fix(scheduler): ignore manual Strix dispatch as merge evidence

ARCHITECTURE.md

@@ -104,13 +104,17 @@ sequenceDiagram
   OC->>SV: PoC command in isolated copy
   SV-->>OC: redacted stdout/stderr + command metadata
   OC-->>PR: APPROVE or request changes
-  MS->>PR: merge only on current-head approval + green checks
+  MS->>PR: merge only on current-head approval + required Strix evidence
 ```
 
 ## Trust boundaries
 
 - Required review workflows execute **base-branch** scripts. A PR that edits
   those workflows cannot widen its own `pull_request_target` token.
+- Merge evidence for Strix is `pull_request_target` and
+  `repository_dispatch` `strix-scan` only. A caller-selected
+  `workflow_dispatch` check run (including Deep) must not park, fail, or
+  satisfy the merge scheduler.
 - Reviewer agents stay `edit: deny`. They judge; they do not implement.
 - Sandbox helpers copy the workspace, drop secret environment values unless
   explicitly allowlisted by **name**, and run subprocesses with `shell=False`.
@@ -145,6 +149,8 @@ trusted `uv` exporter is downloaded from the literal GitHub Releases URL for
   contract.
 - [`docs/doctoring/hourly-nvidia-nim-autofix.md`](docs/doctoring/hourly-nvidia-nim-autofix.md)
   — current increment's repair-worker decision and APA 7th citations.
+- [`docs/doctoring/strix-manual-dispatch-not-merge-evidence.md`](docs/doctoring/strix-manual-dispatch-not-merge-evidence.md)
+  — manual Deep `workflow_dispatch` is not merge evidence.
 - [`docs/doctoring/fast-mlsirm-hourly-review-caller.md`](docs/doctoring/fast-mlsirm-hourly-review-caller.md)
   — product-specific psychometric repair heartbeat and scientific gates.
 - [`docs/doctoring/exact-artifact-sbom-attestation.md`](docs/doctoring/exact-artifact-sbom-attestation.md)

CHANGELOG.md

@@ -39,6 +39,8 @@ Semantic Versioning where the repository publishes a release.
 - 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).
 - Excluded relative `-r` and `--requirement` referrers from generated flat base-lock publication while retaining bounded include syntax diagnostics and discovering independently complete direct `.txt` children of `requirements` directories.
+- Ignore caller-selected `workflow_dispatch` Strix check runs in merge evidence so a manual Deep scan cannot park, fail, or satisfy the scheduler for up to six hours. Required evidence remains `pull_request_target` and `repository_dispatch` `strix-scan`; a successful `strix` commit status can still supersede a failed required check.
+- Materialized base Python locks only when every package line is an exact SHA-256 pin or a bounded relative `-r`/`--requirement` include. A lone `--require-hashes` directive, a dotted include such as `./lock.txt`, or `-r other-hashes.txt` no longer enters the trusted build context.
 - Refused a conflict-scope repository root whose immediate parent is a symbolic link, so a swapped parent cannot redirect the canonical worktree after the last-component check (CWE-367).
 - Materialized base Python locks only when every package line is an exact SHA-256 pin or a bounded relative `-r`/`--requirement` include. A lone `--require-hashes` directive, a dotted include such as `./lock.txt`, or `-r other-hashes.txt` no longer enters the trusted build context.
 - Bounded the Strix quality self-test's deterministic timeout fixtures to 3-second process and 5-second fake-sleep budgets so exact-head policy evidence completes inside the existing job limit without changing production Strix scanner timeouts, providers, credentials, or review semantics.

CLAUDE.md

@@ -123,8 +123,9 @@ repeatable compile command.
   on the worker, never `COPILOT_GITHUB_TOKEN`.
 - **`pull_request_target` trust boundary.** The required review workflows run the *base branch's*
   trusted scripts. A PR that edits the trusted review workflows can fail its own checks until the
-  base branch catches up; a same-head manual `workflow_dispatch` Strix run may supply review evidence
-  but does not replace required PR checks. Do not widen a `pull_request_target` job token to
+  base branch catches up; a same-head manual `workflow_dispatch` Strix run may
+  inform a reviewer but must not park, fail, or satisfy merge evidence, and it
+  does not replace required PR checks. Do not widen a `pull_request_target` job token to
   repository-write permission.
 - **Review output must go through the Python normalizer** (`scripts/ci/opencode_review_normalize_output.py`)
   — it escapes `<`, `>`, `&` when embedding JSON in HTML comments to prevent Markdown-comment

README.md

@@ -45,6 +45,26 @@ reconciliation do not require naruon to be present, imported, or running.
 Every CWL component is standalone and also composable. For this repository
 that means:
 
+
+- `naruon`: approved PRs can become `BEHIND`; the scheduler treats that as an
+  update request, not as a merge signal. GitHub Actions updates the branch with
+  `expected_head_sha`, then the new head is reviewed again.
+- `pg-erd-cloud`: successful bot merges used current-head evidence and
+  `--match-head-commit`; the centralized path keeps that head-SHA guard.
+- `.github`: PRs that edit trusted review workflows can fail because
+  `pull_request_target` runs the base branch's trusted scripts. A same-head
+  manual `workflow_dispatch` Strix run may inform a reviewer, but it does not
+  park, fail, or satisfy merge evidence. Required PR checks stay
+  `pull_request_target` and `repository_dispatch` `strix-scan` until the
+  trusted base branch catches up.
+- `naruon#745`: new OpenCode review-flow work improves Mermaid output by
+  replacing generic risk sketches with changed-file flow DAGs. The central
+  workflow carries that review contract while keeping the self-test drift fix.
+- Cross-repo DX/UX: helpful sibling-repo patterns should be adopted when they
+  reduce maintainer, reviewer, CI-operator, contributor, user, or reader
+  friction. Noisy automation, repeated waiting, false failures, misleading
+  statuses, and URL-only diagnostics are treated as review-experience defects.
+
 | Mode | What happens |
 | --- | --- |
 | **따로 (this repo alone)** | Clone, test, and operate `.github` as the org profile and workflow source. Local quality gates, Cloudflare dry-run, and this repository's own PRs do not depend on naruon or any sibling product checkout. |

docs/doctoring/strix-manual-dispatch-not-merge-evidence.md

@@ -0,0 +1,78 @@
+# Manual Strix `workflow_dispatch` is not merge evidence
+
+검토 기준일: **2026-08-16**
+
+## Incident
+
+The merge scheduler treated every `Strix Security Scan` / `strix` check run as
+required current-head evidence. A caller-selected `workflow_dispatch` run
+publishes that same check name. Official Deep mode can occupy the GitHub-hosted
+360-minute ceiling. An in-progress or failed Deep job on a pull-request head
+therefore parked `strix_evidence_state()`, could replace a later
+`failed_status_checks()` winner for the same workflow/name key, and could
+suppress `repository_dispatch` `strix-scan` retry when the scheduler was not in
+centralized-dispatch mode.
+
+GitHub's Actions UI and `gh workflow run --ref` let a writer choose the
+workflow revision (GitHub, n.d.-a). That revision supplies the workflow
+definition before any trusted-source checkout. Manual Deep remains a reviewer
+tool, not a merge gate.
+
+## Decision
+
+Required Strix merge evidence is only:
+
+- `pull_request_target` check runs, and
+- `repository_dispatch` type `strix-scan` check runs.
+
+The scheduler now:
+
+1. reads `checkSuite.workflowRun.event` on each check run;
+2. ignores `workflow_dispatch` check runs in `is_strix_context`,
+   `strix_evidence_state`, `failed_status_checks`, `action_required_checks`,
+   and `matching_actions_job_id`;
+3. treats a completed required Strix check run as complete even when a later
+   pending `strix` commit status exists;
+4. skips `workflow_dispatch` workflow runs in non-central `active_review_run_refs`
+   so a same-head Deep job cannot return `already_running` and block
+   `strix-scan`.
+
+A successful `strix` commit status can still supersede a failed required check
+run. That path informs a reviewer and remains the documented exception for
+base-branch catch-up. It does not let a Deep check run park merge.
+
+Do not fold this change into ContextualWisdomLab/.github#1054. That pull
+request owns the official `quick|standard|deep` mapping. This record owns
+merge-evidence isolation so #1054 can land without a six-hour merge stall.
+
+## Verification contract
+
+`tests/test_pr_review_merge_scheduler.py` fails if:
+
+1. `PULL_REQUEST_FIELDS_FRAGMENT` drops `workflowRun.event`;
+2. a running or failed `workflow_dispatch` Strix check run changes
+   `strix_evidence_state` or `failed_status_checks` when a required check run
+   is present;
+3. a same-head `workflow_dispatch` workflow run is classified as current
+   evidence in non-central `active_review_run_refs`;
+4. `matching_actions_job_id` returns a Deep job over a required job.
+
+## Rollback
+
+Roll back only if a required `pull_request_target` or `strix-scan` check run is
+observed ignored because GitHub omitted `event` and the run was mis-labeled
+`workflow_dispatch`. Missing `event` stays required evidence.
+
+## References (APA 7th)
+
+GitHub. (n.d.-a). *Manually running a workflow*. GitHub Docs. Retrieved
+August 16, 2026, from
+https://docs.github.com/en/actions/how-tos/manage-workflow-runs/manually-run-a-workflow
+
+GitHub. (n.d.-b). *Workflow syntax for GitHub Actions*. GitHub Docs. Retrieved
+August 16, 2026, from
+https://docs.github.com/en/actions/reference/workflows-and-actions/workflow-syntax
+
+GitHub. (n.d.-c). *Objects: WorkflowRun* (`event`). GitHub GraphQL API.
+Retrieved August 16, 2026, from
+https://docs.github.com/en/graphql/reference/objects#workflowrun

requirements-pip-audit-ci-hashes.txt

@@ -213,9 +213,9 @@ packaging==26.2 \
     # via
     #   pip-audit
     #   pip-requirements-parser
-pip==26.1.2 \
-    --hash=sha256:382ff9f685ee3bc25864f820aa50505825f10f5458ffff07e30a6d96e5715cab \
-    --hash=sha256:f49cd134c61cf2fd75e0ce2676db03e4054504a5a4986d00f8299ae632dc4605
+pip==26.2.1 \
+    --hash=sha256:71138adf1f4ca900cdb7d289c21b7494329f2332b6d85f0e1c42108c0384ed3e \
+    --hash=sha256:f6ad667e89a1fe78046c8f13232b247200f5258d7828f3f7883d660878e0813f
     # via pip-api
 pip-api==0.0.34 \
     --hash=sha256:8b2d7d7c37f2447373aa2cf8b1f60a2f2b27a84e1e9e0294a3f6ef10eb3ba6bb \

scripts/ci/pr_review_merge_scheduler.py

@@ -73,6 +73,7 @@
           detailsUrl
           checkSuite {
             workflowRun {
+              event
               workflow { name }
             }
           }
@@ -953,9 +954,28 @@ def is_opencode_context(node: dict[str, Any]) -> bool:
     return node.get("context") == "opencode-review"
 
 
+def workflow_run_event(node: dict[str, Any]) -> str:
+    """Return the GitHub Actions event that created this check run, if present."""
+    workflow_run = ((node.get("checkSuite") or {}).get("workflowRun") or {})
+    return str(workflow_run.get("event") or "").strip()
+
+
+def is_manual_workflow_dispatch(node: dict[str, Any]) -> bool:
+    """Return whether a check run was created by caller-selected workflow_dispatch."""
+    return workflow_run_event(node) == "workflow_dispatch"
+
+
 def is_strix_context(node: dict[str, Any]) -> bool:
-    """Return whether a check or status context belongs to Strix evidence."""
+    """Return whether a check or status context is required Strix merge evidence.
+
+    Manual ``workflow_dispatch`` check runs can publish the same job name for
+    up to six hours (Deep). They may inform a reviewer but must not park,
+    fail, or satisfy the merge scheduler. A successful ``strix`` commit status
+    can still supersede a failed required check run.
+    """
     if node.get("__typename") == "CheckRun":
+        if is_manual_workflow_dispatch(node):
+            return False
         workflow = (
             ((node.get("checkSuite") or {}).get("workflowRun") or {}).get("workflow")
             or {}
@@ -1062,17 +1082,31 @@ def opencode_in_progress(pr: dict[str, Any], *, stale_after_minutes: int | None
 
 
 def strix_evidence_state(pr: dict[str, Any]) -> str:
-    """Return missing, running, or complete for current-head Strix evidence."""
+    """Return missing, running, or complete for current-head Strix evidence.
+
+    A completed required Strix check run is enough. A later pending
+    ``strix`` commit status from a manual Deep dispatch must not park merge.
+    """
     found = False
+    completed_required_check_run = False
+    running_required = False
     for node in context_nodes(pr):
         if not is_strix_context(node):
             continue
         found = True
         status = (node.get("status") or node.get("state") or "").upper()
-        if status in RUNNING_CHECK_STATES:
-            return "running"
-        if node.get("__typename") == "CheckRun" and status != "COMPLETED":
-            return "running"
+        is_check_run = node.get("__typename") == "CheckRun"
+        if status in RUNNING_CHECK_STATES or (
+            is_check_run and status != "COMPLETED"
+        ):
+            running_required = True
+            continue
+        if is_check_run:
+            completed_required_check_run = True
+    if completed_required_check_run:
+        return "complete"
+    if running_required:
+        return "running"
     return "complete" if found else "missing"
 
 
@@ -1387,6 +1421,8 @@ def failed_status_checks(pr: dict[str, Any]) -> list[str]:
         if node.get("__typename") != "CheckRun":
             status_contexts.append(node)
             continue
+        if is_manual_workflow_dispatch(node):
+            continue
         workflow = (
             (((node.get("checkSuite") or {}).get("workflowRun") or {}).get("workflow") or {}).get("name")
             or ""
@@ -1435,6 +1471,8 @@ def action_required_checks(pr: dict[str, Any]) -> list[str]:
     for node in context_nodes(pr):
         if node.get("__typename") != "CheckRun":
             continue
+        if is_manual_workflow_dispatch(node):
+            continue
         conclusion = (node.get("conclusion") or "").upper()
         if conclusion in ACTION_REQUIRED_CONCLUSIONS:
             required.append(node.get("name") or "check-run")
@@ -1947,6 +1985,8 @@ def active_review_run_refs(
                 continue
             if centralized_dispatch:
                 continue
+            if run_data.get("event") == "workflow_dispatch":
+                continue
             run_head = str(run_data.get("head_sha") or "").lower()
             pull_requests = run_data.get("pull_requests") or []
             if run_head == head: