# WO-05: Verify the integrated repair and deliver scoped commits

Status: complete — verified offline and delivered to `origin/main`

## Acceptance

- [x] Fix the three baseline ledger-enumeration test failures by reading ledger
  record files, without weakening delivery assertions.
- [x] Run affected orchestrator, browser, desktop, Telegram progress, panel,
  ownership, final delivery, and voice regression tests using mocked services.
- [x] Build affected workspace packages and complete integration checks.
- [x] Review the integrated changes against WO-01–04 and resolve relevant gaps.
- [x] Commit only owned changes, fetch the tracked remote, and push verified
  commits to `origin/main` without force or modifying the dirty original checkout.
- [x] Record commit identifiers, commands/results, and any runtime validation
  limitation; keep installation/publication distinct from repository delivery.

## Implementation and verification

Integration checkout: `/tmp/omnius-progress-browser-20261011`, branch
`repair/telegram-progress-browser-20261011`, tracking `origin/main`. Each feature
was reviewed and cherry-picked from its scoped checkout. The original dirty
checkout was neither reset nor modified. Native SQLite was built only in the
isolated dependency installation to run the native-runner fixtures.

### Passing verification

The table counts distinct passing cases once. The first web/voice command also
ran three older visibility cases; those are replaced by the final expanded
15-case visibility suite and are not counted twice.

| Area | Checks | Evidence |
| --- | ---: | --- |
| Orchestrator telemetry, visual evidence, mixed progress/final separation, desktop discovery, observability | 55 | `/tmp/omnius-orchestrator-integration.log` |
| Exact tool discovery and schema contracts | 19 | `/tmp/omnius-discovery-regression.log` |
| Browser downloads, DOM targeting, cancellation, desktop input | 76 | `/tmp/omnius-browser-desktop-final.log` |
| Telegram narration, panels, public activity, native runners/children, steering, typing ownership | 169 | `/tmp/omnius-narration-final-tests.log`; exact ten suites in WO-02 |
| Integrated Telegram native closeout, diagnostic archive/views, desktop capability boundaries | 35 | `/tmp/omnius-telegram-integration-final.log` |
| Daemon split-stream/failure/Stop transport, web client/script, TUI voice | 91 | `/tmp/omnius-web-voice-regression.log` |
| **Distinct selected passing checks** | **445** | Does not claim the entire repository suite is green |

From each package directory, selected tests were run with
`pnpm exec vitest run <files> --maxWorkers=1 --minWorkers=1`.

Orchestrator files:

```text
activity-diagnostic-visibility.test.ts
telegram-progress-accounting.test.ts
completion-provenance-visual-evidence.test.ts
model-resolution-telemetry.test.ts
tool-search-desktop.test.ts
run-observability.test.ts
tool-search-exact-resolution.test.ts
tool-search-schema-fields.test.ts
tool-search-schema-budget.test.ts
```

Execution files:

```text
playwright-browser-download.test.ts
playwright-browser-cancellation.test.ts
playwright-browser-contract.test.ts
playwright-browser-observation.test.ts
dom-summary.test.ts
desktop-click.test.ts
vision-action-loop.test.ts
```

CLI files beyond the ten listed in WO-02:

```text
telegram-desktop-tool-discovery.test.ts
activity-diagnostic-visibility.test.ts
telegram-private-native-closeout.test.ts
web-chat-transport-lifecycle.test.ts
web-ui-client-runtime.test.ts
web-ui-script.test.ts
voice-narration.test.ts
```

`pnpm -r build` passed for all 11 workspace projects selected by the recursive
build command (12 projects including the repository root). Log:
`/tmp/omnius-progress-browser-final-build.log`. CLI `pnpm exec tsc --noEmit`
and scoped `git diff --check` also passed.

The final integration review repaired the raw Telegram diagnostic bypasses,
mixed commentary/final classification, queued continuation writes, current-input
event attribution, explicit no-click keyboard input, workspace-owned downloads,
and bounded native cleanup. No open implementation item remains in WO-01–04.

### Reproduced baseline failures

The assessment's three `EISDIR` failures are fixed; delivery assertions remain
unchanged. Expanded suites also exposed these failures before the relevant
repair, which are recorded rather than silently folded into this work:

| Suite | Baseline result | Reproduction |
| --- | --- | --- |
| `tool-search-preparatory-progress.test.ts` | 5 pass, 3 fail | Same three renewal/already-inline preparatory-credit assertions with runner restored to pre-desktop `fedbcff7`; no receipt-accounting change in this repair |
| `telegram-intake-presentation.test.ts` | 8 fail | Unchanged pre-WO-02 CLI: one source-custody rejection and seven legacy routing fixture timeouts; `/tmp/omnius-intake-baseline-tests.log` |
| `telegram-typing-action-handoff.test.ts` | 13 pass, 2 fail, one unhandled rejection | Same failures at pre-WO-02 integration `70efc6b2`: obsolete intake-schema expectations and missing public attention grant; `/tmp/omnius-typing-handoff-baseline-tests.log` |

WO-02 lists the exact intake/typing case names and errors. These three suites
remain separate follow-up work; they are not a regression attributed to this
repair, nor are they included in the passing total above.

### Delivery

| Commit on `origin/main` | Work |
| --- | --- |
| `ddaaf332` | Durable scope and five work orders |
| `9dc6c04f` | Diagnostic classification and truthful browser evaluation effects |
| `2f8cd60f` | Preserve focused native filename input |
| `d746c515` | Keep mixed progress/work narration out of final candidates |
| `dcccad52` | Live DOM controls and exact scoped targets |
| `ad8cf340` | Scoped desktop discovery and capability term coverage |
| `555c4270` | Native dialog descriptions and real Telegram discovery boundaries |
| `3b08d86d` | Failure-lifecycle bookkeeping telemetry |
| `70efc6b2` | Verified native browser download custody and cleanup |
| `346bbbea` | Telegram narration, coalescing, continuation ownership and raw diagnostic filters |

The tracked remote was fetched before pushes; no force push or shared-history
rewrite was used. This record closes with a separate documentation commit after
verification. Installation/publication and running-session rollout were not
performed. No live inference, GPU workload, active browser/desktop interaction,
or Telegram send was used in validation. Live model compliance with the progress
instruction and provider behavior remain rollout checks, not claimed results.
