# PLAY2 verification

Evidence: [raw logs, scripts, readbacks and screenshots](https://files.bagerbach.com/play2-evidence-3advzlzwqc3p.zip). [Dispatch ledger](https://files.bagerbach.com/dispatches-vjg72wik62jy.tsv).

Player after saving: [master at 12 seconds](https://files.bagerbach.com/master-current-after-flush-v0iikk7824ya.png), [fix at 45 seconds](https://files.bagerbach.com/final-after-flush-c2pe7iswrfyu.png), [mini player at 45 seconds](https://files.bagerbach.com/final-mini-after-sync-tgz2zpiemg8x.png).

Worktree: `/home/christian/.t3/worktrees/PodNotes/fix-synced-progress-handoff`.
Branch: `fix/synced-progress-handoff`. Node v22.23.2, Linux headless, isolated Obsidian 1.14.4. All commands run through `fnm exec --using 22`; scratch is under `/data/home/tmp/podnotes-PLAY2`.

## Reproduction

The baseline was first reproduced on `ea748cc`, before implementation, and repeated on current master `427c9d5` after UI1b merged. The project verify skill created a 60-second local WAV at `Fixtures/Local Fixture.wav`. Its URI was `obsidian://podnotes?episodeName=PLAY2%20Handoff&url=Fixtures%2FLocal%20Fixture.wav&time=10`. The driver opened the player and paused using its visible button. It changed the episode progress in `data.json` to 45 seconds and awaited `onExternalSettingsChange`.

On current master, the player stayed at 12.077712 while memory and disk held 45. `player.save()` followed by `flushSaves()` put memory and disk back at 12.077712. The player screenshots show 00:00:12 both after sync and after flush. A subsequent seek using the visible progress bar saved 30.

Commands: `.obsidian-e2e-artifacts/play2/live-handoff.py master-current`. Exact commands and readbacks are in `master-current-commands.json` and `master-current-live.json`. The initial baseline is recorded in `master-sync.log`, `master-flush.log`, `master-live.json`, `master-unit.log` and `master-e2e.log`.

Five original unit regressions and both new real-app handoff journeys failed before the production fix. The additional regression run had ten expected failures. Five new controls already passed on master: API pause, media-key pause, seek before sync, segment preview and saving a seek through the real writer. No assertion was weakened in the existing tests; the playback fixtures now reflect real browser playing state, and the error-recovery case now ends between checkpoints.

## Initial fix and review

The initial fix adopted 45 in the real media element, full player, mini-player progress and Media Session, and kept 45 after saving. A visible progress-bar seek then saved 30. The complete fresh-vault Obsidian 1.14.4 suite passed 138 tests across 19 files, including restart restore, active playback during sync, quit/unload and popout behavior.

One interrogate pass used claudeAgent/claude-fable-5-1 and codex_pool/gpt-6-astra. GPT identified a native-control seek regression. It was reproduced in the actual Obsidian media element: changing currentTime from 30 to 42 emitted seeking, timeupdate and seeked. The initial fix showed 42 but left disk at 30; current master saved 42. This tests real browser media events on desktop. It does not claim to test iOS fullscreen controls.

The comment review removed four redundant comments, kept the jsdom playback limitation, and passed after one bounded correction of unsupported structural flags. No suppression or unencoded constraint remains in the scoped additions.

## Write-count baseline

`count-session.py master-current` uses `npm run count:writes` on the exact PodNotes data file: two seconds idle, sixteen seconds playback, pause, then a visible seek and flush. Counts were 0 while idle, 2 after playback/pause and 3 after seek/flush. All writes completed, none failed, none remained pending, and the recorder confirmed it was intact and the plugin unchanged.

## Documentation handoff

Held PR #458 must remove the stopped-time note and manual progress-bar correction from sync.md's device handoff instructions. Describe clean paused adoption and the priority of active playback or an unsaved local seek. Keep selecting the desired episode when another episode is loaded, and keep waiting for sync. Replace the claim that every episode played since startup retains local position. In mobile.md's final Limits on phones bullet, remove noting and manually checking the stopped time while retaining pause and sync instructions.

## Scope and follow-ups

The fix uses the existing three-way merge. It adds no schema, settings, public API or URI change, and requires no migration. Existing limitations remain: syncing another selected episode does not replace loaded media; resuming an ended episode starts at zero; value-based merging cannot express the causal priority of a same-value seek before the external callback. The local-file full-player artwork placeholder is visually poor and belongs to the separately planned UI2 work.

## Final local checks

Commit 35d08de corrects native seeks without adding state: a stopped element that differs from the snapshot marks local movement and updates the snapshot baseline. The check runs before restore, timeupdate and persistence. Thirteen new regressions failed against the initial fix. Two saved-seek tests also caught a first attempt that did not update the baseline. They now pass.

After removing three new narrative comments, the parent ran bash .obsidian-e2e-artifacts/play2/verify.sh on Node v22.23.2:

| Command | Result |
| --- | --- |
| npm run lint | PASS |
| npm run format:check | PASS |
| npm run typecheck | PASS |
| npm run build | PASS |
| npm run test | PASS: 2,424 tests in 122 files; Svelte checks pass |
| npm run test:tooling | PASS: 13 tests in 2 files |
| npm run docs:build | PASS: 18 pages, internal links and 112 legacy fragments |
| npm run size | PASS: 311,744 raw bytes, 98,324 gzip bytes at level 9 |

Master 427c9d5 built with the same locked dependencies measured 311,379 raw bytes and 98,239 gzip bytes. The final change adds 365 raw bytes and 85 gzip bytes, leaving 1,676 gzip bytes below the limit. This is a bundle-size comparison, not a performance claim.

## Workflow notes

A native-fix delegate ran one standalone browser probe before the parent instructed it to use only the project Obsidian workflow. It stopped, removed its scratch directory and made no iOS claim. That probe is not acceptance evidence. The final verifier uses real Obsidian exclusively. A draft native-seek-after-end test also had its expectation corrected from finished/60 to the established 20/unfinished behavior.

## Final runtime result

Independent verifier: claudeAgent/claude-opus-5-5, task play2-final-runtime-verification-1. PASS on head 35d08de1e4f6b5e33a39195b586fa32d284d831a. No tracked file changed during verification.

| Run | Result |
| --- | --- |
| python3 .obsidian-e2e-artifacts/play2/live-handoff.py final | Sync 45 adopted in player, media, memory and disk; save/flush retained 45; visible seek saved 30 |
| python3 .obsidian-e2e-artifacts/play2/native-seek.py final | Real native events at 42; save/flush retained 42 |
| python3 .obsidian-e2e-artifacts/play2/immediate-seek-probe.py final | Synchronous raw seek and save recorded 37 before any event; later sync 50 adopted; raw seek 33 before sync 55 retained local priority and saved 33 |
| python3 .obsidian-e2e-artifacts/play2/count-session.py final | Same 0/2/3 writes as master; intact recorder, no failed or pending writes |
| TMPDIR=/data/home/tmp/podnotes-PLAY2 fnm exec --using 22 -- npm run test:e2e | Fresh vault, Obsidian 1.14.4: 139/139 tests in 19 files, exit 0 |
| TMPDIR=/data/home/tmp/podnotes-PLAY2 PODNOTES_OBSIDIAN_APP=/opt/Obsidian/obsidian fnm exec --using 22 -- npm run test:e2e | Fresh vault, Obsidian 1.13.7: 139/139 tests in 19 files, exit 0 |

The parent inspected the JSON readbacks, both final suite logs and all eight final screenshots. Full-player time and progress and mini-player progress match the saved positions. No runtime errors were captured. The final bundle SHA-256 is 9ae57b187587341f6998458d8c83d4cedb932a3eabb481395830645c74982459.

The immediate probe used real native events. Its external data.json write and seek were ordered inside one renderer eval so the file watcher could not apply the sync before the intended seek. This tests the real external-change handler and media element on Linux desktop. Neither iOS nor Android was tested.

The verifier stopped the write recorder before stopping Obsidian. The parent independently reran npm run stop:e2e-obsidian -- --dry-run and saw pids: none. The worktree is clean; the final fetch still points origin/master at 427c9d5. The completed verifier has no pending child runs and was cancelled to stop further delivery.

The minute rounding in the mini player shows 1 min left with 10 to 15 seconds remaining. This predates the fix and is a UI2 consideration, alongside the artwork placeholder. Both observations are outside this playback-persistence change.

Fix Root Causes kept the change at the point that produced stale progress. Foundational Thinking reused playing status, the existing moved flag and the player snapshot instead of adding a second merge path. Prove It Works required before/after persisted values from the same real application and an independent final runtime pass.