fix(strix): retry exact model-quality warnings

CHANGELOG.md

@@ -35,6 +35,9 @@ Semantic Versioning where the repository publishes a release.
 
 ### Fixed
 
+- Retry a Strix provider model that emits its exact model-quality warning via
+  the configured fallback sequence, while retaining fail-closed handling for
+  all other warning, timeout, provider, and vulnerability signals.
 - Parsed `opencode.jsonc` as JSONC (stripping `//` and `/* */` comments outside string literals) in the reasoning-effort guard and its contract tests, instead of raw `json.loads`, which rejected the file the moment it carried its first explanatory comment (added for the `contextual-orchestrator` provider block) with `Expecting property name enclosed in double quotes`. Comment markers inside string values, such as the `$schema` URL, are left untouched.
 - Download the pinned `uv` 0.12.1 exporter from the official GitHub Releases URL instead of `releases.astral.sh`, which now returns HTTP 403 and blocks org-wide OpenCode `coverage-evidence`. The SHA-256 pin is unchanged. The opener may follow one hop onto `release-assets.githubusercontent.com` or `objects.githubusercontent.com` and still rejects every other host, userinfo, non-HTTPS scheme, and nondefault port (ContextualWisdomLab/.github#1109).
 - Compared the trusted `uv` executable's post-install `--version` output against the real GitHub Releases build's full string, `uv 0.12.1 (x86_64-unknown-linux-gnu)`, instead of the bare `uv 0.12.1` the prior check required; the genuine release binary always prints the target triple, so every installation was failing the pin check immediately after the archive download itself was fixed (ContextualWisdomLab/.github#1109).

docs/doctoring/strix-quality-warning-fallback.md

@@ -0,0 +1,49 @@
+# Strix model-quality warning fallback
+
+검토 기준일: **2026-08-21**
+
+## Incident
+
+The central Strix workflow selected the public-repository NVIDIA NIM model
+`nvidia/nemotron-3-super-120b-a12b`. Strix completed a scan and produced only a
+`LOW` finding, but it also emitted `MODEL QUALITY WARNING` because the selected
+model was not a recommended frontier model. The gate correctly classified the
+warning as non-clean evidence, then stopped before trying the configured
+fallback models. This left an actionable low-severity report indistinguishable
+from an unrecoverable provider failure and blocked the target pull request.
+
+## Decision
+
+`scripts/ci/strix_quick_gate.sh` recognizes only Strix's exact model-quality
+warning (`MODEL QUALITY WARNING` / `is not a recommended frontier model for
+Strix`) as retryable model evidence. It remains an infrastructure/failure
+signal, so the scan never passes merely because the warning was seen. The
+existing fallback sequence must obtain a clean result or the gate fails closed.
+
+All other `Warn`, `Warning`, `Fatal`, `Denied`, and `Timeout` output remains a
+hard failure. A `MEDIUM` or higher vulnerability remains blocking even when a
+fallback succeeds; below-threshold findings are handled by the existing
+`STRIX_FAIL_ON_MIN_SEVERITY` policy.
+
+## Verification contract
+
+`tests/test_strix_nvidia_nim_not_found_fallback.py` executes the production
+classifier against a bounded quality-warning log and asserts that the warning
+is wired into `is_model_retryable_error`. The existing shell harness continues
+to cover generic warning signals and fallback failure paths.
+
+## Rollback
+
+If a future Strix release changes the warning wording, add the exact new
+provider-produced wording to the narrow classifier and its regression test.
+Do not remove generic warning failure handling or neutralize the entire warning
+class.
+
+## References (APA 7th)
+
+GitHub. (n.d.). *Workflow syntax for GitHub Actions*. GitHub Docs. Retrieved
+August 21, 2026, from
+https://docs.github.com/en/actions/reference/workflows-and-actions/workflow-syntax
+
+GitHub. (n.d.). *Using workflow run logs*. GitHub Docs. Retrieved August 21,
+2026, from https://docs.github.com/en/actions/how-tos/monitor-workflows/use-workflow-run-logs

scripts/ci/strix_quick_gate.sh

@@ -2944,6 +2944,10 @@ is_llm_token_limit_error() {
 # was interrupted or incomplete.  Used as a guard to prevent the
 # below-threshold override from silently passing an aborted scan.
 has_detected_infrastructure_error() {
+	if is_model_quality_warning; then
+		return 0
+	fi
+
 	if grep -Eiq '(^|[^[:alpha:]])(Fatal|Denied|Warn|Warning)([^[:alpha:]]|$)' "$STRIX_LOG"; then
 		return 0
 	fi
@@ -2990,6 +2994,10 @@ has_detected_infrastructure_error() {
 	return 1
 }
 
+is_model_quality_warning() {
+	grep -Eiq 'MODEL QUALITY WARNING|is not a recommended[[:space:]]+frontier model for Strix' "$STRIX_LOG"
+}
+
 latest_strix_report_dir() {
 	local latest=""
 	local run_dir
@@ -3818,6 +3826,10 @@ is_hallucinated_source_claim_finding() {
 is_model_retryable_error() {
 	local model="$1"
 
+	if is_model_quality_warning; then
+		return 0
+	fi
+
 	if is_vertex_model "$model" && is_vertex_not_found_error; then
 		return 0
 	fi

tests/test_strix_nvidia_nim_not_found_fallback.py

@@ -171,6 +171,41 @@ def test_not_found_skips_same_model_and_enters_cross_model_fallback(self) -> Non
         self.assertIn("is_nvidia_nim_not_found_error", retryable)
         self.assertNotIn("is_nvidia_nim_not_found_error", same_model_retry)
 
+    def test_model_quality_warning_enters_configured_fallback(self) -> None:
+        """Retry a weaker provider model instead of treating its warning as clean evidence."""
+
+        gate_source = STRIX_GATE.read_text(encoding="utf-8")
+        quality_warning = _function_block(
+            gate_source,
+            "is_model_quality_warning",
+        )
+        retryable = _function_block(gate_source, "is_model_retryable_error")
+        self.assertIn("is_model_quality_warning", retryable)
+
+        with tempfile.TemporaryDirectory(prefix="strix-quality-warning-") as temp_dir:
+            log_path = Path(temp_dir) / "strix.log"
+            log_path.write_text(
+                "MODEL QUALITY WARNING\n"
+                "'nvidia_nim/example' is not a recommended frontier model for Strix.\n",
+                encoding="utf-8",
+            )
+            script = "\n".join(
+                (
+                    "set -euo pipefail",
+                    'STRIX_LOG="$1"',
+                    quality_warning,
+                    "is_model_quality_warning",
+                )
+            )
+            completed = subprocess.run(
+                ["bash", "-c", script, "strix-quality-classifier", str(log_path)],
+                check=False,
+                capture_output=True,
+                text=True,
+            )
+
+        self.assertEqual(completed.returncode, 0, completed.stderr)
+
     def test_workflow_uses_available_free_first_nvidia_plan(self) -> None:
         """Prefer a documented hosted NIM and another NIM before GitHub."""