fix(review): stop presenting fallback inventories as completed reviews

.github/workflows/opencode-review-dispatch.yml

@@ -5302,7 +5302,6 @@ jobs:
               --run-attempt "$RUN_ATTEMPT" \
               --coverage-result "${COVERAGE_EVIDENCE_RESULT:-unknown}" \
               >"$body_file"
-            printf '\n%s\n\n%s\n' "## Review outcome" "Coverage is a gate, not the review. This body reviews the changed product files." >>"$body_file"
             create_pull_review "$event" "$(cat "$body_file")"
             # create_pull_review COMMENT rewrites the status comment to Gate
             # result: COMMENT. Restore the coverage gate so a miss never looks

AGENTS.md

@@ -1,6 +1,9 @@
 # AGENTS.md — ContextualWisdomLab .github
 
 <!-- CWL-ENTRY -->
+Fallback file inventories must not claim a completed model review or observed
+runtime flow. See [the reproduction](docs/doctoring/review-fallback-quality-boundary.md).
+
 > **Agents: read the master context FIRST.** Before any work, read [`docs/CWL-MASTER-CONTEXT.md`](docs/CWL-MASTER-CONTEXT.md) (mission · naruon-as-platform + inter-component UML · cross-cutting disciplines · conventions · roadmap · current state), the live **GitHub Project #1** <https://github.com/orgs/ContextualWisdomLab/projects/1> (work/roadmap source of truth), the full spec **ContextualWisdomLab/naruon#974**, the live gap snapshot [`docs/product-technical-gap-baseline.md`](docs/product-technical-gap-baseline.md) (not merge authorization; Figma File ID for this repo is N/A per [`docs/adr/0002-product-technical-gap-baseline.md`](docs/adr/0002-product-technical-gap-baseline.md)), and operate the Project per [`docs/agent-github-project-protocol.md`](docs/agent-github-project-protocol.md). The repo/Project — not any private agent memory — is the source of truth. The standing autonomous operating directive for the continuous PR review→fix→merge→develop loop across the ecosystem is [`docs/product-goal-directive.md`](docs/product-goal-directive.md) — a `/goal` session's 4000-character pointer refers to it; read the full directive before running or configuring any such loop.
 
 Materialize accepts only exact SHA-256 pins, a bounded relative `-r` include

CHANGELOG.md

@@ -1,3 +1,9 @@
+### OpenCode fallback output states its review limits
+
+- Fallback comments identify the changed-file inventory and the next action,
+  instead of claiming completed model review. Remove the duplicate generated
+  diagram and identify inspected public symbols without asserting an API change.
+
 ### Failed-check finding names the Strix sandbox instead of the gateway
 
 - `opencode-review-dispatch.yml`'s `emit_strix_provider_failure_finding` rendered one fixed finding for every `STRIX_PROVIDER_UNAVAILABLE` line, whose Root cause read "The contextual-orchestrator gateway or its discovered provider pool was unavailable for this run". `#1953` had just given the Strix sandbox bootstrap failure its own second verdict token (`STRIX_SANDBOX_UNAVAILABLE`) precisely because that attribution is wrong for it -- the sandbox container never reaches its Caido proxy, so the run dies before the gateway serves anything -- and this consumer re-applied the wrong attribution one step downstream, into the review findings and the failure census. The emitter now branches on the second token: a sandbox verdict gets a finding that names Strix's sandbox, says the verdict does not name the gateway, and tells the reader not to change gateway or provider configuration on its strength. A `STRIX_PROVIDER_UNAVAILABLE` line without the token keeps its existing text verbatim, so the gateway class has no regression surface. No test covered this finding text at all before (`gateway or its discovered provider pool` matched nothing under `tests/`); `tests/test_opencode_dispatch_strix_sandbox_finding.py` now runs the production emitter from the published run block and pins both directions plus the no-signal case. Refs #1953, #1935.

CLAUDE.md

@@ -4,6 +4,9 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 ## Read first
 
+Keep fallback inventories distinct from completed model reviews; reproduce the
+publication contract using [this record](docs/doctoring/review-fallback-quality-boundary.md).
+
 `AGENTS.md` is the canonical agent entry point. Per its instructions, before any work read
 [`docs/CWL-MASTER-CONTEXT.md`](docs/CWL-MASTER-CONTEXT.md) (mission, ecosystem UML, cross-cutting
 disciplines CP-1..CP-5/G6/SEAM, binding engineering conventions in §7, roadmap), the live

docs/doctoring/review-fallback-quality-boundary.md

@@ -0,0 +1,83 @@
+# Review fallback quality boundary
+
+## Observed failure
+
+On 2026-09-08, the latest five Noema/OpenCode formal review objects found in the
+20 newest open `.github` PRs were all fallback/status-derived OpenCode output.
+There were 11 matching review objects in that bounded collection. The selected
+five are not a representative sample of model quality or the organization.
+
+| PR | Review ID | Reviewed commit | Matched head at collection |
+| --- | --- | --- | --- |
+| #2029 | 5136651994 | eb79481bc1696c63273b6c2ca22b5e34f68d0208 | No |
+| #2021 | 5134455762 | 4dec9c6a2e03d2ff3e62b186a47fe99910cfcd63 | No |
+| #2019 | 5134382662 | 100f1e1b0554b48ba2f843c965d414993cf5588f | Yes |
+| #2020 | 5134381479 | 29d9d5ec6a02f5c23b315b70f8315a582dac5d1b | No |
+| #2016 | 5134310293 | 5baffac6fdb754e6444c4dfad6a56e4af89d9fe7 | Yes |
+
+All five described a gate limitation and contained Mermaid, but none supplied
+a behavioral summary, a concrete file responsibility beyond its category, or
+a hypothesis/counterexample/evidence chain. Their diagrams were generic file
+inventories. These observations do not establish an outage cause.
+
+The [#2029 review](https://github.com/ContextualWisdomLab/.github/pull/2029#pullrequestreview-5136651994)
+claimed that OpenCode reviewed the product diff, then displayed two generated
+file maps. `build_fallback_review` created the first map and completion claim;
+`publish_fallback_diff_review` appended another completion claim, and the
+common publisher appended its own map. File classification is not source review.
+
+## Repair and measurement
+
+Label the fallback as an inventory with a next action, remove its duplicate
+diagram, and stop the caller from appending a completed-review claim. Public
+symbols read from supplied files are identified as such, not as changed APIs.
+The common publisher retains its evidence map and gate handling. No inference
+result, approval, or finding is synthesized.
+
+English/Korean regression baseline: 0/2 outputs correctly identify the fallback
+limit on main `7fd571dbcdbae6acf29d8f4ee704d7ba6297e4db`. Reproduce with:
+
+```sh
+uv run --no-project .venv/bin/python -m pytest tests/test_opencode_review_surfaces.py tests/test_opencode_review_comment_helpers.py --cov=scripts.ci.opencode_review_surfaces --cov-report=term-missing --cov-fail-under=100 -q
+```
+
+After repair: 2/2 language cases pass; all 43 surface/helper tests pass with
+100% statement and branch coverage of `opencode_review_surfaces`. Docstrings
+also reach 100%. Actionlint passes with its optional ShellCheck/Pyflakes
+integrations disabled; the full invocation was interrupted after remaining
+silent for several minutes. It is not recorded as a full lint pass.
+
+## Model availability dependency
+
+[Noema run 34128415567](https://github.com/ContextualWisdomLab/.github/actions/runs/34128415567/job/101776606311)
+failed after 1,976.2 seconds with one caller attempt and terminal HTTP 502.
+Its [sidecar artifact](https://github.com/ContextualWisdomLab/.github/actions/runs/34128415567/artifacts/10025654264)
+records `provider_connection_error`, 26 review-phase attempt log entries,
+including 19 timeouts and three HTTP errors. Six routes passing preflight
+did not establish success for the real review request.
+
+The installed owner revision was `414f22973658c4ddc3d4320fcf7acd9b4e8ba991`,
+also the consumer pin observed during this investigation.
+[ContextualWisdomLab/contextual-orchestrator#1094](https://github.com/ContextualWisdomLab/contextual-orchestrator/pull/1094)
+already contains a candidate for structured-response synthesizer failover on
+retryable 502/429/timeout errors. At head
+`1c61eff2da012382255bf8b4e1aa6dd0d6dd05ca`, it remains open with six nonterminal
+checks, nine unresolved threads, and no exact-head formal review.
+Triage that owner candidate before duplicating repair in a consumer. Its scope
+matches the failure, but this historical log does not prove the fix works.
+Adoption requires protected owner merge, immutable release, consumer contract
+verification, and a real Noema replay. Do not add caller retries or paid routes.
+
+This repair improves truthful publication, not model availability. A subsequent
+real model review must still demonstrate behavioral explanation, source-bound
+findings, counterexamples, verification limits, and useful next actions. Use the
+[Greptile reference](https://github.com/BerriAI/litellm/pull/20421#issuecomment-3847832704)
+for concrete behavior/flow explanations, not an unsupported numerical score.
+
+## References
+
+GitHub. (n.d.). *REST API endpoints for pull request reviews*. Retrieved
+September 8, 2026, from https://docs.github.com/en/rest/pulls/reviews
+
+Greptile. (2026, February 4). *Greptile overview* [Pull request comment]. GitHub.
+https://github.com/BerriAI/litellm/pull/20421#issuecomment-3847832704

docs/product-technical-gap-baseline.md

@@ -1,8 +1,19 @@
 # Product and Technical Gap Baseline
 
-작성 기준일: **2026-08-26 10:35 KST**
+## 2026-09-08 review-quality follow-up — fallback output and availability
+
+Proposed: fallback inventories no longer claim completed model review and no
+longer generate a duplicate diagram. English/Korean evidence-limit cases improve
+from 0/2 to 2/2; 43 surface/helper tests pass with 100% module statement/branch
+coverage. This does not establish real model review quality or deployment.
+The observed Noema HTTP 502 failure points to the already-open owner candidate
+ContextualWisdomLab/contextual-orchestrator#1094; triage its exact-head review
+threads before immutable adoption and live replay. See
+[the bounded sample, RCA, and validation limits](doctoring/review-fallback-quality-boundary.md).
+
+작성 기준일: **2026-09-08 13:51 KST**
 대상: **ContextualWisdomLab/.github** 중앙 거버넌스·자동화 레포지터리와 이를 소비하는 naruon 생태계
-현재 보호된 `main`: `826b92394c63deb6981c3a8d16a724d71f85a0d7`
+현재 보호된 `main`: `7fd571dbcdbae6acf29d8f4ee704d7ba6297e4db`
 현재 열린 PR 수: **107** (아래 표에 이 스냅샷의 전체 목록 포함; live API 재수집)
 
 이 문서는 제품·기술·운영 Gap을 현재 문서와 현재 GitHub 상태에 묶어 두는 기준선이다. 새 작업은 먼저 이 문서의 Gap ID를 PR 설명과 테스트 증거에 연결하고, PR의 정확한 exact HEAD·Checks·리뷰를 다시 수집한 뒤 구현한다. 표의 상태는 작성 시점의 관측값이므로, 병합 판단에는 재사용하지 않는다. 이 인벤토리는 스냅샷이며 merge authorization이 아니다.

scripts/ci/opencode_review_surfaces.py

@@ -514,24 +514,28 @@ def build_fallback_review(
     language: str = "english",
     coverage_result: str = "success",
 ) -> str:
-    """Build a source-backed formal review of the actual changed product files."""
+    """List supplied changed files without claiming a completed model review."""
     paths = [posix_path(path) for path in changed_files if str(path).strip()]
     korean = _language(language) == "korean"
     overview = "Pull request overview" if not korean else "Pull request 개요"
     walkthrough = "Changed files" if not korean else "변경 파일"
-    diagram = "Changed behavior" if not korean else "변경 동작"
     findings = "Findings" if not korean else "발견 사항"
     intro = (
-        "OpenCode reviewed the current-head product diff. Coverage is a separate gate."
+        "This fallback is a changed-file inventory, not a completed model review. "
+        "Inspect the workflow failure and rerun the review after resolving it. "
+        "File categories and suggested checks below are not execution evidence."
         if not korean
-        else "OpenCode가 현재 head의 제품 diff를 리뷰했습니다. 커버리지는 별도 게이트입니다."
+        else "이 대체 출력은 변경 파일 목록이며, 모델 리뷰가 완료됐다는 뜻은 아닙니다. "
+        "워크플로 실패 원인을 해결한 뒤 리뷰를 다시 실행하세요. "
+        "아래 파일 분류와 권장 검사는 실행 증거가 아닙니다."
     )
     if not paths:
-        intro = (
+        empty_path_notice = (
             "OpenCode could not list changed product files for this head."
             if not korean
             else "OpenCode가 이 head의 변경 제품 파일을 나열하지 못했습니다."
         )
+        intro = f"{empty_path_notice} {intro}"
     lines = [
         f"## {overview}",
         "",
@@ -544,10 +548,10 @@ def build_fallback_review(
         lines.extend(f"- {_file_role(path)}" for path in paths)
     else:
         lines.append("- No changed product files were supplied to the fallback review.")
-    lines.extend(["", f"## {diagram}", "", emit_mermaid(paths, source_root=source_root).rstrip(), ""])
+    lines.append("")
     symbols = rust_api_symbols(source_root, paths)
     if symbols:
-        api_heading = "Changed API" if not korean else "변경 API"
+        api_heading = "Public symbols in supplied files" if not korean else "제공된 파일의 공개 심볼"
         lines.extend([f"## {api_heading}", ""])
         lines.extend(f"- `{symbol}`" for symbol in symbols)
         lines.append("")

tests/test_opencode_agent_contract.py

@@ -1994,7 +1994,8 @@ def test_workflow_provisions_sandbox_tool_and_reviewer_agent():
     assert "needs.coverage-evidence.result != 'cancelled'" in workflow
     assert "request_changes_for_coverage_evidence_failure" in workflow
     assert "implementation_completeness_scan.py" in workflow
-    assert '"## Review outcome"' in workflow
+    assert "opencode_review_surfaces.py build-fallback-review" in workflow
+    assert "This body reviews the changed product files." not in workflow
     assert '"## Check outcome"' not in workflow
     assert 'update_review_overview "COVERAGE_BLOCKED"' in workflow
     assert "record coverage-evidence blocker states" in workflow

tests/test_opencode_review_surfaces.py

@@ -360,23 +360,64 @@ def test_central_workflow_in_diff_is_a_workflow_surface_not_line_one_finding() -
     )
 
 
-def test_fallback_review_empty_file_list() -> None:
-    """Missing changed-file evidence still produces a distinct review body."""
+@pytest.mark.parametrize("language,notice", [
+    ("english", "This fallback is a changed-file inventory, not a completed model review."),
+    ("korean", "이 대체 출력은 변경 파일 목록이며, 모델 리뷰가 완료됐다는 뜻은 아닙니다."),
+])
+def test_fallback_labels_its_evidence_limit(language: str, notice: str) -> None:
+    """A file inventory cannot claim completed review or observed behavior."""
+    review = surfaces.build_fallback_review(
+        changed_files=["src/example.py"], head_sha=HEAD,
+        run_id="1", run_attempt="1", language=language,
+    )
+    assert notice in review
+    assert "reviewed the current-head product diff" not in review
+    assert "제품 diff를 리뷰했습니다" not in review
+    assert "## Changed behavior" not in review
+    assert "## 변경 동작" not in review
+    assert "```mermaid" not in review  # The publisher appends its single evidence map.
+
+
+@pytest.mark.parametrize(
+    ("language", "notice", "action"),
+    (
+        (
+            "english",
+            "not a completed model review",
+            "Inspect the workflow failure and rerun the review after resolving it.",
+        ),
+        (
+            "korean",
+            "모델 리뷰가 완료됐다는 뜻은 아닙니다.",
+            "워크플로 실패 원인을 해결한 뒤 리뷰를 다시 실행하세요.",
+        ),
+    ),
+)
+def test_fallback_review_empty_file_list(
+    language: str, notice: str, action: str
+) -> None:
+    """Missing changed-file evidence keeps the fallback boundary and next action."""
     review = surfaces.build_fallback_review(
         changed_files=[],
         head_sha=HEAD,
         run_id="1",
         run_attempt="1",
+        language=language,
+    )
+    assert notice in review
+    assert action in review
+    assert (
+        "No changed product files" in review
+        or "변경 제품 파일을 나열하지 못했습니다" in review
     )
-    assert "No changed product files" in review
     assert ".github/workflows/opencode-review.yml:1" not in review
 
 
 def test_fallback_review_rejects_accidental_central_workflow_citation(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
     """A synthesized body may not mention the central workflow unless it changed."""
-    monkeypatch.setattr(surfaces, "CENTRAL_WORKFLOW_ANCHOR", "Coverage is a separate gate")
+    monkeypatch.setattr(surfaces, "_file_role", lambda path: surfaces.CENTRAL_WORKFLOW_ANCHOR)
     with pytest.raises(ValueError, match="must not cite"):
         surfaces.build_fallback_review(
             changed_files=ORIGINWEAVE_47_FILES,
@@ -466,7 +507,7 @@ def test_surfaces_cover_remaining_review_branches(tmp_path: Path) -> None:
         source_root=tmp_path,
         language="korean",
     )
-    assert "변경 API" in review
+    assert "제공된 파일의 공개 심볼" in review
     assert "`Once`" in review
 
 
@@ -793,6 +834,7 @@ def test_publisher_workflow_cannot_replace_review_with_coverage_finding(
     fallback_fn = fallback_fn.split("request_changes_for_coverage_evidence_failure()", 1)[0]
     assert "create_pull_review" in fallback_fn
     assert "request_changes_for_coverage_evidence_failure" in fallback_fn
+    assert "This body reviews the changed product files" not in fallback_fn
     assert fallback_fn.index("create_pull_review") < fallback_fn.index(
         "request_changes_for_coverage_evidence_failure"
     )

tests/test_pr_review_autofix_nvidia_nim_contract.py

@@ -17,7 +17,7 @@
 DOCTORING_RECORD = Path("docs/doctoring/hourly-nvidia-nim-autofix.md")
 CHANGELOG = Path("CHANGELOG.md")
 REVIEW_DISPATCH_WORKFLOW = Path(".github/workflows/opencode-review-dispatch.yml")
-REVIEW_DISPATCH_BLOB_SHA = "d86497b3f43bebbabbb4f504eb5132cdf3b7b293"
+REVIEW_DISPATCH_BLOB_SHA = "1004b82a336fd455d995233fdb437249280886fc"
 
 
 def _workflow_text(path: Path) -> str: