# Runtime Health Remediation Master Tracker

**Authority:** canonical granular tracker for RHR-2026-09-02
**Checked-item rule:** code, focused tests, and named evidence must all exist
**Last reconciled:** 2026-09-27

## October 1 Telegram delivery-truth follow-up

- [ ] [WO-57: Telegram delivery truth, reminder authority, and bounded recovery](WO-57-telegram-delivery-truth-and-recovery.md): require transport-backed visible-response evidence, receipt-backed reminder mutations, non-circular inline research delivery, atomic steering, wrapped WebFetch receipt parsing, and evidence-preserving bounded recovery.

## September 27 evidence-driven objective loop and TUI responsiveness

- [ ] [WO-56: Evidence-driven objective loop and nonblocking execution](WO-56-evidence-driven-objective-loop.md): reconcile every tool result against an immutable request-specific objective, prevent duplicate successful external effects, distinguish verifier failure from artifact failure, suspend and resume natural clarification durably, preserve verified Telegram delivery, and remove blocking privilege execution from the TUI event loop.

## September 26 Case Theory pressure recovery

- [x] [WO-55: Telegram pressure recovery without silent terminal states](WO-55-telegram-pressure-recovery.md): generation-fenced recovery for silent inference, stale active-runner decisions, terminal review, continuation, and exact Telegram delivery receipts.

## September 26 Case Theory router deferral recovery

- [x] [WO-54: Case Theory router deferral recovery](WO-54-case-theory-router-deferral-recovery.md): exact task-identity retirement, sequential mixed-domain repair, one scheduler-owned recovery with continuous typing and truthful exhaustion, plus canonical receipt-backed URLs with inference review and final fail-closed withholding. Implemented in `d6a6cb6e`; 251 focused Telegram tests and CLI build passed.

## September 5 runtime 1.0.701 rapid churn repair

- [x] [WO-53: deferred reflection churn](WO-53-deferred-reflection-churn.md): persistent scoped revision receipts prevent repeated unchanged reflection appends; seven regressions and workspace build passed. [Issue #3](https://github.com/robit-man/open-agents/issues/3) retains the separate context-recovery investigation and corrected budget findings. Existing backlog and live runtime are unchanged.

## September 5 runtime 1.0.700 large attachment follow-up

- [x] [WO-51: large Telegram attachments](WO-51-large-telegram-attachments.md): fixed inbound/cache byte caps removed; streamed download, Local Bot API transport, exact archive aliases and responsive durable Stop verified. Archive implementation `dec18c74` and CLI integration `40a76cea` delivered to `origin/main`.
- [x] [WO-52: steering projection preservation](WO-52-steering-projection-preservation.md): `766b72f3` delivered to `origin/main`; 118 regression tests passed, typecheck and final orchestrator build passed. Current instructions survive conversation/task projection and visibility receipts match actual admitted request content.

The reviewed run rejected a 37.8 MB archive and then looped because the compiler removed its required reconciliation instructions. Ollama was answering requests; no ongoing backend outage was established in the latest journal window. Final verification: clean workspace build and final rebuild passed, execution 1,852 passed / 3 existing skips, CLI full run 2,656 passed plus the final guest-isolation case passed in the focused run (2,657 distinct CLI cases), orchestrator 118 focused regressions passed. Telegram's hosted 20 MB download limit remains external; live larger-file reception requires a configured Local Bot API server and shared directory, documented in [the setup guide](../../telegram-large-files.md). User publication and live acceptance remain pending.

## September 5 runtime 1.0.698 delivery and Ollama follow-up

- [x] [WO-49: authored final delivery](WO-49-authored-final-delivery.md): later authored replies survive earlier prose, tool work and summary-only completion. Commit `bfaabd7f` delivered to `origin/main`; 280 orchestrator and 58 CLI checks passed.
- [x] [WO-50: embedding unavailability](WO-50-embedding-unavailability.md): commit `a3375c63` delivered to `origin/main`; bounded missing-model recovery, truthful vector/job outcomes and exact request-model identity verified. User-requested Nomic re-pull succeeded; passive journal follow-up recorded 51 embedding HTTP 200 responses and no embedding 404s in the captured post-pull window. Clean all-workspace build passed, memory 665/665 and indexer 186/186 passed, twelve additional dependent CLI cases and all seventeen daemon transport cases passed. The three real Telegram answer-delivery cases also passed after the clean rebuild. Publication and live acceptance of source changes remain with the user; existing-data/model-migration limits are explicit in WO-50.

The separate 14:43 PDT router failure was a managed Ollama lane exiting mid-stream, followed by connection refused and recovery. The external broker does not retain its exit status/stderr, so the underlying exit cause remains unresolved; WO-50 records the evidence and external observability follow-up.

## September 5 post-publication field review

[Runtime 1.0.697 field review](LIVE-TELEGRAM-LOG-REVIEW-2026-09-05.md): a useful assessment final was delivered at 13:41 PDT, supported by TypeScript and 21 passing tests. At the reviewed checkpoint, the next implementation run was active. No hard blocker was established; source-read false failures, malformed diagnostic input handling and conversation-mode report coverage triggered the repairs below. Native feature workflow selection and continuous typing are not established by this sample.

The user authorized end-to-end repair of the remaining defects; repository work is now complete:

- [x] [WO-46: shell result authority](WO-46-shell-result-authority.md).
- [x] [WO-47: diagnostic execution contract](WO-47-diagnostic-execution-contract.md).
- [x] [WO-48: conversation result evidence](WO-48-conversation-result-evidence.md).

Repairs `a3c26eca` (shell authority), `d4c37489` (validated asynchronous diagnostics), and `672bef53` (durable conversation evidence and terminal reporting) are delivered to `origin/main`. Clean workspace build passed. Execution: 1,820 passed / 3 existing skips; CLI: 2,575 passed. Orchestrator: 2,847 passed / 1 existing skip in the full run, with its sole outdated receipt fixture corrected and all 10 tests in that suite passing on rerun. All 7,243 distinct non-skipped affected-package tests have a passing result on the final implementation; the workorders retain exact commands, initial failures, limits and evidence paths. Publication and live acceptance remain with the user.


## September 5 terminal result and long-horizon follow-up

Current terminal repair authority: [WO-44](WO-44-evidence-backed-terminal-results.md).

- [x] Evidence-backed terminal report survives automatic completion and reaches Telegram.
- [x] Genuine final answers retain source, task and receipt ownership.
- [x] AIWG research-team mechanisms mapped into existing execution and recovery primitives.
- [x] Native long-horizon feature implementation and 12-scenario acceptance: [workflow WO-01](../long-horizon-feature-delivery/WO-01-native-research-feature-workflow.md).
- [x] Native Ollama tool request and response parity: [WO-45](WO-45-native-ollama-tool-contract.md).
- [x] WO-44 regression verification, review and scoped Git delivery recorded.

WO-44 repository acceptance: repair 691532df delivered to origin/main; orchestrator 2,650 passed / 1 skipped across 218 suites; CLI 2,550 passed across 269 suites; both package builds passed.

Native feature workflow WO-01 and native Ollama transport WO-45 are complete in source commits 3c71f668 and 26d4cbed, delivered to origin/main. Final acceptance: 7,310 passed / 4 skipped across 671 suites in schemas, execution, orchestrator and CLI; clean workspace build and final workspace rebuild passed. The workflow workorder maps all twelve required scenarios to concrete assertions, including actual parent compaction in both loops and child Stop propagation. Publication and live acceptance remain with the user.

## September 5 post-publication Telegram progress follow-up

Current authority: [WO-43](WO-43-telegram-router-progress-boundary.md).

- [x] Remove router failure prose from request-comprehension progress.
- [x] Repair confirmed routing contract/correlation defects.
- [x] Verify typing and delivery regressions, record review, build and scoped delivery.

WO-43 repository acceptance: repair 3f3c37f1 delivered to origin/main; all 2,531 CLI tests passed across 268 suites; clean CLI rebuild passed. Publication and live acceptance remain with the user.

## September 5 tool-quality and live-behavior follow-up

Current follow-up authority: [TOOL-QUALITY-2026-09-05.md](TOOL-QUALITY-2026-09-05.md).

- [x] WO-30 structured tool invocation and prose safety.
- [x] WO-31 complete runtime policy delivery.
- [x] WO-32 evidence-dependent scope reconciliation.
- [x] Tool-family audit findings reproduced, repaired, and delivered.
- [x] Aggregate regression/build verification and publication handoff.
- [x] WO-33: File mutation transactions
- [x] WO-34: Shell authority and process results
- [x] WO-35: Search and exploration isolation
- [x] WO-36: Media evidence integrity
- [x] WO-37: Browser and process lifecycle
- [x] WO-38: Tool contract preservation
- [x] WO-39: Telegram working indicator
- [x] WO-40: Web content and crawl receipts
- [x] WO-41: Verification audit and typed completion consistency
- [x] WO-42: Media execution configuration and cancellation

Repository acceptance: all 13 follow-up orders closed; 6,885 tests passed, 4 skipped across 641 suites; clean rebuild passed for all 11 workspace packages. Source head 7005aa41 and scoped repair commits are recorded in the program ledger. Publication and subsequent live Telegram acceptance remain with the user.

## September 4 field-review follow-up: WO-25 through WO-29

The active Telegram field review found new deterministic defects after the
earlier acceptance runs. Prior checked sections remain historical evidence;
they do not close these follow-up repairs. Current authority is
[TELEGRAM-LONGHAUL-2026-09-04.md](TELEGRAM-LONGHAUL-2026-09-04.md).

- [x] WO-25 memory compiler eligibility and outcome audit.
- [x] WO-26 admin Telegram media alias resolution.
- [x] WO-27 task verification authority separated from mutation observations.
- [x] WO-28 obsolete task evidence retirement after explicit replacement.
- [x] WO-29 retired failure diagnostics and active projections.
- [x] Aggregate tests, clean build, independent review, and scoped Git delivery.
- Publication and live validation are reserved to the user after handoff.

## Phase 0: durable program boundary

- [x] Create the program directory.
- [x] Record the observed production baseline.
- [x] Record cross-cutting architectural decisions.
- [x] Define eight dependency-ordered work orders.
- [x] Define program invariants.
- [x] Define scoped commit and push policy.
- [x] Define live-inference hardware gate.
- [x] Confirm every work-order link resolves.
- [x] Confirm every observed failure has a traceability row.
- [x] Record the starting Git state for both repositories in `AUDIT.md`.

## WO-00: hermetic test boundary

- [x] Inventory all production constructors that can open an inference socket.
- [x] Add an injected inference/backend dependency to `EmotionEngine`.
- [x] Add an injected pool-status dependency to `EmotionEngine`.
- [x] Add an injected clock to time-sensitive inference coordination.
- [x] Add an injected scheduler/admission dependency to background inference.
- [x] Make the production constructor preserve current runtime behavior.
- [x] Make unit-test defaults fail closed on unmocked network access.
- [x] Add a socket-attempt sentinel for orchestrator tests.
- [x] Add a socket-attempt sentinel for CLI tests.
- [x] Replace model-dependent fixtures with deterministic fake responses.
- [x] Add an explicit opt-in integration-test network boundary.
- [x] Redact secrets and message bodies in captured replay fixtures.
- [x] Reproduce the constrained-pool emotion failure with a fake backend.
- [x] Prove the focused unit suite opens zero sockets.
- [x] Record WO-00 acceptance evidence.

## WO-01: inference pressure scheduler

- [x] Define `foreground`, `interactive-control`, and `background` workload classes.
- [x] Define priority, cancellation, yield, and fairness semantics for each class.
- [x] Implement a process-wide inference pressure coordinator.
- [x] Register the main interactive task with the coordinator.
- [x] Register Telegram reply generation with the coordinator.
- [x] Register router and repair inference with the coordinator.
- [x] Register emotion labeling and reflection with the coordinator.
- [x] Register Telegram daydream and reflection sweeps with the coordinator.
- [x] Register embedding-triggered background work with the coordinator.
- [x] Gate background work before corpus/embedding construction.
- [x] Recheck pressure immediately before backend admission.
- [x] Cancel or yield running background work when foreground arrives.
- [x] Enforce one background session per scheduler tick.
- [x] Implement round-robin fairness among eligible background sessions.
- [x] Stop a background sweep after its first timeout or capacity refusal.
- [x] Reuse `TelegramCognitionQueue` rather than a second unbounded queue.
- [x] Add pressure-state telemetry without message bodies.
- [x] Add deterministic foreground-preemption tests.
- [x] Add idle-fairness tests.
- [x] Add background-capacity-stop tests.
- [x] Record WO-01 scoped acceptance evidence.

## WO-02: resumable broker admission

- [x] Remove the universal five-second broker admission header.
- [x] Add request-aware admission options to backend request construction.
- [x] Add a stable logical request ID across retry attempts.
- [x] Include workload class in broker admission metadata.
- [x] Preserve response headers in `InferenceHttpError`.
- [x] Parse retry-after, retryability, and structured broker reason codes.
- [x] Classify capacity wait, reclaimable placement, permanent placement, and liveness faults.
- [x] Keep retryable foreground work alive and cancellable.
- [x] Prevent background work from retrying broker capacity failures.
- [x] Bound interactive-control retries independently from foreground work.
- [x] Coalesce retry diagnostics under one logical request.
- [x] Avoid rewriting the complete prompt body on every queue wait.
- [x] Stop permanent placement failures without a retry loop.
- [x] Implement broker FIFO retention or a resumable admission ticket.
- [x] Return explicit `retryable` and `reason_code` values from the broker.
- [x] Return queue position and retry timing without exposing request content.
- [x] Add cancellation cleanup for retained waiters/tickets.
- [x] Add stale-ticket expiry and idempotent resume behavior.
- [x] Add single-logical-request retry tests in Omnius.
- [x] Add FIFO/resume/cancel tests in ollama-unify.
- [x] Add permanent-refusal no-loop tests.
- [x] Add retryable-failure no-hard-exit tests.
- [x] Record WO-02 acceptance evidence in `WO-02-BROKER-PROTOCOL-EVIDENCE.md`.

## WO-03: workboard and failure lifecycle V2

- [x] Define typed evidence predicate variants.
- [x] Preserve prose labels as presentation only.
- [x] Migrate generated workboard requirements to typed predicates.
- [x] Match evidence by predicate fields, target, revision, and receipt.
- [x] Add exact target/path/range matching.
- [x] Add exact verifier-command/result matching.
- [x] Reject failure observations as verification evidence.
- [x] Reject generic shell calls as verification evidence.
- [x] Aggregate identical diagnostics by stable fingerprint.
- [x] Bound retained diagnostic exemplars.
- [x] Track diagnostic first-seen, last-seen, and count.
- [x] Define one active frontier per assignee/lease.
- [x] Activate dependent cards only after accepted completion.
- [x] Prevent phase advancement after rejected completion.
- [x] Reconcile the entire todo snapshot, not only new items.
- [x] Supersede removed or completed mirrored cards.
- [x] Prevent completed cards from reopening without a new epoch.
- [x] Add target and epoch to failure-card identity.
- [x] Resolve matching failures in memory and workboard atomically.
- [x] Emit a durable resolution receipt.
- [x] Correct `lastSubstantiveProgress` to use a turn identifier.
- [x] Add legacy-read/additive-write migration support.
- [x] Add evidence-present acceptance tests.
- [x] Add diagnostic-cardinality tests.
- [x] Add todo-removal and no-reopen tests.
- [x] Add failure-success-resolution tests.
- [x] Record WO-03 acceptance evidence.

## WO-04: canonical context projection

- [x] Define a side-effect-free model-facing request intermediate.
- [x] Normalize model-visible history on every turn.
- [x] Keep semantic compaction dependent on real pressure.
- [x] Project one canonical active task frame.
- [x] Remove duplicate mutable startup state from model visibility.
- [x] Select either raw tool transaction or compact evidence receipt, not both.
- [x] Preserve current user authority, open requirements, active action, and verifier state.
- [x] Preserve exact source evidence needed by the active action.
- [x] Use typed relevance and dependency closure, not keyword stripping.
- [x] Apply the same byte cap during selection and transport.
- [x] Recompute the trace after final transport serialization.
- [x] Report exact dropped/truncated signals from the sent request.
- [x] Make focus directives visible to every model tier they constrain.
- [x] Define a minimal always-inline tool set.
- [x] Move `adjudicate` to deferred discovery unless a capability contract selects it.
- [x] Promote tools from explicit action/capability contracts.
- [x] Add per-turn promotion TTL.
- [x] Demote tools after TTL expiry or action completion.
- [x] Rebuild tool definitions when TTL/state changes.
- [x] Enforce explicit schema byte/token budgets.
- [x] Add duplicate-projection replay tests.
- [x] Add selection/transport parity tests.
- [x] Add tool-promotion TTL tests.
- [x] Meet the replay target of at most 40,000 input tokens.
- [x] Meet the replay target of at most 8,000 schema characters.
- [x] Record WO-04 acceptance evidence in `WO-04-CANONICAL-PROJECTION-EVIDENCE.md`.

## WO-05: completion readiness authority

- [x] Define `CompletionReadinessSnapshot`.
- [x] Join expected outcome, todo revision, workboard revision, evidence ledger, and delivery state.
- [x] Include the latest target-specific verifier receipt.
- [x] List unresolved authoritative card IDs.
- [x] List unmet expected effects.
- [x] List unverified completion claims.
- [x] Compute a deterministic readiness hash.
- [x] Evaluate readiness in the host before accepting `task_complete`.
- [x] Keep model closeout guidance advisory.
- [x] Allow explicit override only with a persisted reason.
- [x] Prevent workboard finalization while authoritative cards remain active.
- [x] Expand terminal receipts with board hash and todo revision.
- [x] Expand terminal receipts with unresolved IDs and evidence mapping.
- [x] Expand terminal receipts with delivery and verifier receipts.
- [x] Add false-positive shell verification tests.
- [x] Add unresolved-card completion rejection tests.
- [x] Add verified-completion acceptance tests.
- [x] Add explicit-override audit tests.
- [x] Record WO-05 acceptance evidence in `H6-MAIN-TUI-WO-05-CONFIRMATION-EVIDENCE.md`.

## WO-06: durable Telegram intake and health

- [x] Define a persistent poll-health record.
- [x] Track last poll start and completion.
- [x] Track last successful response and last received update.
- [x] Track last persisted and last dispatched update IDs.
- [x] Track consecutive failures and last error class.
- [x] Track fatal stop reason and current owner identity.
- [x] Expose poll health in `/telegram status`.
- [x] Add an explicit long-poll client deadline.
- [x] Surface nonfatal poll failures in bounded telemetry.
- [x] Surface SQLite initialization and write failures.
- [x] Create a transactional update spool keyed by update ID.
- [x] Persist the raw normalized envelope before offset advance.
- [x] Advance only the highest contiguous durable checkpoint.
- [x] Mark dispatch start, success, retryable failure, and terminal failure.
- [x] Replay nonterminal spool entries after restart.
- [x] Make replay idempotent across process crashes.
- [x] Persist all update kinds, including reactions and callbacks.
- [x] Integrate input-ownership records with spool identity.
- [x] Await bridge close and database flush on every exit path.
- [x] Restart the bridge after update/resume when configured.
- [x] Add shadow dual-write and rollback flags.
- [x] Add empty-success versus failing-poll status tests.
- [x] Add crash-before-offset replay tests.
- [x] Add handler-failure replay tests.
- [x] Add reaction/callback persistence tests.
- [x] Add signal/readline shutdown tests.
- [x] Record WO-06 acceptance evidence.

## WO-07: integration, rollout, and canaries

- [x] Run all WO-00 hermetic unit suites.
- [x] Run scheduler preemption and fairness integration tests.
- [x] Run Omnius broker admission simulations.
- [x] Run ollama-unify FIFO/resume/cancel simulations.
- [x] Run workboard production-sequence replays.
- [x] Run failure-resolution replays.
- [x] Run context-budget real-trace replays.
- [x] Run completion-readiness integration tests.
- [x] Run Telegram crash/restart/exact-once tests.
- [x] Run CLI, orchestrator, execution, and memory typechecks.
- [x] Run all workspace builds.
- [x] Capture pre-change broker service health.
- [x] Verify CUDA negotiation policy before service-affecting operations.
- [x] Commit the verified broker repair in its own repository.
- [x] Fetch and push broker `main` without history rewrite.
- [x] Restart the broker through its documented service path.
- [x] Capture post-restart broker health and queue metrics.
- [x] Confirm the active Ollama service is reachable through the broker.
- [x] Enable each Omnius feature in an isolated deterministic shadow replay.
- [x] Compare old and new projections on body-free keyed-HMAC traces.
- [x] Promote one subsystem at a time in the deterministic rollout fixture.
- [x] Exercise each supported rollback switch and legacy-store boundary.
- [x] Confirm endpoint `http://127.0.0.1:11434` and exact available model tag
  `robit/ornith-1.5:9b` without generating tokens.
- [ ] Inspect actual runner GPU placement with `nvidia-smi`.
- [x] Confirm no runner for `robit/ornith-1.5:9b` is currently resident on an
  unintended GPU. Recheck after the permitted zero-token load probe.
- [ ] Run a token-generating live canary only after the full preflight.
- [x] Record deterministic latency, queue, context, completion, and Telegram
  health results.
- [x] Commit only task-owned Omnius files.
- [x] Fetch and push Omnius `main` without history rewrite.
- [x] Reconcile every tracker item with evidence or an explicit unresolved
  reason.

Operational blockers are exact. No user-designated physical A100 UUID was
provided for the live canary, so no zero-token load probe or token-generating
request was sent. The installed package in `telegram_test` also remains an
operator publication boundary. Physical macOS, Windows, and AGX Orin devices
are not attached to this workspace. These boundaries do not reopen a
deterministically repaired P0 or P1 defect.

## WO-19: clean build and optional-dependency reproducibility

- [x] Remove the schemas package's undeclared global clone dependency.
- [x] Prove operational-world migration clones do not alias legacy input.
- [x] Make Helia an explicit optional capability at the execution boundary.
- [x] Return a typed unavailable result when optional Helia support is absent.
- [x] Close the identity-kernel optional-string type hole.
- [x] Declare or correctly route the CLI menus' color dependency.
- [x] Run focused migration, identity-kernel, and TUI menu tests.
- [x] Run affected package typechecks and builds from declared dependencies.
- [x] Prove an isolated clean checkout does not depend on protected generated or
  publish-tree changes.
- [x] Record WO-19 acceptance evidence in `WO-19-CLEAN-BUILD-EVIDENCE.md`.

## WO-23: long-horizon control-plane reconciliation

- [x] Reconcile typed failure outcomes across retry, workboard, and durable
  failure memory.
- [x] Project all unresolved todo leaves and active failure obligations into
  integration closure.
- [x] Keep lexical discovery candidates advisory until evidence adopts them.
- [x] Compact oversized raw discovery independently of total provider context
  capacity and restore validated canonical receipt handles.
- [x] Activate nested Git tracking only from explicit in-workspace target paths
  and reject ambiguous or escaped repositories.
- [x] Consolidate Telegram `edited_message` revisions before UI output,
  cognition, attention inference, or queue admission.
- [x] Pass deterministic regression tests, affected typechecks, and the
  workspace build.
- [x] Commit and push the scoped repair to `main`.

## WO-24: Telegram logical request variant identity

- [x] Correlate empty Telegram streams with broker conflict receipts.
- [x] Confirm the broker correctly binds one logical ID to one exact wire
  contract.
- [x] Give stream-to-unary compatibility fallback a distinct stable ID.
- [x] Give `/v1`-to-native empty recovery a distinct stable ID.
- [x] Give changed `/v1` recovery bodies a distinct stable ID.
- [x] Preserve the original ID across byte-identical queue retries.
- [x] Route Ollama follow-up decisions through native chat with `think:false`.
- [x] Raise the follow-up contract budget from 300 to 1,024 tokens.
- [x] Add a broker-faithful 409 conflict regression harness.
- [x] Pass backend, Telegram transport, observability, and Bot API suites.
- [x] Pass affected package typechecks and the workspace build.
- [ ] Run a token-generating live canary only after the mandatory accelerator
  preflight identifies the requested endpoint, model, and GPU placement.

## Program completion gate

- [x] All repository-controlled deterministic WO acceptance sections pass.
- [ ] All physical-device, installed-package operator, and live-inference WO
  acceptance sections pass.
- [x] No open deterministic P0 or P1 regression remains after independent
  adversarial review and the aggregate suites.
- [x] Broker timeout rate is below 1% in the controlled load harness.
- [x] Background inference admissions are zero while foreground work is active.
- [x] Context replay meets both token and schema budgets without authority loss.
- [x] Workboard diagnostic cardinality remains bounded under repeated failures.
- [x] Telegram crash replay loses no update and duplicates no terminal dispatch.
- [x] Completion cannot pass without exact required evidence.
- [x] Both repositories are clean for task-owned files after scoped delivery.

## Extended work-order tracking

- [x] WO-08 Telegram identity, conversation, and reply semantics passes all
  typed relationship and delivery lifecycle replays.
- [x] WO-09 Telegram admission, API governance, and moderation passes burst,
  permission, approval, and rollback tests.
- [x] WO-10 Telegram observability and native presentation passes golden and
  transport capability tests.
- [x] WO-11 safe web, document, image, and audio capability surface passes
  sandbox, transform, provenance, and delivery tests.
  - [x] Public Telegram tool surface excludes raw shell and arbitrary download.
  - [x] Public page fetch rejects private, local, metadata, and credentialed URLs.
  - [x] Search results carry an explicit unverified-link trust state.
  - [x] Successful page fetches carry final-URL, status, size, and SHA-256 receipts.
  - [x] Acquired media retains source and redirect provenance through delivery.
  - [x] Media acquisition requires an exact candidate from a prior bounded page discovery.
  - [x] Scoped artifacts reject symlink-backed and over-limit files before read.
  - [x] Fixed audio transforms remove source metadata and chapters.
  - [x] Focused deterministic component and delivery suites pass.
  - [x] Enforce verified-link evidence against the final Telegram reply text.
  - [x] Reject a Telegram send kind that conflicts with capability MIME/media kind.
  - [x] Wire created-PDF page rendering into the scoped visual review loop.
  - [x] Move persistent chat artifacts into explicit per-run directories with cleanup.
  - [x] Add local clickable-file presentation coverage.
  - [x] Add deterministic delayed-body/timeout acquisition coverage.
- [x] WO-12 person reminders and self-action schedules pass restart,
  idempotency, delivery, and preemption tests.
- [x] WO-13 ontology work and relationship graph passes scale, lifecycle,
  verifier, and conversation-timeline tests.
- [x] WO-14 adjudication passes planner fallback, full sub-agent, evidence,
  dissent, persistence, and replay tests.
- [x] WO-15 prompt and context policy passes STE, conversational tone,
  deduplication, anti-bleed, and model-tier evaluations.
- [x] WO-16 deterministic terminal, browser-soak, and ARM64 installer suites
  pass.
- [ ] WO-16 physical macOS, Windows, and AGX Orin operator canaries pass.
- [x] WO-17 daemon frontend and discovery surfaces pass endpoint, responsive,
  accessibility, and mocked inference end-to-end tests.
- [x] WO-18 pause, stop, and resume use one durable interruption lifecycle and
  pass cancellation, restart, ownership, external-effect, and isolation tests.
- [x] WO-20 exit and restart preserve one exact typed task continuation. Generic
  context restoration cannot mix or activate unrelated historical artifacts.
- [x] WO-21 task convergence and todo scope prevent stale `tasks N/M` displays,
  cross-runner todo leakage, and activity-only re-engagement.
  - [x] Bind todo tools to immutable runner-owned sessions.
  - [x] Clear the active persisted checklist at a fresh task boundary after
    archival.
  - [x] Count only typed authoritative advancement as progress.
  - [x] Latch an advisory convergence review when the leaf frontier is static.
  - [x] Make loop, reminder, budget, and workboard progress semantics truthful.
  - [x] Pass deterministic race, stasis, display, typecheck, and build tests.
- [x] WO-22 autocomplete footer geometry uses the last painted layout as its
  terminal authority and clears all released rows during idle and active-stream
  contraction.
  - [x] Trace the deferred DirectInput repaint and content-write ordering race.
  - [x] Confirm existing tests do not model final terminal cells in this path.
  - [x] Unify idle and streaming geometry transitions around deterministic
    replay.
  - [x] Cover precommitted multi-match to fewer-match and zero-match contraction.
  - [x] Pass focused footer tests, CLI typecheck, and CLI build.
- [x] A recovered pause no longer strands its session. Submitting a new task
  retires the superseded generation instead of refusing every later prompt,
  while unreconciled external effects still fence admission.
