valent-pipeline 0.19.62 → 0.19.64

package/bin/cli.js

@@ -760,6 +760,7 @@ traceCmd
   .option('--reqs-manifest <path>', 'AC manifest (defaults to stories/<id>/output/reqs-brief.manifest.json)')
   .option('--spec-manifest <path>', 'Spec manifest (defaults to stories/<id>/output/qa-test-spec.manifest.json)')
   .option('--label <name>', 'Evidence-run label whose junit results prove execution (default qa-b-tests)')
+  .option('--run <runId>', 'Bind to an exact evidence run id (overrides label resolution — use when several suites multiplex one label)')
   .action(async (options) => {
     const { traceCheckCmd } = await import('../src/commands/trace.js');
     await traceCheckCmd(options);
@@ -773,6 +774,7 @@ traceCmd
   .option('--reqs-manifest <path>', 'AC manifest path override')
   .option('--spec-manifest <path>', 'Spec manifest path override')
   .option('--label <name>', 'Evidence-run label for junit results (default qa-b-tests)')
+  .option('--run <runId>', 'Bind to an exact evidence run id (overrides label resolution — use when several suites multiplex one label)')
   .option('--out <path>', 'Output path (defaults to stories/<id>/output/traceability-matrix.generated.md)')
   .action(async (options) => {
     const { traceMatrixCmd } = await import('../src/commands/trace.js');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "valent-pipeline",
-  "version": "0.19.62",
+  "version": "0.19.64",
   "description": "v3 multi-agent AI pipeline for software development lifecycle",
   "type": "module",
   "bin": {

package/pipeline/orchestrators/claude-code/sprint.workflow.js

@@ -1951,7 +1951,7 @@ async function runStory(story) {
     // reworks via the QA-B re-execution alone, then falls through to JUDGE at the cap as usual.
     const evGate = await runEvidenceGate(storyId, [])
     recordStage('evidence', evGate)
-    const visual = await runVisualGate(storyId, profiles)
+    const visual = await runVisualGate(storyId, profiles, itemType)
     recordStage('visual', visual)
     phase('Judge')
     const decision = await runJudgeGate(storyId, visual, evGate)
@@ -2234,8 +2234,9 @@ async function runStory(story) {
   recordStage('evidence', evGate)
 
   // Visual validation (revived PMCP): for ui-profile stories, produce pmcp-evidence.md so JUDGE's
-  // evidence pass has the visual artifact it requires. No-op (returns null) for non-ui stories.
-  const visual = await runVisualGate(storyId, profiles)
+  // evidence pass has the visual artifact it requires. No-op (returns null) for non-ui stories, and a
+  // non-blocking N/A for itemType:bug (the QA-A-less bug-fix path mints no checklist — M-192).
+  const visual = await runVisualGate(storyId, profiles, itemType)
   recordStage('visual', visual)
 
   phase('Judge')
@@ -2826,11 +2827,23 @@ async function runStory(story) {
   // exact trigger QA-A uses to produce the checklist (qa-a/write-spec.md Step 9: `ui` in profiles).
   // The agent starts its own dev server: unlike the old always-on topology where QA-B triggered an
   // idle PMCP while QA-B's server was up, here the stages are sequential, so PMCP owns its server.
-  async function runVisualGate(sid, storyProfiles) {
+  async function runVisualGate(sid, storyProfiles, sidItemType = 'story') {
     if (!storyProfiles.includes('ui')) {
       log(`${sid}: no ui profile — visual validation N/A (QA-A wrote no checklist; nothing for PMCP to run)`)
       return null
     }
+    // QA-A-less bug-fix path (owner-ruled M-192): a bug-fix item runs bug-intake -> Build -> CRITIC ->
+    // QA-B -> Judge and STRUCTURALLY SKIPS the Spec phase, so QA-A never mints a
+    // visual-validation-checklist.md. Running PMCP against a checklist that can NEVER exist yields a
+    // 0/0 block-fail (onMissingMcp inherited from the parent story) on a path with no checklist to
+    // drive — a config contradiction, not a real visual defect. Treat it as a non-blocking N/A: no
+    // visual coverage is lost because the parent story's `revalidate` re-drives the FULL visual gate
+    // on the merged fix (its checklist exists). A FULL story always runs QA-A, so a missing checklist
+    // THERE stays a real, blocking fault — this scopes the gate down ONLY for itemType:bug.
+    if (sidItemType === 'bug') {
+      log(`${sid}: ui profile but itemType:bug — the QA-A-less bug-fix path mints no visual-validation-checklist.md; visual validation N/A (owner-ruled M-192). The parent story's revalidate re-drives the full visual gate on the merged fix.`)
+      return { schema: 1, agent: 'pmcp', story: sid, verdict: 'na', skipped: true, mcpAvailable: null, totalCheckpoints: 0, passed: 0, failed: 0, naCheckpoints: 0, files: [], notes: 'QA-A-less bug-fix path: no visual-validation-checklist.md minted (Spec phase skipped for itemType:bug); visual gate N/A (owner-ruled M-192). The parent story revalidate re-drives the full visual gate on the merged fix.' }
+    }
     phase('Visual')
     const naMode = visualOnMissingMcp === 'na'
     const result = await spawn('PMCP', 'pmcp.md',
@@ -2920,7 +2933,13 @@ async function runStory(story) {
       : ''
     // A ui story has a PMCP visual pass; tell the evidence agent the result so Step 9 (Verify PMCP
     // Evidence) is grounded rather than guessing whether visual evidence should exist.
-    const visualNote = visual
+    const visualNote = visual && visual.skipped
+      ? `\n\nThis is a UI-profile BUG-FIX item: the QA-A-less bug-fix path mints no ` +
+        `visual-validation-checklist.md, so PMCP did NOT run and there is NO pmcp-evidence.md. Treat ` +
+        `PMCP evidence as N/A in Step 9 (no checklist ⟹ visual N/A) — do NOT flag its absence as a ` +
+        `missing required file (owner-ruled M-192). The parent story's revalidate re-drives the full ` +
+        `visual gate on the merged fix.`
+      : visual
       ? `\n\nThis is a UI story: a visual-validation (PMCP) pass already ran and wrote ` +
         `pmcp-evidence.md (verdict: ${visual.verdict}, ${visual.passed || 0}/${visual.totalCheckpoints || 0} ` +
         `checkpoint(s) passed${visual.mcpAvailable === false ? `, browser MCP unavailable` : ''}). ` +

package/pipeline/steps/judge/evidence-review.md

@@ -6,7 +6,7 @@ Runs after bug review (Steps 1-4) approves with no P4->P1-P3 escalations.
 
 Read all inputs, record in `inputsRead`. If any REQUIRED file missing, set blocker and STOP.
 
-Required: `execution-report.md`, `traceability-matrix.md`, `bugs.md`, `qa-test-spec.md` — all in `{story_output_dir}/`. Also `pmcp-evidence.md` **if this is a UI story** — i.e. `ui` is in `{testing_profiles}` (equivalently, `visual-validation-checklist.md` exists). If there is no `ui` profile, PMCP evidence is N/A — do NOT treat its absence as a missing required file.
+Required: `execution-report.md`, `traceability-matrix.md`, `bugs.md`, `qa-test-spec.md` — all in `{story_output_dir}/`. Also `pmcp-evidence.md` **if this is a UI story** — i.e. `ui` is in `{testing_profiles}` **AND** `visual-validation-checklist.md` exists. The checklist's existence is the operative trigger: a `ui`-profile item with **no** `visual-validation-checklist.md` has PMCP **N/A** — do NOT treat the absence of `pmcp-evidence.md` as a missing required file. This is exactly the **QA-A-less bug-fix path** (an `itemType:bug` fix runs bug-intake → Build → CRITIC → QA-B → Judge and structurally skips the Spec phase, so QA-A mints no checklist); its visual coverage is re-driven in full by the parent story's `revalidate` on the merged fix (owner-ruled M-192). If there is no `ui` profile at all, PMCP evidence is likewise N/A.
 
 **Machine evidence (when `stories/{story_id}/evidence/` exists — required in that case):** the newest JSON under `evidence/asserts/` (the deterministic green-gate's verdict), `evidence/pins.jsonl` (which commit each gate reviewed), and the run records under `evidence/runs/` (frozen junit reports + hashes, written only by the CLI). **ATDD stories add** `evidence/atdd-red.json` (the pre-implementation failing run: per-case statuses, the pre-dev SHA, the acceptance-source snapshot hashes) and — for a proven story — `evidence/proof.json` (red run + green run + the CLI-hashed diff between: the story's headline evidence). These are the authoritative numbers — the markdown artifacts are the human-readable companion. If the evidence directory does not exist, this is a LEGACY story: apply the legacy fallbacks below and flag every count as self-reported.
 

package/src/commands/trace.js

@@ -9,9 +9,21 @@
 import { readFileSync, existsSync, mkdirSync, writeFileSync } from 'fs';
 import { join, isAbsolute, dirname } from 'path';
 import { traceCheck, traceMatrix } from '../lib/trace.js';
-import { evidenceDir, latestRun, resolveRoot } from '../lib/evidence.js';
+import { evidenceDir, resolveRunForAssert, requiredCaseIdsForRun, resolveRoot } from '../lib/evidence.js';
 import { parseJunit } from '../lib/junit.js';
 
+/** Load an exact evidence run by id (for `--run`), or null if absent/unreadable. */
+function loadRunById(evDir, runId) {
+  const dir = join(evDir, 'runs', runId);
+  const recPath = join(dir, 'evidence.json');
+  if (!existsSync(recPath)) return null;
+  try {
+    return { runId, dir, record: JSON.parse(readFileSync(recPath, 'utf-8')) };
+  } catch {
+    return null;
+  }
+}
+
 function readJson(path, what) {
   try {
     return JSON.parse(readFileSync(path, 'utf-8'));
@@ -41,11 +53,24 @@ function loadInputs(options) {
     return { noop: 'spec manifest predates the per-case `ac` field — skipping (no-op pass). Re-run QA-A to gain mechanical AC coverage.' };
   }
 
-  // Executed testcases from the newest report-bearing evidence run (junit re-parsed from the
-  // frozen copies — never from a transcription).
+  // Executed testcases from the evidence run that ACTUALLY carries this story's required acceptance
+  // cases — NOT merely the newest same-label run. A REVALIDATE (and any multi-suite cycle) re-runs the
+  // full npm suite under the `qa-b-tests` label ALONGSIDE the acceptance run; binding to the newest by
+  // label alone globs the full suite (which carries ZERO of the story's required cases) and false-fails
+  // every required case as "unexecuted" — label-multiplexing, the M-11/.45/.60 class, of which the
+  // trace-check was the last unfixed surface (the `evidence assert` + verdict-crosscheck paths already
+  // resolve this way). resolveRunForAssert binds to the same-SHA sibling carrying a required case id and
+  // preserves the #5 crash-loud invariant (a report-less newest is never silently skipped). An explicit
+  // --run pins exactly, for the operator who wants to override resolution. (junit re-parsed from the
+  // frozen copies — never from a transcription.)
   let junitCases = null;
   let runId = null;
-  const run = latestRun(evidenceDir(root, story), { label: options.label || 'qa-b-tests', requireReport: true });
+  const evDir = evidenceDir(root, story);
+  const label = options.label || 'qa-b-tests';
+  const requiredCaseIds = requiredCaseIdsForRun(cases);
+  const run = options.run
+    ? loadRunById(evDir, options.run)
+    : resolveRunForAssert(evDir, { label, requiredCaseIds });
   if (run) {
     runId = run.runId;
     junitCases = [];