docs(mobile): G2 evidence — T054/T055 proven, T056 3/5 drills green

[h4yfans]

What

• T054 → [x]: device proof (iPhone 12 Pro: conformance 26/26 + round-trip 6/6, 2026-08-24 21:07) + simulator closure run after the full fix chain (22:50 — bulk ladder applied, nothing stuck, list complete).
• T055 → [x]: Maestro us1-unlock-browse.yaml green on the iPhone 17 Pro simulator (2026-08-25 01:44) — stored session short-circuits sign-in/unlock as designed, browse→open-note asserted. One flow fix: the note body sets accessibilityLabel="Note content" (no testID), so the final assert matches the label instead of id:.
• T056 → 3/5 annotated: kill-switch ✓ (staging client_policies flip → Read-only banner on next bg→fg, no restart; clears on re-enable), version-gate ✓ (99.0.0 floor → update banner → clears on NULL), attribution ✓ (staging D1: zero ios write rows across sync_items/crdt_* — US1 pull-only proven).
• G2 checkpoint block refreshed: remaining items are the final phone pass after fix(mobile): note_bodies.path is not unique — a title collision no longer kills the apply batch #1826, the <5 s visibility drill (owner call on trigger path), and SC-004 (owner call on seeding the 10k vault).

Verification

Staging client_policies restored to writes_enabled=1, min_write_version=NULL after the drills (verified by query). Screenshots + attribution JSON captured in the session.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[h4yfans]

Two mobile sync fixes landed on this branch during the T056 drills (both proven on simulator against staging):

1. Body edits were invisible to mobile — the record feed never carries body changes (updates push content: null; body saves do not touch the record row), and the pull engine fetched CRDT bodies only for changedNoteIds. Desktop peers see body edits because desktop probes every note's CRDT sequence each pass. Mobile now probes the 30-day recent window each pass (30 s floor), mirroring desktop; older bodies stay on-demand.
2. An open app never re-pulled — the only triggers were fg/online transitions and the BG task, so visibility of a desktop edit was set by the next app switch (uniform ~43 s in the drill). A 5 s foreground poll now runs while the app is active.

20-trial drill results (desktop → staging → simulator, app foreground): 0 misses; body median 32.2 s / p95 34.0 s; metadata median 28.1 s / p95 33.1 s (best trial 7.9 s). The floor is set by the DESKTOP push side: PushCoordinator drains the outbox and CRDT state on 30-second cycles (CrdtSnapshotScheduler + push cycle), so <5 s end-to-end is not reachable for any edit type without making desktop pushes event-driven — flagging that as an owner decision (Phase 4 write-path work is the natural home).

🤖 Generated with Claude Code

[github-actions]

React Doctor found no new issues. 🎉

_{Reviewed by React Doctor for commit 92090b1.}