wogiflow 2.29.8 → 2.29.9

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "wogiflow",
-  "version": "2.29.8",
+  "version": "2.29.9",
   "description": "AI-powered development workflow management system with multi-model support",
   "main": "lib/index.js",
   "bin": {
@@ -10,7 +10,7 @@
   },
   "scripts": {
     "flow": "./scripts/flow",
-    "test": "NODE_ENV=test node --test tests/auto-compact-prompt.test.js tests/flow-paths.test.js tests/flow-io.test.js tests/flow-audit-gates.test.js tests/flow-standards-hook-three-layer.test.js tests/flow-config-loader.test.js tests/flow-damage-control.test.js tests/flow-output.test.js tests/flow-constants.test.js tests/flow-session-state.test.js tests/flow-hooks-integration.test.js tests/flow-utils.test.js tests/flow-security.test.js tests/flow-memory-db.test.js tests/flow-durable-session.test.js tests/flow-skill-matcher.test.js tests/flow-bridge.test.js tests/flow-proactive-compact.test.js tests/flow-cascade-completion.test.js tests/flow-capture-gate.test.js tests/flow-correction-detector-hybrid.test.js tests/flow-promote.test.js tests/flow-archive-runs.test.js tests/flow-memory.test.js tests/flow-hooks-pre-tool-helpers.test.js tests/flow-hooks-bugfix-scope-gate.test.js tests/flow-hooks-routing-gate.test.js tests/flow-hooks-phase-read-gate.test.js tests/flow-hooks-commit-log-gate.test.js tests/flow-hooks-deploy-gate.test.js tests/flow-hooks-todowrite-gate.test.js tests/flow-hooks-git-safety-gate.test.js tests/flow-hooks-scope-mutation-gate.test.js tests/flow-hooks-strike-gate.test.js tests/flow-hooks-component-check.test.js tests/flow-hooks-scope-gate.test.js tests/flow-hooks-implementation-gate.test.js tests/flow-hooks-research-gate.test.js tests/flow-hooks-loop-check.test.js tests/flow-hooks-manager-boundary-gate.test.js tests/flow-hooks-phase-gate.test.js tests/flow-hooks-pre-tool-orchestrator.test.js tests/flow-hooks-observation-capture.test.js tests/flow-hooks-task-gate.test.js tests/flow-durable-session-suspension.test.js tests/flow-health-mcp-scopes.test.js tests/flow-lean-config.test.js tests/flow-workspace-autopickup.test.js tests/flow-worker-boundary-gate.test.js tests/flow-worker-question-classifier.test.js tests/flow-completion-truth-gate-contradictions.test.js tests/flow-structure-sensor.test.js tests/flow-workspace-dispatch-tracking.test.js tests/workspace-ipc-sqlite.test.js tests/workspace-ipc-multi-worker.test.js tests/flow-story-gates.test.js tests/flow-workspace-restart-handoff.test.js tests/flow-wogi-claude-wrapper.test.js tests/flow-wave1-integrations.test.js tests/flow-wave2-integrations.test.js tests/flow-wave3-integrations.test.js tests/flow-commit-claims-gate.test.js tests/auto-review.test.js tests/gate-telemetry-surface.test.js tests/agents-md-alias.test.js tests/flow-skill-manage.test.js tests/fuzzy-patch.test.js tests/mode-schema.test.js tests/flow-feature-dossier.test.js tests/flow-autonomous-mode.test.js tests/flow-epic-cascade.test.js tests/flow-workspace-summary.test.js tests/flow-hooks-research-evidence-gate.test.js tests/flow-worker-mcp-strip.test.js tests/flow-orchestrate-corrections.test.js tests/flow-source-fidelity.test.js tests/flow-hooks-long-input-enforcement.test.js tests/workspace-channel-tracking.test.js tests/flow-hooks-deletion-log.test.js tests/flow-task-boundary-reset.test.js tests/flow-deferral-gate.test.js tests/flow-research-required-gate.test.js && NODE_ENV=test node tests/run-quality-gates.test.js",
+    "test": "NODE_ENV=test node --test tests/auto-compact-prompt.test.js tests/flow-paths.test.js tests/flow-io.test.js tests/flow-audit-gates.test.js tests/flow-standards-hook-three-layer.test.js tests/flow-correction-detector-reconcile.test.js tests/flow-correction-backfill.test.js tests/flow-audit-gates-feature-output-health.test.js tests/flow-config-loader.test.js tests/flow-damage-control.test.js tests/flow-output.test.js tests/flow-constants.test.js tests/flow-session-state.test.js tests/flow-hooks-integration.test.js tests/flow-utils.test.js tests/flow-security.test.js tests/flow-memory-db.test.js tests/flow-durable-session.test.js tests/flow-skill-matcher.test.js tests/flow-bridge.test.js tests/flow-proactive-compact.test.js tests/flow-cascade-completion.test.js tests/flow-capture-gate.test.js tests/flow-correction-detector-hybrid.test.js tests/flow-promote.test.js tests/flow-archive-runs.test.js tests/flow-memory.test.js tests/flow-hooks-pre-tool-helpers.test.js tests/flow-hooks-bugfix-scope-gate.test.js tests/flow-hooks-routing-gate.test.js tests/flow-hooks-phase-read-gate.test.js tests/flow-hooks-commit-log-gate.test.js tests/flow-hooks-deploy-gate.test.js tests/flow-hooks-todowrite-gate.test.js tests/flow-hooks-git-safety-gate.test.js tests/flow-hooks-scope-mutation-gate.test.js tests/flow-hooks-strike-gate.test.js tests/flow-hooks-component-check.test.js tests/flow-hooks-scope-gate.test.js tests/flow-hooks-implementation-gate.test.js tests/flow-hooks-research-gate.test.js tests/flow-hooks-loop-check.test.js tests/flow-hooks-manager-boundary-gate.test.js tests/flow-hooks-phase-gate.test.js tests/flow-hooks-pre-tool-orchestrator.test.js tests/flow-hooks-observation-capture.test.js tests/flow-hooks-task-gate.test.js tests/flow-durable-session-suspension.test.js tests/flow-health-mcp-scopes.test.js tests/flow-lean-config.test.js tests/flow-workspace-autopickup.test.js tests/flow-worker-boundary-gate.test.js tests/flow-worker-question-classifier.test.js tests/flow-completion-truth-gate-contradictions.test.js tests/flow-structure-sensor.test.js tests/flow-workspace-dispatch-tracking.test.js tests/workspace-ipc-sqlite.test.js tests/workspace-ipc-multi-worker.test.js tests/flow-story-gates.test.js tests/flow-workspace-restart-handoff.test.js tests/flow-wogi-claude-wrapper.test.js tests/flow-wave1-integrations.test.js tests/flow-wave2-integrations.test.js tests/flow-wave3-integrations.test.js tests/flow-commit-claims-gate.test.js tests/auto-review.test.js tests/gate-telemetry-surface.test.js tests/agents-md-alias.test.js tests/flow-skill-manage.test.js tests/fuzzy-patch.test.js tests/mode-schema.test.js tests/flow-feature-dossier.test.js tests/flow-autonomous-mode.test.js tests/flow-epic-cascade.test.js tests/flow-workspace-summary.test.js tests/flow-hooks-research-evidence-gate.test.js tests/flow-worker-mcp-strip.test.js tests/flow-orchestrate-corrections.test.js tests/flow-source-fidelity.test.js tests/flow-hooks-long-input-enforcement.test.js tests/workspace-channel-tracking.test.js tests/flow-hooks-deletion-log.test.js tests/flow-task-boundary-reset.test.js tests/flow-deferral-gate.test.js tests/flow-research-required-gate.test.js && NODE_ENV=test node tests/run-quality-gates.test.js",
     "test:syntax": "find scripts/ lib/ -name '*.js' -not -path '*/node_modules/*' -exec node --check {} +",
     "lint": "eslint scripts/ lib/ tests/",
     "lint:ci": "eslint scripts/ lib/ tests/ --max-warnings 0",

package/scripts/flow-audit-gates.js

@@ -34,6 +34,7 @@ const fs = require('node:fs');
 const path = require('node:path');
 
 const { PATHS, safeJsonParse } = require('./flow-utils');
+const { safeJsonParseString } = require('./flow-io');
 
 // ============================================================
 // Score Cap Thresholds
@@ -641,6 +642,114 @@ function compareTrend(currentResults, previousAudit) {
 // Main: Run All Gates
 // ============================================================
 
+/**
+ * Gate: Feature Output Health (wf-6c58953a)
+ *
+ * Inspects DATA produced by features, not just CODE that produces it.
+ * Catches "silent feature no-op" — feature runs without errors, persists
+ * data, but the persisted data has all-null structured fields. This class
+ * is invisible to traditional code review/lint/typecheck/tests.
+ *
+ * Discovered 2026-05-09 when wogiflow-cli investigation found the
+ * correction-extractor was capturing user frustration but writing null
+ * structured fields. The /wogi-audit ran B+ and missed it because every
+ * agent inspects code, not output.
+ *
+ * Rule registry — explicit per-file checks, NOT a generic walker (per
+ * challenge round: blanket "all-null is bug" is false-positive city).
+ *
+ * @param {string} [projectRoot=PATHS.root] — project to inspect (default: current)
+ * @returns {Object} gate result with severity + findings
+ */
+function checkFeatureOutputHealth(projectRoot = PATHS.root) {
+  const findings = [];
+  const stateDir = path.join(projectRoot, '.workflow', 'state');
+  const corrDir = path.join(projectRoot, '.workflow', 'corrections');
+
+  // ---- Rule 1: pending-corrections.json null-fields ratio ----
+  // Note: pending-corrections.json is a top-level ARRAY, so safeJsonParse
+  // (which rejects arrays) won't work. Use file-read + safeJsonParseString.
+  const pcPath = path.join(stateDir, 'pending-corrections.json');
+  if (fs.existsSync(pcPath)) {
+    let records = [];
+    try {
+      const content = fs.readFileSync(pcPath, 'utf-8');
+      records = safeJsonParseString(content, []);
+    } catch (_err) { /* fail-open */ }
+    const arr = Array.isArray(records) ? records : [];
+    if (arr.length > 0) {
+      const nullCount = arr.filter(r =>
+        r && typeof r === 'object' &&
+        (r.whatWasWrong == null) &&
+        (r.whatUserWants == null)
+      ).length;
+      const ratio = nullCount / arr.length;
+      if (ratio >= 0.5) {
+        findings.push({
+          rule: 'pending-corrections-null-fields',
+          severity: ratio === 1 ? 'high' : 'medium',
+          message: `${nullCount}/${arr.length} (${Math.round(ratio * 100)}%) pending-corrections records have null structured fields. Likely correction-detector extraction failure. Run \`flow-correction-backfill\` or restore via Layer 2 enrichment.`,
+          evidence: `${path.relative(projectRoot, pcPath)}: ${arr.length} records analyzed; ${nullCount} fully null`
+        });
+      }
+    }
+  }
+
+  // ---- Rule 2: prompt-history × corrections cross-reference ----
+  // prompt-history.json is also typically a top-level array.
+  const phPath = path.join(stateDir, 'prompt-history.json');
+  if (fs.existsSync(phPath)) {
+    let ph = [];
+    try {
+      const content = fs.readFileSync(phPath, 'utf-8');
+      ph = safeJsonParseString(content, []);
+    } catch (_err) { /* fail-open */ }
+    const phArr = Array.isArray(ph) ? ph : (ph && Array.isArray(ph.prompts) ? ph.prompts : []);
+
+    // Frustration markers (regex per known-pattern set)
+    const frustrationRe = /\b(don'?t|stop|wait|actually|why did|why is|you keep|you always|fucking|seriously)\b/i;
+    let frustrationCount = 0;
+    for (const entry of phArr) {
+      if (!entry || typeof entry !== 'object') continue;
+      const text = entry.prompt || entry.text || entry.userMessage || '';
+      if (typeof text === 'string' && frustrationRe.test(text)) frustrationCount++;
+    }
+
+    let corrCount = 0;
+    if (fs.existsSync(corrDir)) {
+      try {
+        corrCount = fs.readdirSync(corrDir).filter(f => f.endsWith('.md')).length;
+      } catch (_err) { /* fail-open */ }
+    }
+
+    if (frustrationCount >= 3 && corrCount === 0) {
+      findings.push({
+        rule: 'prompt-history-vs-corrections-mismatch',
+        severity: 'high',
+        message: `prompt-history.json has ${frustrationCount} frustration markers but corrections/ is empty. Correction-extractor pipeline appears non-functional (captures input, fails to materialize records).`,
+        evidence: `prompt-history: ${frustrationCount} matches across ${phArr.length} entries; corrections/: ${corrCount} files`
+      });
+    }
+  }
+
+  // Determine overall gate severity
+  const hasHigh = findings.some(f => f.severity === 'high');
+  const hasMed = findings.some(f => f.severity === 'medium');
+  const severity = hasHigh ? 'high' : hasMed ? 'medium' : 'pass';
+
+  return {
+    gate: 'feature-output-health',
+    exists: true,
+    passed: findings.length === 0,
+    findings,
+    severity,
+    scoreCap: 100, // doesn't cap score directly; surfaces as audit findings
+    message: findings.length === 0
+      ? 'Feature output health: no issues detected'
+      : `Feature output health: ${findings.length} finding(s) — ${findings.map(f => f.rule).join(', ')}`
+  };
+}
+
 /**
  * Run all Gate 0 checks and return consolidated results.
  * @returns {Object} gate results with score cap
@@ -654,6 +763,7 @@ function runAllGates() {
   gates.push(checkLintConfigIntegrity());
   gates.push(checkTests());
   gates.push(checkScriptCompleteness());
+  gates.push(checkFeatureOutputHealth());
 
   const cap = calculateScoreCap(gates);
   const framework = detectFramework();
@@ -716,6 +826,14 @@ function main() {
       console.log(JSON.stringify(checkScriptCompleteness(), null, 2));
       break;
 
+    case 'feature-output-health': {
+      // Optional --project=<path> argument for cross-project audit
+      const projArg = process.argv.find(a => a.startsWith('--project='));
+      const projectRoot = projArg ? projArg.slice('--project='.length) : PATHS.root;
+      console.log(JSON.stringify(checkFeatureOutputHealth(projectRoot), null, 2));
+      break;
+    }
+
     case 'eslint-disable':
       console.log(JSON.stringify(countEslintDisables(), null, 2));
       break;
@@ -827,6 +945,7 @@ module.exports = {
   checkTests,
   parseTestErrorCount, // wf-e111d850: exposed for unit testing
   checkScriptCompleteness,
+  checkFeatureOutputHealth, // wf-6c58953a: feature output health gate
 
   // Extended checks
   countEslintDisables,

package/scripts/flow-correction-backfill.js

@@ -0,0 +1,148 @@
+#!/usr/bin/env node
+
+/**
+ * Wogi Flow — Pending-Corrections Backfill (wf-6c58953a)
+ *
+ * Backfills records in `.workflow/state/pending-corrections.json` that have
+ * null `whatWasWrong` / `whatUserWants` fields. The fix lands at code level
+ * (flow-correction-detector.js Layer 1+2 reconciliation), but historical
+ * records persisted before the fix already have null fields. This tool
+ * applies the same deterministic-fallback extraction retroactively.
+ *
+ * Strategy:
+ *   - Read pending-corrections.json
+ *   - For each record where userMessage is populated AND
+ *     (whatWasWrong is null OR whatUserWants is null)
+ *   - Apply deterministic extraction: whatWasWrong = first 200 chars of
+ *     userMessage; whatUserWants stays null (intent inference is an LLM job
+ *     — honest null > wrong guess; live extractor will populate going forward)
+ *   - Mark `enrichmentSource: "backfill-<date>"` so consumers can distinguish
+ *     backfilled from live extractions
+ *   - Atomic write: write-temp + rename
+ *
+ * Usage:
+ *   node scripts/flow-correction-backfill.js                        # current project
+ *   node scripts/flow-correction-backfill.js --project=<path>      # explicit project
+ *   node scripts/flow-correction-backfill.js --dry-run             # report only
+ */
+
+'use strict';
+
+const fs = require('node:fs');
+const path = require('node:path');
+
+const { PATHS } = require('./flow-utils');
+const { safeJsonParseString } = require('./flow-io');
+const { deterministicWhatWasWrong } = require('./flow-correction-detector');
+
+const BACKFILL_DATE = new Date().toISOString().slice(0, 10); // YYYY-MM-DD
+
+/**
+ * Backfill a single project's pending-corrections.json.
+ *
+ * @param {string} projectRoot — project directory containing .workflow/
+ * @param {Object} [opts]
+ * @param {boolean} [opts.dryRun=false] — if true, return what WOULD change without writing
+ * @returns {{ found: number, backfilled: number, alreadyPopulated: number, written: boolean, path: string|null, dryRun: boolean }}
+ */
+function backfillPendingCorrections(projectRoot, opts = {}) {
+  const { dryRun = false } = opts;
+  const pcPath = path.join(projectRoot, '.workflow', 'state', 'pending-corrections.json');
+
+  const result = {
+    found: 0,
+    backfilled: 0,
+    alreadyPopulated: 0,
+    written: false,
+    path: null,
+    dryRun
+  };
+
+  if (!fs.existsSync(pcPath)) {
+    result.path = pcPath;
+    return result;
+  }
+
+  let content;
+  try {
+    content = fs.readFileSync(pcPath, 'utf-8');
+  } catch (err) {
+    throw new Error(`Cannot read pending-corrections at ${pcPath}: ${err.message}`);
+  }
+
+  const records = safeJsonParseString(content, []);
+  if (!Array.isArray(records)) {
+    throw new Error(`Expected array at ${pcPath}; got ${typeof records}`);
+  }
+
+  result.found = records.length;
+  result.path = pcPath;
+
+  let changed = false;
+  for (const r of records) {
+    if (!r || typeof r !== 'object') continue;
+    const userMsg = r.userMessage;
+    if (typeof userMsg !== 'string' || !userMsg.trim()) continue;
+
+    const needsFill = (r.whatWasWrong == null) && (r.whatUserWants == null);
+    if (!needsFill) {
+      result.alreadyPopulated += 1;
+      continue;
+    }
+
+    // Apply deterministic extraction (whatWasWrong only — whatUserWants
+    // stays null; intent inference is the live extractor's job going forward)
+    r.whatWasWrong = deterministicWhatWasWrong(userMsg);
+    r.enrichmentSource = `backfill-${BACKFILL_DATE}`;
+    result.backfilled += 1;
+    changed = true;
+  }
+
+  if (changed && !dryRun) {
+    // Atomic write: write-temp + rename
+    const tmpPath = `${pcPath}.tmp-${process.pid}`;
+    fs.writeFileSync(tmpPath, JSON.stringify(records, null, 2) + '\n');
+    fs.renameSync(tmpPath, pcPath);
+    result.written = true;
+  }
+
+  return result;
+}
+
+// ============================================================
+// CLI
+// ============================================================
+
+function main() {
+  const argv = process.argv.slice(2);
+  const projArg = argv.find(a => a.startsWith('--project='));
+  const dryRun = argv.includes('--dry-run');
+
+  const projectRoot = projArg ? projArg.slice('--project='.length) : PATHS.root;
+
+  let result;
+  try {
+    result = backfillPendingCorrections(projectRoot, { dryRun });
+  } catch (err) {
+    console.error(`Error: ${err.message}`);
+    process.exit(1);
+  }
+
+  console.log(JSON.stringify({
+    project: projectRoot,
+    pendingCorrectionsPath: result.path,
+    found: result.found,
+    backfilled: result.backfilled,
+    alreadyPopulated: result.alreadyPopulated,
+    written: result.written,
+    dryRun: result.dryRun
+  }, null, 2));
+}
+
+module.exports = {
+  backfillPendingCorrections
+};
+
+if (require.main === module) {
+  main();
+}

package/scripts/flow-correction-detector.js

@@ -329,14 +329,106 @@ function recordHybridTelemetry(verdict, runCtx = {}) {
   }
 }
 
+// ============================================================================
+// Layer 1 + Layer 2 Reconciliation (wf-6c58953a)
+// ============================================================================
+
+/**
+ * Deterministic fallback for `whatWasWrong` — preserves the user's literal
+ * frustration text when LLM extraction is unavailable or fails. Better than
+ * null: a 200-char excerpt is honest signal; null is data loss.
+ *
+ * @param {string} message — user message
+ * @returns {string|null}
+ */
+function deterministicWhatWasWrong(message) {
+  if (typeof message !== 'string') return null;
+  const trimmed = message.trim();
+  if (!trimmed) return null;
+  return trimmed.slice(0, 200);
+}
+
+/**
+ * Reconcile Layer 1 (keyword classifier) + Layer 2 (Haiku LLM) results.
+ *
+ * Pre-fix bug: Layer 1 returned `{whatWasWrong: null, whatUserWants: null}`
+ * when keyword matched, never calling Layer 2. The user's actual frustration
+ * was captured but structured fields stayed null — silent feature no-op.
+ *
+ * Post-fix design:
+ *   - Layer 1 hit + Layer 2 success: trust Layer 1's classification (high-
+ *     precision keyword match), use Layer 2's strings if non-null else
+ *     deterministic fallback. Record `llmDisagreed` if Layer 2 said
+ *     `isCorrection: false` (e.g., user said "I'm just asking a question").
+ *   - Layer 1 hit + Layer 2 fail/skip: deterministic fallback for `whatWasWrong`
+ *     (first 200 chars). `whatUserWants` stays null (intent inference is an
+ *     LLM job; honest null > wrong guess).
+ *   - Layer 1 miss + Layer 2 success: Layer 2 is primary classifier (existing path).
+ *   - Both miss: not a correction.
+ *
+ * Pure function — testable in isolation, no LLM mock needed.
+ *
+ * @param {Object|null} layer1 — Layer 1 result {isCorrection, confidence, correctionType, method, matchedPattern}
+ * @param {Object|null} layer2 — Layer 2 (LLM) result {isCorrection, confidence, correctionType, whatWasWrong, whatUserWants}
+ * @param {string} trimmed — trimmed user message (for deterministic fallback)
+ * @returns {Object|null} reconciled record OR null if not a correction
+ */
+function reconcileExtraction(layer1, layer2, trimmed) {
+  // Both layers ran
+  if (layer1 && layer2) {
+    const what = layer2.whatWasWrong || deterministicWhatWasWrong(trimmed);
+    const wants = layer2.whatUserWants || null;
+    return {
+      isCorrection: true, // Layer 1 high-precision keyword match wins binary
+      confidence: layer1.confidence,
+      correctionType: layer1.correctionType || layer2.correctionType || 'behavior',
+      whatWasWrong: what,
+      whatUserWants: wants,
+      method: 'keyword+ai',
+      matchedPattern: layer1.matchedPattern,
+      enrichmentSource: layer2.whatWasWrong ? 'haiku' : 'deterministic-fallback',
+      llmDisagreed: layer2.isCorrection === false,
+    };
+  }
+  // Layer 1 only (Layer 2 unavailable: no API key, network error, etc.)
+  if (layer1) {
+    return {
+      isCorrection: true,
+      confidence: layer1.confidence,
+      correctionType: layer1.correctionType || 'behavior',
+      whatWasWrong: deterministicWhatWasWrong(trimmed),
+      whatUserWants: null,
+      method: layer1.method,
+      matchedPattern: layer1.matchedPattern,
+      enrichmentSource: 'deterministic-fallback',
+    };
+  }
+  // Layer 2 only (Layer 1 missed)
+  if (layer2 && layer2.isCorrection) {
+    return {
+      isCorrection: true,
+      confidence: layer2.confidence,
+      correctionType: layer2.correctionType || null,
+      whatWasWrong: layer2.whatWasWrong || null,
+      whatUserWants: layer2.whatUserWants || null,
+      method: 'ai',
+      enrichmentSource: 'haiku',
+    };
+  }
+  // Both missed → not a correction
+  return null;
+}
+
 // ============================================================================
 // AI-Based Detection (Haiku — language-agnostic)
 // ============================================================================
 
 /**
  * Detect if a message is a correction using Claude Haiku.
- * This is the ONLY detection method — no regex fallback.
- * Works in any language.
+ * Hybrid: Layer 1 keyword classifier (fast) + Layer 2 Haiku enrichment.
+ *
+ * wf-6c58953a (2026-05-09): Layer 1 hit no longer short-circuits structured
+ * extraction. See reconcileExtraction() for the post-fix design rationale.
  *
  * @param {string} userMessage - The user's message
  * @param {string} previousContext - Summary of what the AI was doing
@@ -354,8 +446,12 @@ async function detectCorrection(userMessage, previousContext = '') {
     return { isCorrection: false, confidence: 0, method: 'skipped', reason: 'length-filter' };
   }
 
-  // Layer 1 (wf-e6d65edf) — keyword pre-classifier. Skips Haiku entirely on a hit.
+  // Layer 1 (wf-e6d65edf) — keyword pre-classifier.
+  // wf-6c58953a: NO longer short-circuits structured extraction. Layer 1's
+  // classification is captured; reconcile with Layer 2 (or deterministic
+  // fallback when Layer 2 unavailable) at end.
   const hybridCfg = getHybridConfig();
+  let layer1Result = null;
   if (hybridCfg.hybridEnabled) {
     const matched = findKeywordMatch(trimmed);
     if (matched) {
@@ -368,12 +464,10 @@ async function detectCorrection(userMessage, previousContext = '') {
         confidence: conf,
         durationMs: Date.now() - start,
       });
-      return {
+      layer1Result = {
         isCorrection: true,
         confidence: conf,
         correctionType: 'behavior',
-        whatWasWrong: null,
-        whatUserWants: null,
         method: 'keyword',
         matchedPattern: matched.phrase,
       };
@@ -383,6 +477,10 @@ async function detectCorrection(userMessage, previousContext = '') {
   // Check if API key is available
   const apiKey = process.env.ANTHROPIC_API_KEY;
   if (!apiKey) {
+    // wf-6c58953a: Layer 1 hit + no API key → deterministic fallback (not null)
+    if (layer1Result) {
+      return reconcileExtraction(layer1Result, null, trimmed);
+    }
     return { isCorrection: false, confidence: 0, method: 'skipped', reason: 'no-api-key' };
   }
 
@@ -492,11 +590,19 @@ Respond with JSON only (no markdown, no explanation):
       durationMs: Date.now() - start,
     });
 
+    // wf-6c58953a: reconcile Layer 1 + Layer 2 (or just Layer 2 if Layer 1 missed)
+    const reconciled = reconcileExtraction(layer1Result, aiResult, trimmed);
+    if (reconciled) return reconciled;
+    // Both layers say no-correction
     return aiResult;
   } catch (err) {
     if (process.env.DEBUG) {
       console.error(`[DEBUG] AI correction detection failed: ${err.message}`);
     }
+    // wf-6c58953a: Layer 2 failure with Layer 1 hit → deterministic fallback
+    if (layer1Result) {
+      return reconcileExtraction(layer1Result, null, trimmed);
+    }
     return { isCorrection: false, confidence: 0, method: 'ai', reason: err.message };
   }
 }
@@ -1295,11 +1401,15 @@ function correlateWithPriorGates(correction) {
 // ============================================================================
 
 module.exports = {
-  // Detection (AI-only)
+  // Detection (hybrid Layer 1 + Layer 2)
   detectCorrection,
   batchAnalyzePrompts,
   spawnBackgroundDetection,
 
+  // wf-6c58953a: reconciliation helpers exposed for unit testing + backfill
+  reconcileExtraction,
+  deterministicWhatWasWrong,
+
   // Queue management
   loadPendingCorrections,
   queuePendingCorrection,