Repository navigation
fix(diagnostics): preserve rerun custody and synchronized inspection - #99
Conversation
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars. ⚙️ Run configuration
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
Independent review is active on exact detached739c1aef0ddd85c9200f65b9f8155d170737b8b3. Read-only admission check passes103 paths;96 and98 files match their reviewed commits,97 worker matches and the extra app seed guard/tests are being independently reviewed.7f393815→739 changes only plan/ledger; no earlier529 receipt is being treated as final evidence. Evidence access limit: your generated |
|
Independent final-tree source review is clear: complete103-path admission passes; companion96/98 bytes match exactly;97worker matches; the added seed guard/tests preserve the generic schema and enforce actual app prefix generation, repeated-DT time and qualified-native diagnostic origin before cold classification/pinning.7f393815→739 changes only plan/ledger. Fresh isolated target at published739c1aef just passed51 application tests,22 recording tests and the extreme-value oracle contract (one existing measured lifecycle test explicitly ignored). No previous529 or individual-branch execution receipt is being substituted. Native workflows and strict Clippy are in progress on this exact head. Acceptance audit distinguishes remaining implementation from inputs: full accepted configurable-provider binding/history/physical composition, typed accepted provider outcome persistence/specs/cold reader, durable cross-track rollout comparison, broader residual/Courant metrics and per-attempt cost samples remain future cloud-testable work under numerical/API/recording ownership. The absent aa82/Torch/parity bundle separately blocks measured A1. These are full-plan completeness boundaries, not newly observed regressions in PR99's declared bounded fixes. Two available legacy modal-provider consumer tests omitted by current ai-pressure ignored-test selection have now passed with its exact pinned checkpoint/export, without Torch: pressure-session replay/fallback and24-interval production stepping (23model selections, intentional unsupported-dt fallback, final-gate rollback). Final-tree rerun is planned; minimal follow-on workflow selectors are recorded on95. No shared/workflow edits by this reviewer. |
|
A further quantitative oracle defect is now an executed counterexample in an UNCHANGED separate tool: The native oracle also has an existing1e-14 absolute tolerance floor; values below it are only checked to that stated precision. I am treating that as an explicit limitation, not silently tightening or relaxing acceptance. The range repair will preserve current tolerances and divergence bounds. To keep99's reviewed integration stable, I will prepare a separate isolated follow-on draft limited to this tool and a new automatically discovered Cargo range contract, based on739c1ae. No99 source/inventory/receipt edit or merge. This gives the lead a transferable tested fix without duplicating active shared edits. Final99 fresh native/qualification review continues separately. |
|
Published the separately scoped native comparison verifier fix as draft PR100: #100 Branch Final PR99 independent checks have now also completed: clean exact739 qualification (18 commands), fresh packet normal/-O (26+26, no skips), and real eframe/Xvfb/llvmpipe cold-open/probe/export/missing-runtime failure/recovery/known-inflight Cancel/restart/window-Close. I am consolidating those fresh final-source receipts next. These are new independent runs, not transferred component or absent retained lead receipts. |
|
Independent final PR99 review and fresh exact-final-source validation complete. Bounded source/product approval; final hosted CI still required before lead merge. No shared sources changed by this review. This portable receipt is separate from the inaccessible retained lead receipts at7f: those remain uninspected. My new actual qualifier/native runs target final739 directly. The full local observation has SHA256 {
"schema": "rheon-independent-final99-observation-v1",
"recorded_utc": "2026-10-15T22:29:27.166715+00:00",
"subject": {
"pr": 99,
"head": "739c1aef0ddd85c9200f65b9f8155d170737b8b3",
"tree": "e9778d147b9e774a1d59b8262277a2b900060add",
"worktree": "/workspace/rheon-review99",
"clean": true,
"dedicated_target": "/tmp/rheon-final99-review-target",
"source_revision_stamp": "739c1aef0ddd85c9200f65b9f8155d170737b8b3"
},
"verdict": "Bounded source admission and independently executed final-source tests, qualification, packet and standalone native workflows PASS. Hosted CI and lead integration remain separate. Full A1-A8 acceptance is not claimed.",
"evidence_custody": "Fresh runs on final739; not transferred from component commits,529 or7f. Retained lead /tmp receipts remain absent in this execution environment and are not represented as inspected.",
"source_review": {
"complete_paths": 103,
"changed": 5,
"unchanged": 98,
"missing": 0,
"mismatched_pins": 0,
"inventory_sha256": "2974c09d19972bff7a91108e56dce88dbd3bd94f2e7a4eb7b19bf6643314ebb9",
"review_binding_sha256": "482983c09d1669847d4ea5c93f3282d65cbec4731cd243ec2e51d34d97c3399b",
"reviews": [
"/tmp/rheon-review99-admission-review.json",
"/tmp/rheon-review99-seed-review.json"
],
"seed_contract": "Actual prefix generation and repeated-DT time bits, qualified diagnostic origin/status validated before any cold classification/pin. Generic reader remains broader."
},
"tests": {
"app": {
"passed": 51,
"failed": 0,
"ignored": 1,
"ignored_boundary": "Existing measured-model case absent; not passed"
},
"diagnostic_recording": {
"passed": 22,
"failed": 0,
"ignored": 0
},
"diagnostic_oracle": {
"passed": 1,
"python_calls": 72
},
"strict_scoped_clippy": "passed",
"modal_consumer": {
"source_head": "739c1aef0ddd85c9200f65b9f8155d170737b8b3",
"source_tree": "e9778d147b9e774a1d59b8262277a2b900060add",
"command": "RHEON_AI_EXPORT=/workspace/.rheon-m23-review-assets/trajectory-v2-modal.bin cargo test --locked --no-default-features --test pressure_session_contract --test provider_step_contract -- --ignored --nocapture --test-threads=1",
"checkpoint_sha256": "25533a2fbaa5ab0f5fb574dc6307a2b8d6a686deb7db867d0d03338d4d70d5a9",
"export_sha256": "902ad8d77b16ae42f22f90bf71c22c19ae2e3cbe0cc0f4cc3d770930d4049aea",
"tests": [
{
"name": "frozen_model_selection_and_unsupported_domain_fallback_are_visible",
"passed": 1,
"failed": 0,
"ignored": 0,
"filtered": 3,
"seconds": 31.03
},
{
"name": "real_model_advances_production_session_for_twenty_four_intervals",
"passed": 1,
"failed": 0,
"ignored": 0,
"filtered": 6,
"seconds": 29.68,
"selected_model_intervals": "23/24"
}
],
"log_sha256": "34639bc13c8114a18d9d33a869e3d0328dc87e095be543baa20dde1dfb69aaac",
"boundary": "Existing legacy modal provider and supported/unsupported domain/rollback contracts only. This is not the missing aa82 deployment or configurable accepted-provider composition.",
"coverage_gap": "ai-pressure.yml supplies the existing export but only scopes solver_provider_contract; pressure_session_contract/provider_step_contract ignored consumers omitted. Minimal existing test-list addition is implementable; no workflow edit made in this reviewer worktree."
}
},
"qualification": {
"source_head": "739c1aef0ddd85c9200f65b9f8155d170737b8b3",
"source_tree": "e9778d147b9e774a1d59b8262277a2b900060add",
"clean": true,
"status": "passed",
"commands": 18,
"lean_checked": false,
"executable_sha256": "46ea4b2af04f5a54e264ecbf2a9bd04b862e74e04ef849f980af90a8d7037460",
"records_sha256": "ecdd0877f17997c577d5183cd928c27c802f9ffdcee8324f0802fe3841fddb45",
"receipt_sha256": "c7eb831fa49449b2687ef2acedbdc73d0e705c51a7d11237fd3bf3ca7824e780",
"stepping_authorized": false
},
"packet": {
"status": "passed",
"source_head": "739c1aef0ddd85c9200f65b9f8155d170737b8b3",
"source_tree": "e9778d147b9e774a1d59b8262277a2b900060add",
"normal_tests": 26,
"optimized_tests": 26,
"skips": 0,
"data_sha256": "3d71f88fd2e62b2f0b051bc783d7829f6cb131544418737a9e14dabe4b73078b",
"primary_qualification_sha256": "c7eb831fa49449b2687ef2acedbdc73d0e705c51a7d11237fd3bf3ca7824e780",
"receipt_sha256": "67eeb1ba41b35034402c62e0994e1cc25a422fa0305f2ece9aefd7a0b2c33caf",
"fluid_solves": 0
},
"limits": [
"No hardware/GPU, measuredaa82 bundle/Torch parity, descendants or hard shutdown deadline proof.",
"No fresh actual historicalf76 executor-guard or root desktop window workflow in this final739 observation; current app manufactured guard tests passed, prior receipts not transferred.",
"No accepted configurable learned-provider composition, accepted provider codec/capture, rollout benchmark/cost completeness or full scientific plan acceptance.",
"Existing archive is replay evidence, not restart checkpoint.",
"Lead owns all merges; no branch deleted and unrelated work preserved."
],
"observation_sha256": "9e6beda6f3db951f2122a5ddac655a4717cd17477e6d2e9454d323e23c268fe3",
"native_summary": {
"platform": "Linux X11 eframe via owned Xvfb :91, software Mesa25.0.7/llvmpipe LLVM19.1.7",
"binary": {
"path": "/workspace/.rheon-final99-review/bin/rheon-ai-compare-739c1ae",
"sha256": "9cb642a4088b7e3035a43f18398711af9c5d3fc47fba4d62ce137fa83a6a2a10"
},
"raw_probe_oracle": {
"head": "739c1aef0ddd85c9200f65b9f8155d170737b8b3",
"archive": "/workspace/.rheon-final99-review/original/run-1792102550875214206-119153/attempts.sqlite",
"archive_sha256": "df7425fd201b3ef9f940f7ece91855f3024ad5922603b217c53b4f1b392ad679",
"probes": [
{
"cell": [
8,
8,
0
],
"physical_m": [
0.53125,
0.53125,
0.03125
],
"candidate_pa": 0.0,
"native_pa": -43.22597225471727
},
{
"cell": [
8,
8,
3
],
"physical_m": [
0.53125,
0.53125,
0.21875
],
"candidate_pa": 0.0,
"native_pa": -44.05157682964029
}
],
"cold_without_manifest_or_new_child": true,
"window": "2097155"
},
"failed_saved_rerun": {
"head": "739c1aef0ddd85c9200f65b9f8155d170737b8b3",
"failed_finished": "/workspace/.rheon-final99-review/reruns/run-1792102758227319215-119872/finished.json",
"files": [
"completion.json",
"finished.json",
"intent.json",
"scene.json"
],
"exact_spec_parent": true,
"original_files_byte_exact": 25,
"child_pid_log_unchanged": true,
"archive_created": false,
"oracle_export": "/workspace/.rheon-final99-review/original/run-1792102550875214206-119153/diagnostic-comparison-0-0.json",
"oracle_modes": [
{
"mode": 0,
"positive": "independent diagnostic pressure oracle passed: 4096 cells",
"negative": "metric rmse"
},
{
"mode": 1,
"positive": "independent diagnostic pressure oracle passed: 4096 cells",
"negative": "metric rmse"
},
{
"mode": 2,
"positive": "independent diagnostic pressure oracle passed: 4096 cells",
"negative": "metric rmse"
}
],
"observed": {
"terminal": "failed",
"detail": "No such file or directory (os error 2)",
"statistics": {
"native_prefix_steps": 0,
"produced_attempts": 0,
"dispatched_calls": 0,
"qualified": 0,
"rejected": 0,
"unavailable": 0,
"recorded_attempts": 0,
"last_advanced_generation": null,
"model_start_seconds": 0.0,
"model_prediction_seconds": 0.0,
"run_wall_seconds": 0.002548285
}
}
},
"recovery": {
"head": "739c1aef0ddd85c9200f65b9f8155d170737b8b3",
"recovered_finished": "/workspace/.rheon-final99-review/reruns/run-1792102790431813641-119872/finished.json",
"source": "/workspace/.rheon-final99-review/original/run-1792102550875214206-119153/finished.json",
"exact_spec_and_parent": true,
"original_files_byte_exact": 25,
"manifest_restored_sha256": "521d5243a73418c89ba3ea95c584e2ece1b4533f2c634f1f96fbe34ee917cabd",
"immediate_child_observed_alive_then_absent": 120238,
"observed": {
"terminal": "completed",
"detail": "Typed diagnostic horizon complete; model drained",
"statistics": {
"native_prefix_steps": 15,
"produced_attempts": 3,
"dispatched_calls": 3,
"qualified": 0,
"rejected": 3,
"unavailable": 0,
"recorded_attempts": 3,
"last_advanced_generation": 18,
"model_start_seconds": 0.903933363,
"model_prediction_seconds": 0.029505278000000003,
"run_wall_seconds": 9.523871132
}
}
},
"restart_close_during_prediction": {
"head": "739c1aef0ddd85c9200f65b9f8155d170737b8b3",
"finished": "/workspace/.rheon-final99-review/slow-runs/run-1792102993621389313-126945/finished.json",
"restart_distinct_run": true,
"old_files_byte_exact": 22,
"inflight_marker_before_Close": {
"pid": 127917,
"request_id": 16
},
"immediate_child_observed_alive_then_absent": 127917,
"unadvanced_frame16_absent": true,
"observed": {
"terminal": "cancelled",
"detail": "Cancelled before diagnostic commit; recorded attempt retained",
"statistics": {
"native_prefix_steps": 15,
"produced_attempts": 1,
"dispatched_calls": 1,
"qualified": 0,
"rejected": 0,
"unavailable": 1,
"recorded_attempts": 1,
"last_advanced_generation": null,
"model_start_seconds": 0.88298352,
"model_prediction_seconds": 0.042680571,
"run_wall_seconds": 8.33019771
}
}
},
"manual_window_observations": [
"Cold reopen with deployment manifest absent verified retained archive; no child spawned.",
"Left probe atZ0, Zslider remapped both pane markers and footer toZ3; right-pane click anddifference toggle retained physical identity. Exact values independently decoded fromraw SQLite IEEE754 bytes; manually inspected screenshots, noOCR.",
"Export action produced actual4096-cell summary; independent diagnostic oracle passes normal/-O/PYTHONOPTIMIZE=1; RMSE+1 negative rejected by exactmetric name all3.",
"Cold failed-run open cleared prior fields/probe and displayed Missing archive/Failed0.",
"Cold cancelled-run open verified retained pin and showed NoCandidate/NotAdvanced with actual native qualified history, no substitute fields.",
"Fresh Start created distinct owner/run; Cancel and WM_DELETE Close each issued during positively observed in-flight request16. Each retained one unavailable/unadvanced attempt and no frame16.",
"Fast and slow native app sessions both exited0 after normal WM_PROTOCOLS close. Immediate children positively alive before relevant operations and absent after."
],
"original_source_preexisting_files_unchanged": 25,
"cancelled_source_files_unchanged_after_restart": 22,
"zero_manifest_restored_sha256": "521d5243a73418c89ba3ea95c584e2ece1b4533f2c634f1f96fbe34ee917cabd",
"slow_manifest_sha256": "1df7f2daee0b260eeadd1f979b6bbd865763d746777ac1f1d0202089eac06e86",
"cancel": {
"head": "739c1aef0ddd85c9200f65b9f8155d170737b8b3",
"finished": "/workspace/.rheon-final99-review/slow-runs/run-1792102906268054372-126945/finished.json",
"window": "2097155",
"inflight_marker_before_Cancel": {
"pid": 126988,
"request_id": 16
},
"immediate_child_observed_alive_then_absent": 126988,
"observed": {
"terminal": "cancelled",
"detail": "Cancelled before diagnostic commit; recorded attempt retained",
"statistics": {
"native_prefix_steps": 15,
"produced_attempts": 1,
"dispatched_calls": 1,
"qualified": 0,
"rejected": 0,
"unavailable": 1,
"recorded_attempts": 1,
"last_advanced_generation": null,
"model_start_seconds": 0.975539014,
"model_prediction_seconds": 0.05785129,
"run_wall_seconds": 8.298990423
}
}
}
},
"artifact_sha256": {
"/tmp/rheon-review99-admission-review.json": "bd1ddca527c829dcf935c9a479b426672190eb6f9487a2d851a545e9ac3de4e6",
"/tmp/rheon-review99-native-after-cancel.png": "f5aea52b30acebf00c171a3c9d7f98657e22e3517320feb93170be42d6b82c1d",
"/tmp/rheon-review99-native-cancel-cold-verified.png": "9f260aab66cb09ace5b5b21bbaa5a1a8832ccc8324474187c916643e044eb16f",
"/tmp/rheon-review99-native-cancel.json": "b6500d958029b940e31dbd9ae96a1c0617177bbdf410d84955422346a6d85125",
"/tmp/rheon-review99-native-close-during-prediction.json": "42702913228eb5898d530b12e2ea0ec581cf12eecaf5a22ce5ccafdbdad6d8d4",
"/tmp/rheon-review99-native-cold-loaded.png": "cc0c11cb2995af9a4305f85ab83d87ded5de792964c4c32bf6f3a9ada3b26fb6",
"/tmp/rheon-review99-native-failed-cold-cleared.png": "f6afe9c08f182e71a268306c9057889998eac490333340e66f703f7988cd072f",
"/tmp/rheon-review99-native-failed-rerun.json": "6b2d76d9c96e5beba32b72c4a1abcdabe7dcce605e99d0058824695e9f77ad29",
"/tmp/rheon-review99-native-original.json": "dc77c40578f956279802cc10578e246c46ac89c426c4b122b1e98054a6c13f94",
"/tmp/rheon-review99-native-probe-oracle.json": "d9b80ea55d0ae25ca052e98457e625f520495ddd43b6e5f616a2732a16f8a123",
"/tmp/rheon-review99-native-probe-z0.png": "0dd4eae7430979c397f2a9f384da568e407005e2824e2f6155884675db7543a2",
"/tmp/rheon-review99-native-probe-z3.png": "57d0f0bdc3b7b11d3c0f8c1bca2fa52fea8c9ef763871c37f2370a6a9d4a6684",
"/tmp/rheon-review99-native-recovery.json": "83c23478f08b04b8e1359d17da518ed3672541af0720342de4303a5f77c8a6d0",
"/tmp/rheon-review99-native-restart-warmup.png": "c498b946903f96c6a005bda50ddae396d66fcf2c26b99cd9b198072008b49346",
"/tmp/rheon-review99-native-right-probe-difference.png": "154b59d70f8542d5de06f61470b33719e29d5d6b89158a609b1f4214b3aa7449",
"/tmp/rheon-review99-seed-review.json": "103d60c6d59594ff9d7989e592728989707bf97e1a8ee9fbb04fd3877c41c396",
"/tmp/rheon-review99-qualified/qualification.json": "c7eb831fa49449b2687ef2acedbdc73d0e705c51a7d11237fd3bf3ca7824e780",
"/tmp/rheon-review99-packet/qualification.json": "67eeb1ba41b35034402c62e0994e1cc25a422fa0305f2ece9aefd7a0b2c33caf"
}
}Follow-on draft PR100 at Remaining code versus genuine input gaps: accepted configurable Session/provider composition and accepted-provider capture/codec/rollout metrics require allocated implementation; the exact aa82 weights/deployment/normalization/runtime/parity packet are genuinely absent. Existing trajectory-v2 consumer contracts now independently passed on739 (23/24model selections plus fallback/rollback), but are not a substitute for aa82. The existing CI test-list omission of those two ignored consumer binaries is implementable without new assets; no workflow edit made here. All merges remain lead-owned; no branches deleted. |
|
Hosted CI identity follow-through: fetched synthetic merge At this snapshot, default116695861409, desktop116695860955 and Pages/build116695860776 are still pending/in progress; no all-green or merge readiness claim. Final independent source/product review is complete and portable at #99 (comment). Draft follow-on100/4ac6799 remains separately gated and stacked; preserve99 branch until dependent branch no longer requires it. All merges remain lead-owned. |
1 participant
A frozen diagnostic rerun could lose its intent when the runtime package disappeared or changed. Cold run-set verification also accepted generic seed histories that could not come from this application's native prefix. Preserve intent and parent before mutable package checks, then bind application histories to their exact prefix generation, repeated-DT clock and native-qualified origin/status. Failed startup retains honest Failed metadata without fabricating an archive or inference; genuine within-horizon recording-capacity tails remain valid.
Compose the seven allocated files from #96, #97 and #98, plus the independently reviewed app-only seed correction. The pressure oracle uses a scaled independent
fsumnorm for representable tiny/large values. Cold/live inspection caches reductions by exact strongly retained immutable publication identity, remaps X/Y probes onto the selected Z, and labels actual native and candidate observations separately. Explicit export and rerun still reopen and verify source custody independently of the display cache. Generic recording/schema, numerical acceptance, existing owners and legacy frame controls retain their contracts.Independent GPT-6.1 Sol review approved the final source hashes and identified the additional seed gap before admission. The unchanged conservative selector still covers all 103 paths; no gate or inventory scope is narrowed. Original companion PRs and branches remain intact.
Validation on clean source
7f393815e23238ee6e07fe509cd993f6a40ba3fe, tree9acd04b4ecfa6e7c87f37b07a3cab8d63a39bd8f; final follow-through changes only plan/ledger:-O, with zero skips. Historical qualification receipts remain tied to their original subjects.-OandPYTHONOPTIMIZE=1oracles; perturbed RMSE is rejected in all modes.API contract, review/admission and evidence ledger. Native observation SHA256
622d2f3bae768269991f6e51a084f71e40d4b9222a3ef46cc342783f74a29139; generated artifacts stay outside Git. An initial local cache-seeding run was discarded; only corrected isolated-target validation is used.Includes the diagnostic runtime now merged through #95 at25ab73f, following #94; retargeted to main with an identical proposed merge tree. Keep draft until applicable exact-head hosted checks pass, then retarget/merge under existing lead authority. Full measured A1, accepted learned-provider A2 and unsupported platforms remain open; no training, substitute checkpoint, hardware/latency/deadline claim or README #90 merge.