fix: address prior PR feedback — CI mutation grep, docs, eval DAG

.github/workflows/ci.yml

@@ -78,7 +78,8 @@ jobs:
       - name: Mutation test (storage_json)
         run: |
           timeout 900 cargo mutants -f '*storage_json*' 2>&1 | tee mutation.log || true
-          MISSED=$(grep -E '[0-9]+ missed' mutation.log | tail -1 | grep -oE '[0-9]+' | head -1 || echo "0")
+          # Extract the missed count (e.g. "14 mutants ... 2 missed, 9 caught" -> 2), not the total
+          MISSED=$(grep -E '[0-9]+ missed' mutation.log | tail -1 | sed -n 's/.* \([0-9][0-9]*\) missed.*/\1/p' | head -1 || echo "0")
           echo "missed=$MISSED" >> "$GITHUB_OUTPUT"
         id: mutation
       - name: Check mutation baseline

README.md

@@ -1002,6 +1002,8 @@ The summary includes:
 
 Contributions are welcome! Please open an issue first to discuss what you would like to change.
 
+**PR workflow:** Open a PR → ensure CI is green (version, lint, security, test, mutation, review) → merge when ready. Use a short test plan in the PR description. Small, focused PRs are preferred.
+
 ### Local Development Checks
 
 Enable the repository-managed git hooks after cloning:

docs/mutation-testing.md

@@ -15,7 +15,7 @@ cargo mutants -f '*storage_pg*'
 cargo mutants -f '*cost*'
 ```
 
-CI runs mutation on `*storage_json*` with a timeout; see [CI job](#ci) below.
+CI runs mutation on `*storage_json*` with a timeout; see [CI job](#ci) below. The `*state*` and `*storage_pg*` globs are not run in CI (too many mutants, longer runtime); run locally when auditing those areas.
 
 ## Known equivalent / accepted mutants
 
@@ -37,6 +37,17 @@ After adding targeted tests (refresh_summary, get_event_stats exact aggregates/s
 
 If new mutants appear in these regions, add assertions that would fail on the wrong operator/formula, or add them to this table with a one-line rationale.
 
+### state.rs and storage_pg.rs (local only)
+
+Mutation on `*state*` and `*storage_pg*` is not in CI. Summary from a local run (interrupted; full run is slow):
+
+| Glob          | Mutants | Notes |
+|---------------|---------|--------|
+| `*state*`     | ~90     | Many missed: arithmetic in cost/age, `load_reviews_from_disk`, `mark_running` / `complete_review` / `fail_review` / `prune_old_reviews` (no-op or wrong operator), `current_timestamp`, `ReviewEventBuilder`. Killing these would need more unit tests or integration tests that assert side effects. |
+| `*storage_pg*`| ~67     | Many missed: `migrate`, `is_empty`, `parse_comment_status`, `save_review` / `get_review` / `list_reviews` / `delete_review` / `save_event` / `list_events` (stubbed return or wrong operator). Would need PostgreSQL-backed tests or contract tests to kill. |
+
+To re-run: `cargo mutants -f '*state*'` and `cargo mutants -f '*storage_pg*'` (allow several minutes each).
+
 ## Pre-push vs CI
 
 - **Pre-push** (`.githooks/pre-push`): Runs unit tests, `cargo audit`, web build+test. Does *not* run mutation (too slow for every push).

src/commands/eval/runner/execute/dag.rs

@@ -188,6 +188,7 @@ fn stage_hints(stage: EvalFixtureStage) -> DagNodeExecutionHints {
             side_effects: false,
             subgraph: Some("review_pipeline".to_string()),
         },
+        // Linear chain (each depends on previous); parallelizable kept for consistency / future DAG shape.
         EvalFixtureStage::ExpectationMatching
         | EvalFixtureStage::CommentCountValidation
         | EvalFixtureStage::BenchmarkMetrics => DagNodeExecutionHints {