Skip to content

Reconcile managed room-learner stack with current RuView main - #2513

Open
proffesor-for-testing wants to merge 158 commits into
mainfrom
integration/ruv-main-reconcile-20261110
Open

proffesor-for-testing wants to merge 158 commits into
mainfrom
integration/ruv-main-reconcile-20261110

Conversation

@proffesor-for-testing

@proffesor-for-testing proffesor-for-testing commented Nov 10, 2026 •

Copy link
Copy Markdown
Collaborator

CI/review follow-up complete; ready for rUv review. Current head 4f70c5eafe28a17395f1588cc15f5e64ba932d7d against main 094e1400, GitHub merge candidate 994e5896, tree af593d44: all 11 workflows are SUCCESS. Full workspace/downstream checks and all 37 required Release cases passed with zero selected failures/ignores. The synthetic preparation/input-order defect is corrected and the five-second held-publication regression passes. Production capture/clock/buffering and all original guards/caps remain unchanged. Final current-head review: NO BLOCK within code/CI scope; owner review and separate exact-head merge authorization remain. No server PR was merged by this lane.

Current CI: full workflow43279683707, security43279683676, bench43279683922, QEMU/fuzz43279683678. Current-head MQTT43279683661, firmware, Python, forecast and policy workflows have passed. All current-head checks are terminal SUCCESS. Local equivalent focused fixture: 205.18s PASS with five-second delayed preparation, all80 windows, fit/cold/evaluate; 45 parent classifier tests and YAML/Rust syntax checks pass. That local prototype does not substitute for terminal CI on this head.

Reconcile the managed pairing, companion, room-learner, recovery and temporal stack with RuView main while retaining original PR ancestry. PR head:4f70c5eafe28a17395f1588cc15f5e64ba932d7d; target:main. No main merge or release is performed by this update.

Final changes

  • Preserve the managed ruview-unified dependency alongside upstream optional/default-off FANN, Python/witness work, selected-model withholding, recovery and prepared training.
  • Retain reviewed ZIP7 allocation, Wasmtime/zstd and SQLx SQLite repairs without weakening audit policy or source provenance.
  • Repair the five CI bootstrap failures with a digest-pinned non-root Mosquitto sidecar, digest-pinned SDK QEMU/runtime, hosted Clang18 and scoped signed Canonical HTTPS APT sources on Ubuntu24.04. Preserve broker integration, QEMU/fuzz matrix/budgets, workspace, temporal, socket/security/backport, CIR, FANN, auth and bench gates.
  • Reduce AETHER attention/GNN temporary allocations without changing scalar accumulation order, APIs, parameter layout, architecture, gradients, epochs or production time/evaluation budgets. Preserve malformed/missing-value rejection, including zero-width attention. Frozen scalar oracles and perturbation/golden tests protect numerical equivalence.

Prior head68/main094 qualification — historical evidence

GitHub generated merge 9b8f707 has exact parents main094e140 and PR68cd049, tree2b732f00e133526b310d634a792a219aaba4d86d. It was pushed unchanged to feat/2513-main-ci-20261112 solely to qualify the combined tree; no competing PR or history rewrite.

All seven workflow conclusions are SUCCESS:

Workflow Evidence
Full CI 43188188760
Bench regression 43188188802
QEMU/fuzz (all11 jobs) 43188188492
Firmware builds 43188188532
Security scanning 43188188705
Forecast Rust 43188188669
Python package 43188188553

Full Rust workspace and downstream gates pass. Retained temporal receipt: clean worktree,1096 registered tests; 37 explicitly selected Release cases passed with zero selected failures/ignores, including C1/C2/C3/C4, both budget343 paths, C4B and C4C. C4C executed all48 default epochs,395 parameters,38,373 objective evaluations/64 training cohorts under the unchanged60,000ms fit and100,000 evaluation caps, then128 durable windows/held-out/cold-selection checks. Its full lifecycle took487.68s; this is not its fit duration. Release binary SHA256:f2cf611a9808895e9a6680e41d7aec1c8c678532bc9a41cb585b1d48bfb18b79.

On the unchanged PR head itself, supported fresh dispatches also pass MQTT43183390004 and bench43183609967. These are distinct from combined-tree receipts.

Retained failures and limits

  • The original PR-event runs remained queued with zero jobs and inconsistent cancellation/rerun responses; fresh supported dispatch and the combined-tree push restored execution. Their exact backend cause is unproved.
  • Standalone head68 CI43183605678 passed Debug/C1–C3 but hit the existing240s operator-episode watchdog at76/80 durable windows. It did not reach C4C. The same unchanged operator test passes locally and in the successful current-main candidate. Retain this timing/reproducibility risk; not every standalone run is green. No watchdog, acceptance threshold or production fit budget was relaxed.
  • A later checkpoint-encoding prototype preserved bytes but showed no end-to-end improvement (local operator baseline215.08s versus candidate225.14s amid other build activity). It was removed and never pushed. Only the reviewed AETHER optimization is in this PR.
  • Inherited advisory Docker steps still fail on the missing root Dockerfile/image reference despite their successful job conclusion. No image was published; overall CI is not container-release qualification. Performance/API-doc jobs skipped according to existing conditions remain skipped, not executed proof.
  • No optimized app/helper was installed; helper198 remains a separate frozen app-lane artifact. Source/CI results do not establish physical sensing accuracy or customer acceptance.

Review and dependencies

The prior68 qualification below is historical after this test/CI-only head update. The synthetic preparation/input-order defect now has a controlled reproduction and correction; fresh current-head qualification and rUv review remain open. A separate exact-head human merge decision is required. Recheck changed main/head compatibility before relying on this frozen snapshot. #2528 remains a separate draft radar PR; its firmware source and hardware acceptance are not implicitly merged here.

Do not close original PRs on approval alone. Only after the approved candidate actually lands, refresh live heads and verify ancestry/content parity before the historical conditional cleanup. No original PR has been closed by this update.

Complete the prepared local integration of upstream a1ca023 into the companion candidate. Runtime helper provenance remains pinned to the previously verified clean source.
Move the companion pairing design to ADR-419 so upstream ADR-383 remains the HTTP request-target validation decision.
Recheck approved-device capacity under the approval lock before persistence, retaining pending requests when full. Build the exact Host allowlist entry from the prepared identity authority used by the QR, including generated and reloaded identities.

Regressions reproduced the old failures and now pass. Verified actual workspace binary check, nine companion tests and the real binary authority test, plus 28 isolated module tests and independent fix-delta review. No Host wildcard or validation bypass.
Resolve Tokio 1.51.6 with its compatible Mio, socket2 and macro dependencies. Preserve the scoped companion stack and existing audit policy; no upstream vendor or framework adoption. Existing LRU unsoundness and yanked-package warnings remain separately recorded.
Preserve main archive/cache compatibility and companion socket gates. Retain the audited main dependency graph while restoring companion/MAT privacy changes.
@proffesor-for-testing
proffesor-for-testing marked this pull request as ready for review November 10, 2026 14:48
Retain unified and FANN dependencies, reviewed ZIP backports and current SQLite runtime guard. Use byte-verified Wasmtime cache metadata from main without its unconsumed original manifest.
@proffesor-for-testing proffesor-for-testing changed the title Reconcile managed app stack with rUv sensor-fusion main for local baseline Reconcile managed room-learner stack with current RuView main Nov 11, 2026
@proffesor-for-testing

Copy link
Copy Markdown
Collaborator Author

Conditional authored-PR cleanup analysis refreshed against published head b8f731b854cf1760262082992c75b4e9e51d23c9. Authenticated author proffesor-for-testing; all five REST pages checked:48 authored open PRs including2513, eight drafts. Of the47 other PRs:

35 exact heads included: #2501, #2500, #2495, #2487, #2481, #2479, #2475, #2474, #2471, #2470, #2467, #2465, #2461, #2459, #2452, #2450, #2449, #2447, #2446, #2444, #2439, #2438, #2434, #2427, #2424, #2423, #2417, #2407, #2402, #2401, #2386, #2379, #2374, #2350, #2249.

Two superseded, not ancestor/patch-equivalent claims: #1760 is covered by stronger provisioning already present (the same provisioning source exercised by62 mocked controls is retained); #2342's TLS dependency intent is matched/exceeded by the preserved current-main Python lock (quinn-proto0.11.19, rustls0.23.46, rustls-webpki0.103.16, webpki-roots1.0.11). This is source-intent equivalence, not a fresh published-wheel security qualification.

Keep these10: #2499 identity-failure authority repair; #2445 dual native/UART firmware transport; #2498 guarded paired node labels; #2497/#2453/#2458 reference/comparison alternatives; #2496 offline pose-training split repair; #2395/#2400 firmware signing/release; #1145 Docker runtime work. Each still has unique scoped changes; do not wholesale merge them or close as delivered by2513.

Closure conditions: approval alone is not closure. Wait for the approved2513 candidate to land, then re-read every original head/state. A normal merge permits ancestry proof; a squash requires candidate-tree/content parity or explicit reconciliation of intervening changes. The candidate tree here is 866c8605d8b7112d1726b75c480ed2ee75f7ed02. GitHub may not close stacked PRs automatically. Any changed original head must be reclassified. No PR closures/main merge performed by this analysis.

Executed by Codex with a separate read-only inventory reviewer. Repository/API base SHA snapshots lagged live target refs in14 rows; live target refs were verified separately rather than treating those snapshots as today's branches.

@proffesor-for-testing

Copy link
Copy Markdown
Collaborator Author

CI follow-up at candidate dc0661f47b9ff5b537e9d13da7cec9e91dbb0aad:

  • The prior MQTT cancellation was the 20-minute job limit: unrelated long learner library tests prevented the real broker step from starting. The workflow now runs its MQTT namespace (99 passing tests and one existing ignore locally) before the unchanged real Mosquitto integration step. Full workspace/temporal CI remains unchanged. YAML parsing and broker-step preservation pass.
  • The first new MQTT run failed earlier, during apt installation, with Temporary failure resolving 'azure.archive.ubuntu.com'; no unit/broker test result was produced. Retried failed run 43039199137 once.
  • Fuzz job 130586129291 failed on the same apt/DNS error before fuzz compilation. Retried failed jobs in run 43039199315 after its other jobs completed.
  • Rust Workspace Tests is still running. This is not a green/no-block CI verdict, and the hosted broker result is pending. No main merge, release or original-PR closure has occurred; review/merge authorization must name this current head.

@proffesor-for-testing proffesor-for-testing left a comment •

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

REMAND for final merge disposition: the qualified current-main candidate is green, while rUv review and the known standalone timing risk still need a decision. This COMMENT is not formal human approval.

Frozen tuple, refreshed November 12 at 16:00 UTC:

  • PR head: 68cd04927ac503ae68313edc249c6ea8627319c2.
  • Main: 094e1400719ded1283c67d23b9c737f6738a3244.
  • GitHub merge ref: 9b8f70754d036a77900d936b93795616d2242dd5; tree 2b732f00e133526b310d634a792a219aaba4d86d.
  • The PR remains CLEAN/MERGEABLE, with rUv requested and no submitted owner review.

EXECUTED receipts close all five original CI areas:

  • MQTT on head68: 43183390004 passed feature units and actual pinned Mosquitto integration; the owned broker stopped successfully.
  • QEMU and fuzz on the merge candidate: 43188188492, all 11 jobs passed, including actual fuzz testing.
  • Rust workspace and downstream checks: 43188188760, SUCCESS. All 37 mandatory Release cases passed, zero selected failures or ignores. Default-48 C4C completed 48 epochs, 395 parameters and 38,373 objective evaluations under the unchanged 60,000 ms / 100,000 caps. Full lifecycle 487.68 s is not fit time.
  • Bench: 43188188802, SUCCESS. Head68 also separately passed 43183609967.
  • Firmware, security, forecast Rust and Python workflows also remain SUCCESS on the merge candidate.

Retained counterevidence: standalone head68 43183605678 hit the unchanged 240-second operator guard at 76/80 windows before C4C. Child938 repeated it. These failures are not erased or claimed causally fixed.

Further probes did not reproduce the failure: child diagnostic full CI8370 43232492184 passed Debug plus all 38 Release cases and downstream sockets/FANN/CIR/witness checks; direct Release0e490242 passed all 38; failure-only comparison31b15c55 passed all nine C4 cases with normal-path logging removed. Controlled segment76 and partial-row77 storage holds both resumed to 80 durable windows and completed fit/cold/evaluate. These are additional probes, not substitutes for the parent's separate 37-case receipt or hardware qualification.

Child #2527 now has test/CI-only live evidence and lazy watchdog state at56d43b5a, with current-head gates still running. Parent head68 is unchanged. No speculative capture repair or timeout/budget relaxation was introduced. Keep the intermittent standalone timing risk visible for rUv's decision; refresh qualification if either parent/main ref changes. No main merge, release or helper installation was performed by this lane.

@proffesor-for-testing
proffesor-for-testing marked this pull request as draft November 12, 2026 18:25

@proffesor-for-testing proffesor-for-testing left a comment •

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

NO BLOCK for the reviewed code and CI at head 4f70c5eafe28a17395f1588cc15f5e64ba932d7d against main 094e1400719ded1283c67d23b9c737f6738a3244. This COMMENT is not formal human approval or merge authorization.

The current GitHub merge candidate 994e5896053579ae8e35b36f2bc8818160a37aad, tree af593d44129b957c917c14d2c1a6ad397c1dadc5, matches those exact parents. All 11 current-head workflows are terminal SUCCESS, including full CI43279683707, MQTT43279683661, QEMU/fuzz43279683678, security43279683676 and bench43279683922. Rust workspace, sockets, patched archive/cache compatibility, FANN, CIR and both witness guards passed.

The retained Release receipt is clean, has 1096 registered tests and 37 required passes with zero selected failures/ignores. C4: nine cases PASS526.41s. C4C: PASS629.70s for its full lifecycle; this is not fit duration. Test binary SHA256 fb559047d32f72eb4e1b4d859d22177954ab74ab63c3b39efd69a75407d9849b.

Root cause and repair: diagnostic211 exposed 22 final-Check preparation/restart gaps, no deferred ACK and no clock lead. The synthetic driver generated input before segment preparation completed; host latency repeatedly consumed its first-window buffer. The fixture now waits for verified preparation before generating synthetic packets, asserts no preparation credited as observation, and includes an actual five-second publication hold. Local full-journey regression also passed205.18s. All80 windows/phases/fit/cold/evaluation assertions, original240s phase wall guard, production clock/backpressure rules and default-48 training/evaluation caps are retained. Real-paced/backpressure tests are unchanged.

Opposite-assumption checks: no rows/flags are manually promoted, the wait shares the original phase-start Instant, bad/missing ACKs cannot produce success, and PR verification cannot invoke the legacy publisher/notifier. Publication/version jobs were skipped for verification, not qualified. No production capture repair, physical sensing/throughput qualification or repair of the separate historical Empty Check is claimed.

Current source/CI follow-up is complete. Ready for rUv's owner review and separate exact-head merge decision. Refresh qualification if head/main changes. No main merge, release or helper198 replacement was performed by this lane.

@proffesor-for-testing

Copy link
Copy Markdown
Collaborator Author

Dependency-security follow-up for the held stack: fresh November14 audit on descendant #2555 at783e69d4 found RUSTSEC-2026-0523/0524 (quinn-proto0.11.18) and RUSTSEC-2026-0538 (quinn-udp0.5.14), inherited from this parent lockfile. Prior green checks at4f70c5e do not include the new advisory result.

Separate draft #2556 at ef9dd1b patches only the two lock entries and required UDP dependency edges. Local locked compile checks and the unchanged repository-policy audit pass. #2555 now stacks on that repair and its current combined-head Rust Dependency Audit passes; fresh standalone security run43602259821 also passed its Rust audit. Other hosted checks remain in progress.

This parent and installed helper are unchanged. The merge hold remains; account for #2556 in the stack review before treating dependency security as current. No new audit ignores or approval/merge claim.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants