Skip to content

Reconcile C5 and variable-source protected Retry with reviewed parent - #2527

Draft
proffesor-for-testing wants to merge 14 commits into
integration/ruv-main-reconcile-20261110from
prepare/build1102-20261111
Draft

proffesor-for-testing wants to merge 14 commits into
integration/ruv-main-reconcile-20261110from
prepare/build1102-20261111

Conversation

@proffesor-for-testing

@proffesor-for-testing proffesor-for-testing commented Nov 11, 2026 •

Copy link
Copy Markdown
Collaborator

Reconcile C5/variable-source/protected-Retry with the corrected parent #2513, preserving source identity, original deadlines and replay fences. Current head 3d26dc1a1b796d0bd45b933a28d1edf25ebf97ca, branch prepare/build1102-20261111 → integration/ruv-main-reconcile-20261110 at parent 4f70c5eafe28a17395f1588cc15f5e64ba932d7d. All five current-head workflows are SUCCESS. Code/CI review has NO BLOCK; draft is retained for the parent landing/retargeting dependency.

Changes:

  • Explicit C5 8-bit/53-position admission and truthful PHY diagnostics; versioned 2–8-source acquisition/recovery identity with legacy coverage.
  • Protected Retry retains confirmed native prefixes, failed-tail replay floors and the original deadline, including lost-ACK recovery.
  • Parent CI bootstrap repairs and numerical-equivalent AETHER allocation reuse, with existing fit/evaluation caps retained.
  • Native-persistence Retry is explicitly classified in Debug and mandatory/exact in Release: prior37 cases plus Retry,38 total.
  • Live Release batch logs retain exit status, exact counts, binary custody and partial evidence; failed output forwarding reaps the child. Watchdog metadata is lazy/failure-only.
  • Correct the synthetic operator fixture: await segment preparation at its supplied receiver coordinate before generating packets. The final Check deliberately holds publication five seconds. No preparation time is credited as observation. All80 windows, all phases, original240s guard and fit/cold/evaluation assertions remain; production timing/capture and real-paced/backpressure tests are unchanged.
  • Verification-only PR/manual events skip the legacy publisher/notifier; existing main-push behavior is unchanged.

Root evidence: failing56 and diagnostic211 reproduced phase7 at76 durable windows. Internal state showed22 gap/restart cycles, no deferred ACK or clock lead, and the writer still busy. A separate controlled three-delay probe recreated the same reset state. This was synthetic input/preparation ordering, not an unpolled ACK. No claim is made that this repairs the separate historical physical Empty Check.

Executed validation:

  • Current code:55 classifier/runner self-tests pass, including actual IPC streaming/exit/cleanup controls; original buffering fails the early-progress negative control. Additional stderr-flood and SIGTERM-ignore probes pass.
  • Equivalent local full-operator prototype with the same barrier/five-second delay passes in205.18s:80 windows plus fit/cold/evaluate. This is synthetic fixture evidence, not exact-head CI or hardware acceptance.
  • Original production reconciliation treebbcf:102 focused library and9 API cases pass; workspace/CIR bench compilation passes. Later changes are CI/test-only.
  • Earlier diagnostic full CI8370 and direct Release0e490242 each passed38 selected Release cases; failure-only31 passed9 C4 cases. Their positive evidence and the subsequent failures are retained, not substituted for new-head checks.

Current-head qualification is terminal SUCCESS: full CI43280148006, security43280152033, bench43280146942. All 38 required Release cases and full workspace/downstream checks passed, with zero selected failures/ignores. Keep draft until the parent dependency and base are reconciled. Final review. Installed helper198, room/capture/device data and other sessions' app work are unchanged by this lane. No server PR has been merged.

Combines main dea1cc2 with the reviewed C5/helper source. Retains both unified-input and FANN dependencies, ZIP backports and SQLite repair. Resolves metadata/provenance conflicts without changing the held main PR.

Agent: Codex; host: DraganMac; session: ruview-phone-room-sync-20261111
Add an explicit 2-8-source profile with bounded receiver arrival bursts, dynamic input/model/evaluation/replay bindings and isolated recovery. Preserve earlier three-source contracts and pending recovery. Keep operator conditions, reserved checks and model selection explicit.

Verified locally: 10 targeted variable-profile tests, 66 scoped regressions, 27 API tests, FANN and SQLite controls. Synthetic software checks do not qualify the physical room.

Agent: Codex; host: DraganMac; session: ruview-phone-room-sync-20261111
…al recovery

Retain structural, gap, replay and native-window limits for partial checkpoints; validate new arrival aggregate rates over complete episode parts without paused intervals. Discover actual temporal journals read-only and keep old/expired runs visible without renewal, ignoring retired heads.

107 focused Release tests pass. Independent review NO BLOCK on exact six-file diff 4d2afb47439756b1ab3077972ac6558a1a39b356e2e260fcf02cb4933cb97c53. No physical qualification claim.

Agent: Codex; host: DraganMac; session: ruview-phone-room-sync-20261111
@proffesor-for-testing proffesor-for-testing changed the title Support C5 native input, variable-source captures and protected Retry Reconcile C5 and variable-source protected Retry with reviewed parent Nov 12, 2026

@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 pending terminal hosted verification. No new source blocker found in the scoped parent reconciliation at bbcf38e (parents19852807+68cd0492; treed9cdb89efc49437458dd1479d21c3464266dd2d2).

STATIC: exactly seven incoming CI/helper/AETHER paths match the reviewed parent. Published C5/native/variable-source/Retry/recovery/API paths are unchanged from198. This does not silently adopt dirty worktree experiments or alter installed helper198/artifacts. Full CI/security are explicitly dispatched because the stacked base is excluded by their PR branch filters.

EXECUTED: Release library compiles with1121 registered tests.102 unique selected library tests and9 server API tests pass with0failed/ignored; selections are matched to actual registries and retained tree/executable hashes. Coverage includes all256 C5 header flag variants, explicit packing/PHY claims, receiver-only clock authority,2–8-source bounds, four-source fit/cold identity, reserved validation separation, old recovery namespaces/cross-profile ciphertext denial, authenticated recovery status and protected Retry prefix/expiry/replay controls. These are in-process/synthetic fixture checks, not physical sensing qualification.

The historical bench failure was APT mirror DNS before compilation; the current hosted install step passes using the parent's signedCanonical helper. Current security/QEMU/fuzz/Python/data-policy/fix-marker workflows pass. Local workspace and CIR feature benches compile/link successfully; hosted workspace/Release and bench gates remain active, so do not infer hosted terminal success. Parent#2513 qualification does not substitute for this child's distinct runtime tests.

Second pass checked profile-selection identity and old-default discovery, namespace/key separation, flag/header binding and API negotiation, unchanged production budgets and caller compatibility. No sanitation, expiry, original deadline, anti-replay, confirmation or model-selection authority was relaxed by reconciliation. No main merge, install, firmware or data action performed. Existing#2513 standalone timing/advisoryDocker caveats remain separate.

This is a Codex COMMENT, not independent human approval of the entire historical stack. Exact next gate: terminal current-head CI/bench, resolve any demonstrated failure, then parent/human review and separately qualified helper/native/physical acceptance. Keep draft and target parent.

@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 pending current-head hosted verification at 93893c1, tree1b6825b421e18d15f87d0a94e2a967d70f4ce732 (parent mergebbcf; parent2513=68cd0492). No new source blocker found in the bounded reconciliation/profile correction. This is a Codex COMMENT, not human approval of the historical stack.

EXECUTED failure/remedy: old bench failed Ubuntu mirror DNS before compilation; parent merge's entire Linux bench workflow now passes. First full Debug run then exposed one native Retry journey missing from the existing Release policy (120s watchdog,1094pass/1fail/29ignore). Current correction keeps all assertions/watchdogs/production limits, registers the case transparently in Debug and requires its exact active Release execution. Selection grows37→38, never zero/ignored coverage.

EXECUTED:52 policy self-tests pass, actual Debug classifier passes, exact unchanged Retry Release case passes11.94s with0ignored. Removing its name or ignored classification from the actual registry is rejected. On parent mergebbcf,102 selected library+9 API Release tests and local workspace/CIR compilation passed; these are historical production-equivalent receipts, not invented current-hosted results.

STATIC: post-bbcf delta is ONLYscripts/ci/temporal_test_profiles.py, its self-tests and two test-only classification/precondition lines. Shipping C5/variable-source/Retry/recovery/API code is unchanged from198; AETHER changes are the reviewed parent optimization. Scope covers flag/header binding, source-count/resource bounds, receiver-only clock authority, namespace/ciphertext/old-profile isolation, reserved checks and original prefix/expiry/replay controls. No data/permission/model authority was expanded.

Current security and whole bench workflows PASS. Full CI passes Debug but fails Release C4 legacy operator-eight240s watchdog; C1/C2/C3pass, C4C/newRetry not reached. Isolated diagnostic43227196104 atb3c10477 adds only bounded test metadata and one fixed-case workflow; PR938 remains unchanged. Keep draft and parent dependency. Native installed/helper/physical sensing remains unverified by this source lane. No main merge, artifact replacement or device action. Next gate: terminal938 hosted counts/gates, fix demonstrated failures, then parent/human review and separate helper/native acceptance.

Investigation update, 2026-11-12: isolated Ubuntu operator journey passed1case/80durable windows(283.94s), and original nine-case C4batch passed9/0failed/0ignored(405.30s), runs43227196104/43230046573. Individual240s guard unchanged. A separate controlled Mac probe actually held finalCheck segment publication, triggered temporal_storage_busy_pause, released it, resumed and completed80durable windows(1passed258.78s); unshipped experiment restored and PR worktree clean. These refute a deterministic C4 predecessor failure and that specific pause/resume failure; they do not establish the root cause of the earlier repeated failures.

Original fullCI with bounded test-only progress diagnostics remains43232492184 at8370be07. Complete original Release profile sequence/all38mandatory cases is43236354700 at diagnostic-only0e490242. No production repair, source-head update or main merge is claimed. PR2527 remains93893c11; REMAND remains until the unexplained Release failure is resolved or separately accepted by the maintainer with this evidence.

@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 merge readiness at 56d43b5a5a94eb21af9c840ffd5ab72f0aea512c: current-head Rust/bench checks and the parent/owner decision remain open. No demonstrated source defect was found in the three-file CI/test follow-up. This is a Codex authored-change adversarial review, not formal human approval.

Scope: commits ac999c1a and 56d43b5a after 93893c11; the profile runner, its tests, and the operator watchdog's lazy failure message. Production capture, packet pacing, original deadlines, test/fit budgets, auth and storage authority are unchanged. Deployment/IAM/hardware mutation lenses are outside this diff.

EXECUTED:

  • 55 self-tests pass. The original buffered behavior fails the early-progress negative control.
  • Real child processes prove evidence is visible before exit; stdout/stderr and exit7 are retained; a broken console reaps the child and preserves a partial log.
  • Adversarial probes retain262,144 stderr bytes without hiding exit7, and kill/reap a child ignoring SIGTERM after 5.02 s (exit-9) while retaining its partial log.
  • A real Rust libtest I/O smoke passes using a labeled experimental binary; this is not clean-head runtime qualification.
  • Full diagnostic CI8370 and direct Release0e490242 each pass all 38 required Release cases. The full workflow is terminal SUCCESS. Both produced libtest SHA256 804bff0009bbb5c2413f004e55919ec52497c4954bd694c80aa2830cf9d0d487.
  • Failure-only diagnostic31b15c55 passed nine C4 cases in 532.80 s. Its entire v2/, vendor/ and firmware tree matches this head.

STATIC: executable hash verification remains immediately before invocation; nonzero-exit, exact pass-count and zero-ignore checks are retained. The opposite-assumption pass found no new command/shell authority, lost exit status or stranded direct child in the executed probes. The watchdog keeps its original 240-second expression and only formats scalar state if it fails.

Limits: the earlier938/68 watchdog failure remains unexplained, not claimed production-fixed. No hardware/model accuracy qualification follows. Current56 security and bench are SUCCESS. Exact-head CI43248905947 is now FAILED in C4; C4C and the new Retry batch were not reached in this run. The cancelled ac99 benchmark was superseded; the replacement56 run is retained. Earlier diagnostic passes are distinct from terminal current-head evidence.

Next gate: terminal exact56 Rust/Release/bench results, refresh review, and obtain the parent/rUv decision. Keep draft/merge hold until that is reconciled. Installed helper198 and other sessions' app/device/data work are unchanged by this lane.

EXECUTED recurrence, November12 at16:50UTC: current56 C4 batch685.86s,8passed/1failed/0ignored. The original240s phase guard reports phase7: retained76, durable76, observed0, durableObserved0, pendingtrue, stageobserving, lead-in0, runtimecapture, no durability/capture error. This is a concrete current-head blocker, not merely historical risk. Lazy failure state and live streaming worked. Keep REMAND while distinguishing repeated storage-busy restarts from deferred ACK/clock admission; no timeout or production guard change is authorized by this diagnosis.

@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 3d26dc1a1b796d0bd45b933a28d1edf25ebf97ca, based on parent 4f70c5eafe28a17395f1588cc15f5e64ba932d7d. Keep draft for the parent landing/retargeting dependency. This COMMENT is not formal human approval or merge authorization.

All five current-head workflows are terminal SUCCESS: full CI43280148006, security43280152033, bench43280146942, data policy and fix markers. Full Rust workspace, sockets, compatibility, FANN/CIR and witness guards passed. The clean Release receipt registers1121 tests and executes all38 required cases with zero selected failures/ignores: C4 nine cases495.70s, C4C599.84s (whole lifecycle), variable Retry17.23s. Binary SHA256 a0db11cd5c5270f94df8c1061d123cda20258b35893bbad24e8095819ef86e02; runner SHA256 940a7738841df3573f911cd1721ca6c8530652ff06df52f5e80e27195bb3d9d6.

The parent preparation barrier corrects the 22-restart synthetic fixture failure without changing production capture/clock/buffer behavior or raising the original240s guard. The five-second preparation hold passes the full80-window/fit/cold/evaluate journey. All previous37 Release cases plus exact variable Retry remain mandatory. C5/2–8-source/Retry production code remains covered by prior focused cases and the new full workspace run; original deadlines/replay fences are retained.

Runner adversarial controls:55 self-tests pass; original buffering fails the early-progress negative control. Actual processes retain stdout/stderr/nonzero exits, expose progress before exit, retain262144 stderr bytes, and kill/reap a SIGTERM-ignoring child on output failure after5.02s while keeping partial evidence. Hash and exact-count checks remain before invocation/after execution. No new authority, shell expansion or source defect was demonstrated in the scoped follow-up.

Limits: synthetic code/CI qualification is distinct from hardware/customer/model-accuracy acceptance. Historical physical Empty Check remains separate. Installed helper198 and app/device/data state are unchanged by this lane. Parent must land and the child base be reconciled before a main merge; no child merge into the held parent is implied. Current CI/review work is complete, with ownership/merge decisions still human.

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.

1 participant