interpolation maps progress directly from 0 to 1.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Possibly related issues

• pingdotgg/t3code#28113: The change targets the work-row shimmer’s per-frame redraws by using timed progress steps.
• pingdotgg/t3code#27325: The change targets the work-row shimmer, but the issue also reports other streaming power costs.
• pingdotgg/t3code#9661: The change addresses the work-row shimmer portion of the performance audit.

Suggested reviewers: juliusmarminge, maria-rcks

No merge-blocking issue was identified. Android Release performance remains unverified.

^Failed Failed checks: 1 inconclusive

Check name	Status	Explanation	Resolution
Description check	Inconclusive	The description covers the problem (native timing loop updating every frame, including the pause), the change (12 steps per sweep, no updates during the pause, guards retained), and verification (befo…	Complete the "Scope and approval" section. Link the triaged bug issue or the discussion that gives explicit maintainer approval of the direction and scope, with the approval comment. If the change is a very small, focused fix of an obvious …

^Passed Passed checks: 3 passed

Check name	Status	Explanation
Title check	Passed	The title "fix(mobile): stop the work-row shimmer from updating every frame" follows conventional commit form and names the main change. The changeset replaces the looping NativeAnimated.timing with s…
Linked Issues check	Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	Passed	Check skipped because no linked issues were found for this pull request.

Full details: Title check

Explanation

The title "fix(mobile): stop the work-row shimmer from updating every frame" follows conventional commit form and names the main change. The changeset replaces the looping NativeAnimated.timing with startShimmerSteps, which advances the native mask in 12 steps per sweep and emits no updates during the pause. The title is specific and matches the summary and objectives.

Full details: Description check

Explanation

The description covers the problem (native timing loop updating every frame, including the pause), the change (12 steps per sweep, no updates during the pause, guards retained), and verification (before/after recordings on iOS and Android, a check:changed run, and cadence measurements). It links issue #28113. It does not include a triaged approval comment or discussion link in the Scope and approval section, which the template requires. It also does not clearly state that the PR is a focused fix of an obvious bug without prior approval. The description is mostly complete, but the scope and approval section is missing its required evidence.

Resolution

Complete the "Scope and approval" section. Link the triaged bug issue or the discussion that gives explicit maintainer approval of the direction and scope, with the approval comment. If the change is a very small, focused fix of an obvious bug, state why it qualifies without prior approval. Also state the Problem section in one or two sentences with expected behavior, since the description currently only uses the What's fixed heading. Keep the Problem, Change, and Scope and approval headings from the template.

Generate unit tests (beta)

• Create a new PR

R5 evidence audit

R5: resolved. Severity was blocker for missing native cadence evidence. The new trace establishes the work-row's native sweep and pause cadence in the instrumented iOS Debug simulator. No R5 blocker remains within that scope. R6, the Android Release comparison, remains open with the parent. This is not a PR readiness assessment.

Reviewed PR https://github.com/pingdotgg/t3code/pull/28530 at head 2f5131494415552c16c8c8e72380d6c9a6051889 against main 7195b699547b538e49f758e314dde5c8c5d907cf. GitHub still reported those exact revisions and a draft PR during this audit. No Julius commits appeared in its two-commit history. R1–R3 and R4 retain the parent's resolved status; I did not re-review them. R4's supplied evidence is comment https://github.com/pingdotgg/t3code/pull/28530#issuecomment-6558055997.

Evidence directory: /home/shivam/Developer/.state/fun/runs/r_0bbogunlphz77a00/out/. I inspected the report, observer source, RN native source copies, both raw native logs, attribution, analysis script and CSVs, fixture/player, backend events, fingerprints, manifests, LLDB logs, video metadata, both turn-1 contact sheets and the full-size active head image.

Independent recount

I wrote and ran /tmp/issue-28113/ios-cadence-independent-counts.py, without running the supplied analyzer or modifying its outputs. Full independent results are in /tmp/issue-28113/ios-cadence-independent-counts.json. All 16 observed window boundaries and their write counts match native-windows.csv.

Cycles were discovered from observed native zero resets on main and observed returns to 1/12 on head. Each complete head group contains the ordered native values 1/12 through 12/12 within float precision. Every head root write pairs with a native set command within 1 ms. Both native interpolation outputs follow every positive root write, including all 55 and 56 positive head writes. Each head lifetime also contains four initial zeros and a final partial sweep. Counts were not inferred from configured timer durations.

Attribution and windows

I segmented logs by observer-loaded PID before reconstructing graphs. Main's earlier PID 53628 contains only observer setup, so it cannot contaminate measured PID 56410. Head PID is 60098. All eight hooks succeeded in both measured processes.

Each sustained root has two paths through interpolation, translateX transform, style and props to connected view tags. Main paths end at views 1780/1808 and 2104/2132; head paths end at 1362/1390 and 1720/1748. Their paired ranges are ±72 to ∓129, with main's flat third endpoint. This matches the exact ShimmeringWorkContent source, its mask and countertranslation. The shorter 84-point Thinking graphs are excluded. Backend run ordinals 2–5 independently show the same execute item active for 12.516, 12.517, 12.514 and 12.515 seconds, aligning with these root lifetimes and the pictured Running Replay row.

Main counts use [observed zero reset, first plateau write) and [plateau write, next reset). Head counts include first through last positive step, then the open interval until the next 1/12. These are different boundaries and must be named. The head quiet gaps are 1564.115–1566.870 ms, including the next sweep's first-step delay. In every matched 1449 ms interior pause window, main has 82–86 root writes and target display steps; head has zero. Root and interpolation nodes remain mounted throughout. Head repeat periods are 2849.741–2916.973 ms, so an exact 2800 ms runtime period is unproven.

Observer and provenance limits

observer.m wraps native setters and NativeAnimated manager callbacks while preserving their original implementations. It contains no per-write breakpoint or JS scheduler instrumentation. Main's sustained roots have actual infinite native frame drivers with 169 samples. Head has zero native animation starts and zero manager display steps anywhere in its measured process log. Setter observations alone would not prove display/compositor idleness; this trace separately covers RN NativeAnimated callbacks.

The observer loads before the measured lifetimes. Flush markers are after the respective drops: main 494745.998/494806.827, head 494968.730/495181.630. The four flush transcripts attach, flush and detach with no breakpoint commands. The delayed final head flush is about 191 seconds after its row dropped. The command sequence and logs support the reported absence of measurement-time debugger stops, although the archive is not a continuous debugger audit. Self-timed bookkeeping is 97.256 ms/6179 calls on main and 10.868 ms/625 on head. These omit some dispatch/OS overhead and double-count nested display-event bookkeeping. They are not an uninstrumented performance control.

I byte-compared both archived work-log sources and the head helper with the specified git objects. They match. The replay player matches both revisions after normalizing only SCENARIOS. The fixture hash is a77c199c1d3a83188fc2002e8456c02aafe65dec57bc3fad4fb68f0247a001b6. Both 118-source fingerprint records are byte-identical at 887f84d7e8ee2e1063658bf884c9d17f60d6cbb3. Native config/dependency inputs have no revision diff. The cached client is arm64 simulator Debug, T3 Code Dev 2.2.0, RN 0.88.0, on shiv's iPad mini A17 Pro/iOS 27.0 simulator. Its 323-file manifest hash matches 50a059c6c63a9d0f17ab10bb736489f33c61eac903ba54a16099ec95df2bb6d5, and its listed executable/framework hashes match the report. The original app bundle is not in this archive, so installed binary identity rests on the run's install/load records and manifest rather than my independent binary rehash. No new build was claimed or performed.

All 55 artifact-manifest entries and all four MP4 hashes verify. Contact sheets visibly cover trigger, Running Replay and completion. They supplement the native trace and do not establish its cadence. The trace's periods do not use video frame rate. The measurement JSON preserves selected invocations and complete outputs for those invocations; it does not preserve every setup/Metro/compiler command's raw output. This limits command-level provenance, without undermining the counted native events.

Safe claim and remaining work

Safe: on these exact JS revisions using the matching instrumented Debug native client, eight complete observed cycles per revision change from 81–82 native progress writes during each sweep and 83–88 during its plateau to 12 positive writes and zero quiet-gap writes. Head also has zero observed RN NativeAnimated manager display callbacks in the captured process.

Unproven: physical iPhone behavior, Release or uninstrumented timing, power/battery savings, GPU/compositor redraw counts or app-wide idleness. This subtask also does not prove focus/background/Reduce Motion regressions, Android Release behavior, or a full PR-ready result. A pending timeout spans the pause; say "no progress updates during the pause," not "no timer exists."

Parent should publish/link the new R5 evidence and retain these limits. R6 is still open. No device UI, build, app test, server, source edit, post or push was performed by this reviewer.

Native timestamps are relative to root creation, rounded to 0.001 ms. Native values are printed to nine significant digits. These are observations, including setup zeros and final partial sweeps.