fix(claude): preserve valid reviews and reject failed executions

.flow/specs/fn-291-preserve-successful-claude-reviews-with.json

@@ -0,0 +1,230 @@
+{
+  "branch_name": "investigate/qwen-native-review-20261010",
+  "completion_review_status": "ship",
+  "completion_reviewed_at": "2026-10-24T17:48:08.722718Z",
+  "created_at": "2026-10-10T18:09:50.382435Z",
+  "default_impl": null,
+  "default_review": null,
+  "default_sync": null,
+  "depends_on_epics": [],
+  "id": "fn-291-preserve-successful-claude-reviews-with",
+  "impl_review_rounds": {
+    "fn-291-preserve-successful-claude-reviews-with.1": 0
+  },
+  "next_task": 1,
+  "no_plan": true,
+  "plan_review_rounds": 0,
+  "plan_review_status": "unknown",
+  "plan_reviewed_at": null,
+  "review_attempts": [
+    {
+      "artifact_sha256": "4ca87d4612f0f139fad9d70d97b6b92f908be393129979fd82ad23e5709a5616",
+      "backend": "codex",
+      "base_sha": "39308b309fde394f4f18e8ac12f1fb7593844404",
+      "counter_kind": "impl",
+      "effort": "high",
+      "failure_class": null,
+      "finalized": {
+        "digest": "complete",
+        "receipt": "complete",
+        "status": "not_applicable"
+      },
+      "findings_digest": {
+        "backend": "codex",
+        "digest_truncated": false,
+        "items": [
+          {
+            "chainRoot": "finding-b6e46a51078fc3c76fec253a8b1c93d0",
+            "classification": "introduced",
+            "findingId": "finding-b6e46a51078fc3c76fec253a8b1c93d0",
+            "firstSeenThisRound": true,
+            "severity": "P2",
+            "status": "open"
+          }
+        ],
+        "reviewKind": "implementation"
+      },
+      "forced": false,
+      "hash_epoch": 0,
+      "head_sha": "32c14fb147da17ee537a232e90e0686c1271167f",
+      "head_sha_observed": true,
+      "kind": "impl",
+      "model": "gpt-6-sol",
+      "outcome": "verdict",
+      "output_bytes": 716,
+      "output_sha256": "9cf39239039cceba7d8f57155a4ea7a09948fe9cd0de97e974de0301b9d42e8a",
+      "reservation_id": "eb7a62e8eb824cf9bbfe1369d481d136",
+      "round_consumed": true,
+      "scope": "impl:fn-291-preserve-successful-claude-reviews-with.1",
+      "superseded_by": null,
+      "task": "fn-291-preserve-successful-claude-reviews-with.1",
+      "timestamp": "2026-10-10T19:51:00.215112Z",
+      "verdict": "SHIP"
+    },
+    {
+      "artifact_sha256": "3a027791fbf5eed79ea33f261c8fea92b95fa36dbb0245b20e5529af81a0a3b8",
+      "backend": "codex",
+      "base_sha": "39308b309fde394f4f18e8ac12f1fb7593844404",
+      "counter_kind": "impl",
+      "effort": "high",
+      "failure_class": null,
+      "finalized": {
+        "digest": "complete",
+        "receipt": "complete",
+        "status": "not_applicable"
+      },
+      "findings_digest": {
+        "backend": "codex",
+        "digest_truncated": false,
+        "items": [
+          {
+            "chainRoot": "finding-87fddff3b082189fffcdf2bc64e3e20d",
+            "classification": "introduced",
+            "findingId": "finding-87fddff3b082189fffcdf2bc64e3e20d",
+            "firstSeenThisRound": true,
+            "severity": "P1",
+            "status": "open"
+          },
+          {
+            "chainRoot": "finding-3c93daef51011df855fb29e8a1b99367",
+            "classification": "introduced",
+            "findingId": "finding-3c93daef51011df855fb29e8a1b99367",
+            "firstSeenThisRound": true,
+            "severity": "P1",
+            "status": "open"
+          }
+        ],
+        "reviewKind": "implementation"
+      },
+      "forced": false,
+      "hash_epoch": 1,
+      "head_sha": "acfd179ae82bda87e659e97ac642823cc4217a74",
+      "head_sha_observed": true,
+      "kind": "impl",
+      "model": "gpt-6-sol",
+      "outcome": "verdict",
+      "output_bytes": 2753,
+      "output_sha256": "61fec53f23d8143511e0f9df65977a7dc587b3ce571c6e872bce1a1ca242bf14",
+      "reservation_id": "5b419962d64d4ba897276ab3e25fe4c2",
+      "round_consumed": true,
+      "scope": "impl:fn-291-preserve-successful-claude-reviews-with.1",
+      "superseded_by": null,
+      "task": "fn-291-preserve-successful-claude-reviews-with.1",
+      "timestamp": "2026-10-10T20:05:03.253139Z",
+      "verdict": "NEEDS_WORK"
+    },
+    {
+      "artifact_sha256": "b54227d6fc7189d68c0288e7519cdb480d2e537ceac3e2409a343f66d5365d30",
+      "backend": "codex",
+      "base_sha": "39308b309fde394f4f18e8ac12f1fb7593844404",
+      "counter_kind": "impl",
+      "effort": "high",
+      "failure_class": null,
+      "finalized": {
+        "digest": "complete",
+        "receipt": "complete",
+        "status": "not_applicable"
+      },
+      "findings_digest": {
+        "backend": "codex",
+        "digest_truncated": false,
+        "items": [
+          {
+            "chainRoot": "finding-87fddff3b082189fffcdf2bc64e3e20d",
+            "classification": "introduced",
+            "findingId": "finding-87fddff3b082189fffcdf2bc64e3e20d",
+            "firstSeenThisRound": false,
+            "severity": "P1",
+            "status": "fixed"
+          },
+          {
+            "chainRoot": "finding-3c93daef51011df855fb29e8a1b99367",
+            "classification": "introduced",
+            "findingId": "finding-3c93daef51011df855fb29e8a1b99367",
+            "firstSeenThisRound": false,
+            "severity": "P1",
+            "status": "fixed"
+          }
+        ],
+        "reviewKind": "implementation"
+      },
+      "forced": false,
+      "hash_epoch": 1,
+      "head_sha": "57eaa175861a8cd1b51f5eceb515ec765726e890",
+      "head_sha_observed": true,
+      "kind": "impl",
+      "model": "gpt-6-sol",
+      "outcome": "verdict",
+      "output_bytes": 233491,
+      "output_sha256": "ccdc485cade3db4a8c5a56d5ce8edf00140e1877ff205d0193c85861d0889c80",
+      "reservation_id": "dc7b695a7b0744aa84eddd07382ea5c3",
+      "round_consumed": true,
+      "scope": "impl:fn-291-preserve-successful-claude-reviews-with.1",
+      "superseded_by": null,
+      "task": "fn-291-preserve-successful-claude-reviews-with.1",
+      "timestamp": "2026-10-11T07:31:54.352070Z",
+      "tool_calls": 29,
+      "verdict": "SHIP"
+    },
+    {
+      "artifact_sha256": "e3a1c8b56620269fbae948cb97f21c083de854b4396142f51954b732e7269039",
+      "backend": "copilot",
+      "base_sha": "f70767b9242370bdd7078b1f953bf2a849781178",
+      "counter_kind": "plan",
+      "effort": "high",
+      "failure_class": null,
+      "finalized": {
+        "digest": "complete",
+        "receipt": "complete",
+        "status": "complete"
+      },
+      "findings_digest": {
+        "backend": "copilot",
+        "digest_truncated": false,
+        "items": [],
+        "reviewKind": "completion"
+      },
+      "forced": false,
+      "hash_epoch": 0,
+      "head_sha": "b8b9a916e163940fb0c6a312f63a6c9152def466",
+      "head_sha_observed": true,
+      "kind": "completion",
+      "model": "claude-opus-5.5",
+      "outcome": "verdict",
+      "output_bytes": 6073,
+      "output_sha256": "e7843e272f758bd4944e46e748d332bddad8ba0f311cfcfc3269eade66815dbb",
+      "reservation_id": "ebaab2d7c6084af3b94da5ecd97ab660",
+      "round_consumed": true,
+      "scope": "completion",
+      "superseded_by": null,
+      "task": null,
+      "timestamp": "2026-10-24T17:48:08.719083Z",
+      "verdict": "SHIP"
+    }
+  ],
+  "review_hash_epoch": {
+    "impl:fn-291-preserve-successful-claude-reviews-with.1": 2,
+    "plan#completion": 1
+  },
+  "review_pending_rounds": {},
+  "review_reservations": {},
+  "review_transport_failures": {
+    "completion": 0,
+    "impl:fn-291-preserve-successful-claude-reviews-with.1": 0
+  },
+  "spec_path": ".flow/specs/fn-291-preserve-successful-claude-reviews-with.md",
+  "status": "done",
+  "title": "Preserve successful Claude reviews with unknown-model diagnostics",
+  "tracker": {
+    "baseHashFlow": null,
+    "baseHashTracker": null,
+    "depRelations": [],
+    "id": null,
+    "identifier": null,
+    "lastSyncedAt": null,
+    "mergeBaseFlow": null,
+    "mergeBaseTracker": null,
+    "url": null
+  },
+  "updated_at": "2026-10-24T17:48:08.723373Z"
+}

.flow/specs/fn-291-preserve-successful-claude-reviews-with.md

@@ -0,0 +1,33 @@
+# Preserve successful Claude reviews with unknown-model diagnostics
+
+## Goal & Context
+
+Claude Code can return a successful JSON result for a custom gateway model while emitting an unrecognized-model diagnostic on stderr. The native review backend currently discards that terminal review and refunds its attempt. Preserve valid successful reviews without weakening transport failures.
+
+## Architecture & Data Models
+
+Keep the shared review driver, backend registry, receipt schema and finalizers. Correct only Claude result classification and update the existing transport documentation.
+
+## API Contracts
+
+The sole JSON result envelope remains the reviewer output authority. Stderr stays diagnostic. Explicit models never fall back to another model.
+
+## Edge Cases & Constraints
+
+Invalid JSON, wrong envelopes, error envelopes, nonzero process exits and timeouts remain failures even if their text contains a verdict. Preserve genuine unavailable-model fallback and its cache semantics.
+
+## Acceptance Criteria
+
+- **R1:** A successful Claude JSON result accompanied by an unrecognized-model diagnostic can finalize official plan, implementation and completion reviews. Errors: malformed or wrong-type output and error envelopes cannot become verdicts.
+- **R2:** Explicit custom models are retained on success, with no fallback or false availability-cache entry. Errors: genuine unavailable-model responses keep the existing bounded ladder; explicit selections fail without substitution.
+- **R3:** A failed process cannot finalize a verdict found in its output, and attempts are refunded through the official handler. Errors: nonzero exits, timeouts and unavailable/error envelopes publish no new valid receipt.
+- **R4:** Implementation re-review retains its session and previous findings after a successful warning-bearing result. Errors: stale identity and failed re-review retain existing safeguards.
+- **R5:** Existing transport documentation accurately distinguishes successful custom-model diagnostics from failures. No error surface beyond the documented result contract.
+
+## Boundaries
+
+No new backend, model-specific option, new config key, prompt rewrite, provider fallback, global installation or evaluation trace in the public branch. Standing G1/G2 remain applicable.
+
+## Decision Context
+
+A real terminal custom-model response and a native official review demonstrate the false failure. Documented aliases do not remove the diagnostic. Fixing the existing backend preserves the current architecture and avoids a redundant integration.