@mindrian_os/cli 2.0.0-beta.3 → 2.0.0-beta.33

package/lib/core/navigation-engine.cjs

@@ -0,0 +1,1696 @@
+#!/usr/bin/env node
+'use strict';
+
+/*
+ * Copyright (c) 2026 Mindrian. BSL 1.1.
+ *
+ * Phase 91-00 -- Navigation Engine core (L5 Decision layer)
+ * ==========================================================
+ * Single export `decide(turn, context) -> decision` composes the five
+ * input signals (ICM scope + SQL relations + Feynman-MINTO reasoning +
+ * BRAIN.md derivations + intent/persona) through a structured decision
+ * function (NOT a weighted score per locked decision D-02) and returns
+ * a typed decision struct with full trace.
+ *
+ * Contract source of truth:
+ *   .planning/research/navigation-engine-brain-interface.md (v1, frozen
+ *   at Phase 90-09).
+ *
+ * Canon Part 8 posture (READ THIS FIRST):
+ *   This module is a PURE LOCAL READER. It NEVER queries Brain. It
+ *   NEVER reads BRAIN.md directly via fs.readFileSync; the only allowed
+ *   read path is folder-memory.cjs readQuadruple. It NEVER constructs
+ *   Cypher queries. It NEVER embeds BRAIN.md body strings in network
+ *   payloads. The two PERMITTED brain-client touches are
+ *   brain-client.isAvailable() (boolean, no network when cached) and
+ *   brain-client.schema() (scalar, no user content leaves the local
+ *   process). Both are optional; the engine works fully offline.
+ *
+ * Per Section 2.4: a single readQuadruple result MAY be cached for the
+ * duration of one decide() call. Cross-call caching is FORBIDDEN.
+ *
+ * Per locked decision D-02: this engine is NOT a weighted score across
+ * the five signals. Brain's section weights (0.35 + 0.20 + 0.15 + ...)
+ * stay INSIDE the BRAIN.md contribution block; the staleness multiplier
+ * scales that block as a whole. Inter-signal composition is rule-based
+ * with explicit precedence, not arithmetic.
+ *
+ * License: BSL 1.1.
+ */
+
+const shared = require('./navigation-engine-shared.cjs');
+const folderMemory = require('./folder-memory.cjs'); // for readQuadruple
+// CASC-02 (Phase 142): the navigation.cjs chokepoint is the ONLY allowed local
+// graph read path (Canon Part 9: SQL is the local mind). decide() reads the
+// navigated neighborhood (getRoomContext's Leg C, getNeighborhood-ranked) that the
+// caller threaded onto context.roomContext through this same chokepoint. We require
+// it here as the Part-8 wiring anchor; the caller owns the db handle, so this module
+// never opens room.db and adds ZERO new Brain egress surface (the neighborhood rides
+// in via context as scalars/slugs/scores only).
+const navigation = require('./navigation.cjs'); // chokepoint anchor (Part 9)
+// Phase 144 (NAV-01): the Phase 143 sensor spine. decide() CONSUMES the
+// candidate reaches dispatchSensors PRODUCES; a fired reach maps to a CANONICAL
+// VERB so the router (READ-ONLY for 144) flips routing_source legacy -> engine as
+// a pure consequence (its Precedence Rule 1: validateVerb===true). dispatchSensors
+// is pure / sync / LOCAL-first (no Brain, no network), so this require adds ZERO
+// Brain egress surface. We never edit the sensor module (Phase 143 fence).
+const { dispatchSensors } = require('./insight-sensors.cjs');
+// Phase 203-03: the reusable-expert ctx-assembly producer. Reads CONFIRMED
+// SyntheticExpert nodes from a caller-threaded room.db handle at ctx-assembly time
+// and threads the closed reusable-expert scalars onto sensorCtx (mirroring the
+// MED-01 cortex producer below), so the registered sensorExpertSkill can fire the
+// context_block save-as-skill offer. The db read is the producer's job; the sensor
+// stays pure. The engine never OPENS room.db -- it uses the caller-threaded handle.
+const { detectExpertSkillCandidates } = require('./sensors/sensor-expert-skill.cjs');
+// Phase 244 Plan 05 (TRIG-01): the content-relevance ctx-assembly producer.
+// Queries the ALREADY-SHIPPED tri-modal-index.lexicalSearch (bm25 FTS5) over a
+// caller-threaded room.db handle at ctx-assembly time and threads the closed
+// content-relevance scalars onto sensorCtx (mirroring the SENS-11 producer
+// above), so the registered sensorContentRelevance can fire the context_block
+// content-relevance offer. The db read is the producer's job; the sensor
+// stays pure. The engine never OPENS room.db -- it uses the caller-threaded handle.
+const { detectContentRelevance } = require('./sensors/sensor-content-relevance.cjs');
+// Phase 177-09 (BCH-07): the SEAM 3 behavioral-channel kill-switch is read from
+// calibration-gate.cjs (default LOCKED, shipped in 177-08) via the require idiom
+// -- NEVER a re-typed `false` literal. The require is LAZY (inside
+// readBehavioralChannelArmed, called from resolveFireSkill) to avoid the import
+// cycle: calibration-gate requires f-selector-ranker, which requires THIS module
+// for RECOMMENDED_CONFIDENCE_FLOOR -- a top-level require here would force
+// f-selector-ranker to load before this module finishes exporting, leaving its
+// computed CEILING NaN. Lazy require lets this module finish loading first.
+// calibration-gate.cjs is pure-local (no Brain, no network) -> ZERO Brain egress.
+function readBehavioralChannelArmed() {
+  return require('./navigation/calibration-gate.cjs').BEHAVIORAL_CHANNEL_ARMED;
+}
+// brain-client is OPTIONAL: required only if downstream callers want to
+// pass an explicit brainAvailable hint, but isAvailable() is also fine
+// from the call site. We do NOT require() it here to keep the dependency
+// graph minimal and to make grep-guards trivially passable.
+
+const STALENESS_MULTIPLIERS = shared.STALENESS_MULTIPLIERS;
+// The dead weighted-section-score import was removed (Phase 150-04 MEM-07):
+// composition stays rule-based per 150-CONTEXT D-03; it had no live consumer.
+const REQUIRED_SECTION_KEYS = shared.REQUIRED_SECTION_KEYS;
+const OPTIONAL_SECTION_KEYS = shared.OPTIONAL_SECTION_KEYS;
+const CANONICAL_VERBS = shared.CANONICAL_VERBS;
+
+// Wicked-score escalation threshold per Canon Appendix E rule R4.
+const WICKED_ESCALATION_THRESHOLD = 8;
+
+// Confidence floor for Section 6 RECOMMENDED gate.
+const RECOMMENDED_CONFIDENCE_FLOOR = 0.7;
+
+// ---------- Pattern parsers (tolerant; never throw) ----------
+
+/**
+ * Parse pattern_matches body for `confidence: <float>` candidates.
+ * Returns the highest confidence found in [0, 1], or null when no
+ * candidate is parseable. Tolerant of body whitespace, multi-line, and
+ * forward-compat extra fields.
+ */
+function extractHighestConfidence(body) {
+  if (typeof body !== 'string' || body.length === 0) return null;
+  const re = /confidence\s*:\s*([01](?:\.\d+)?|0?\.\d+)/gi;
+  let best = null;
+  let m;
+  while ((m = re.exec(body)) !== null) {
+    const v = parseFloat(m[1]);
+    if (!isNaN(v) && v >= 0.0 && v <= 1.0) {
+      if (best === null || v > best) best = v;
+    }
+  }
+  return best;
+}
+
+/**
+ * Extract the verb associated with the highest-confidence candidate in
+ * pattern_matches. Returns null when no candidate is parseable. Verbs
+ * MUST appear before the `(confidence: ...)` parenthetical, matching
+ * Canon Part 3 vocabulary.
+ *
+ * Body shape (per Plan 90-01 deriveSection emit):
+ *   - Run Methodology (confidence: 0.85, source: SWOT)
+ *   - Reformulate (confidence: 0.72, source: 5 Whys)
+ */
+function extractTopCandidateVerb(body) {
+  if (typeof body !== 'string' || body.length === 0) return null;
+  const lines = body.split(/\r?\n/);
+  let bestConf = -1;
+  let bestVerb = null;
+  for (const line of lines) {
+    // Match `<prefix> <verb> (confidence: 0.NN ...)`
+    const m = /^[\s\-*]*(.+?)\s*\(\s*confidence\s*:\s*([01](?:\.\d+)?|0?\.\d+)/i.exec(line);
+    if (!m) continue;
+    const verb = m[1].trim();
+    const conf = parseFloat(m[2]);
+    if (isNaN(conf)) continue;
+    // Resolve verb against canonical vocabulary (case-insensitive
+    // contains-match; closed set).
+    let resolved = null;
+    for (const v of CANONICAL_VERBS) {
+      if (verb.toLowerCase() === v.toLowerCase()) {
+        resolved = v;
+        break;
+      }
+    }
+    if (resolved !== null && conf > bestConf) {
+      bestConf = conf;
+      bestVerb = resolved;
+    }
+  }
+  return bestVerb;
+}
+
+/**
+ * Phase 245 (245-05) -- the Brain pattern_matches OBSERVATION, computed
+ * UNCONDITIONALLY on every decide() turn where a parseable candidate exists.
+ *
+ * WHY THIS EXISTS (D-25, 245-RESEARCH.md F-2). resolveFireSkill() below has a
+ * strict precedence order, and its step (2) -- a fired sensor reach -- RETURNS
+ * EARLY. Step (3), the Brain pattern_matches branch, is therefore UNREACHABLE on
+ * any turn where a sensor fired, which is essentially every observed turn. D-24
+ * diagnosed the Brain verb as STARVED: the only two pre-existing callers of
+ * extractTopCandidateVerb are that unreachable step (3) and the RECOMMENDED-marker
+ * gate, which is itself gated on tierMode === 'mode_a' && !attributionBreach &&
+ * confidence >= 0.70 and which discards the verb after using it as a boolean. On a
+ * normal sensor-fires turn the verb string existed NOWHERE downstream.
+ *
+ * This observation is deliberately UNGATED: not on tierMode, not on weightApplied,
+ * not on attributionBreach, not on RECOMMENDED_CONFIDENCE_FLOOR. Those gates govern
+ * whether the verb may FIRE a skill or RENDER a marker. This is an observation of
+ * what Brain suggested; the consumer (245-07's fusion) applies its own tier policy.
+ *
+ * NOT a substitute for, and not derivable from, fire_skill or
+ * context_assembly.decision_grounding: 245-RESEARCH.md Pitfall 2 is explicit that
+ * on a sensor-fires turn decision_grounding reads the SENSOR's reach_id, never
+ * 'brain_verb'. Reading the Brain verb off either would silently re-implement the
+ * exact starvation this function exists to fix. These fields are the ONLY source.
+ *
+ * Part 8: verb is a frozen CANONICAL_VERBS member or null (extractTopCandidateVerb
+ * resolves by exact case-insensitive match against the closed set, so injected free
+ * prose yields null); confidence is a bare number or null. No body text, no source
+ * string, no section name is copied out.
+ *
+ * Soft-fail: a malformed brain object can never throw out of decide(), matching the
+ * discipline every other trace builder in this file uses.
+ *
+ * @param {object|null} brain  quadruple.brain, or null
+ * @returns {{verb: (string|null), confidence: (number|null)}}
+ */
+function brainPatternVerbObservation(brain) {
+  try {
+    const pm = brain && brain.sections && brain.sections.pattern_matches;
+    if (pm && typeof pm.body === 'string' && pm.body.length > 0) {
+      return {
+        verb: extractTopCandidateVerb(pm.body),
+        confidence: extractHighestConfidence(pm.body),
+      };
+    }
+  } catch (_e) {
+    // fall through to the null observation
+  }
+  return { verb: null, confidence: null };
+}
+
+/**
+ * Parse wicked_indicators body for `wicked_score: <int>`. Returns 0 when
+ * no score is parseable.
+ */
+function extractWickedScore(body) {
+  if (typeof body !== 'string' || body.length === 0) return 0;
+  const m = /wicked_score\s*:\s*(\d+)/i.exec(body);
+  if (!m) return 0;
+  const v = parseInt(m[1], 10);
+  if (isNaN(v) || v < 0) return 0;
+  return v;
+}
+
+// ---------- Trace builders ----------
+
+function buildIcmScope(quadruple, turn) {
+  const room = quadruple && quadruple.room;
+  return {
+    section_path: turn && turn.sectionPath ? turn.sectionPath : null,
+    room_exists: room && room.exists ? true : false,
+    artifact_count: quadruple && quadruple.state ? (quadruple.state.artifact_count || 0) : 0,
+  };
+}
+
+function buildSqlSignals(_quadruple, _turn) {
+  // Phase 91-00 ships the engine surface; the SQL signal is consumed
+  // from a future plan that wires room.db edge counts. For this plan
+  // the slot is structural (so /mos:explain-decision can render it) but
+  // contributes no rules. Returning an object (not null) keeps the
+  // five-signal capture invariant met (Test 20).
+  return { edges_consulted: 0, contradictions_detected: 0 };
+}
+
+function buildMintoReasoning(quadruple) {
+  const r = quadruple && quadruple.reasoning;
+  if (!r) {
+    return { exists: false, reasoning_health_score: 0, governing_thought: null, is_stale: true };
+  }
+  return {
+    exists: r.exists === true,
+    reasoning_health_score:
+      typeof r.reasoning_health_score === 'number' ? r.reasoning_health_score : 0,
+    governing_thought: typeof r.governing_thought === 'string' ? r.governing_thought : null,
+    is_stale: r.is_stale === true,
+  };
+}
+
+function buildIntentPersona(context) {
+  const persona = context && context.userPersona ? context.userPersona : null;
+  const intent = context && context.intentSignal ? context.intentSignal : null;
+  return {
+    archetype: persona && persona.archetype ? persona.archetype : null,
+    problem_type: persona && persona.problem_type ? persona.problem_type : null,
+    venture_stage: persona && persona.venture_stage ? persona.venture_stage : null,
+    intent: intent && intent.intent ? intent.intent : null,
+    intent_confidence: intent && typeof intent.confidence === 'number' ? intent.confidence : null,
+  };
+}
+
+// ---------- CASC-02: navigated-neighborhood trace builder ----------
+
+/**
+ * buildNavigatedNeighborhood(roomContext) -> { ranked: [...] } | null
+ *
+ * CASC-02 (Phase 142): reflect the getNeighborhood-ranked Leg C of a getRoomContext
+ * fusion (threaded onto context.roomContext by the caller through the navigation.cjs
+ * chokepoint) into the decision trace. This is what makes the Phase 109 spine
+ * NAVIGATE: routing reflects the ranked graph neighborhood, not merely stored-edge
+ * presence.
+ *
+ * Canon Part 8 (HARD): the projection carries SCALARS / SLUGS / SCORES ONLY -- node
+ * id, type, incoming edge type, depth, and ranking score. It NEVER copies a node's
+ * raw body / prose / properties. The neighborhood is a LOCAL routing signal; it never
+ * rides toward the Brain.
+ *
+ * Defensive: a null / malformed / Promise / empty roomContext returns null so decide()
+ * falls through to the legacy behavior with the trace field at its safe default. Never
+ * throws. A thenable (un-awaited getRoomContext Promise) has no synchronous
+ * relevantNodes, so it degrades to null -- the caller is expected to await and pass the
+ * RESOLVED fusion object (the Task-1 contract).
+ */
+function buildNavigatedNeighborhood(roomContext) {
+  if (!roomContext || typeof roomContext !== 'object') return null;
+  // A bare Promise / thenable cannot be read synchronously here; degrade to null.
+  if (typeof roomContext.then === 'function' && !Array.isArray(roomContext.relevantNodes)) {
+    return null;
+  }
+  const nodes = roomContext.relevantNodes;
+  if (!Array.isArray(nodes) || nodes.length === 0) return null;
+
+  const ranked = [];
+  for (const n of nodes) {
+    if (!n || typeof n !== 'object') continue;
+    // Scalars / slugs / scores ONLY (Part 8). No properties, no claim/summary prose.
+    ranked.push({
+      id: typeof n.id === 'string' ? n.id : null,
+      type: typeof n.type === 'string' ? n.type : null,
+      edgeTypeIn: typeof n.edgeTypeIn === 'string' ? n.edgeTypeIn : null,
+      depth: typeof n.depth === 'number' ? n.depth : null,
+      score: typeof n.score === 'number' ? n.score : null,
+    });
+  }
+  if (ranked.length === 0) return null;
+
+  const seedNodeId = roomContext._meta && typeof roomContext._meta.seedNodeId === 'string'
+    ? roomContext._meta.seedNodeId
+    : null;
+
+  return {
+    ranked: ranked,
+    count: ranked.length,
+    seed_node_id: seedNodeId,
+    // top_node_id is the highest-ranked neighbor slug -- a scalar routing handle.
+    top_node_id: ranked[0].id,
+  };
+}
+
+// ---------- Phase 144 Task 2: Zep-shaped LOCAL trace.context_assembly ----------
+
+/**
+ * buildContextAssembly(navigatedNeighborhood, sensorReaches, decisionGrounding)
+ *   -> { user_summary, facts, decision_grounding }
+ *
+ * Phase 144 (Task 2): adopt Zep's Smart-Context-Assembly OUTPUT SHAPE inside the
+ * decision TRACE ONLY. Every byte is LOCAL trace JSON; NOTHING reaches the Brain.
+ *
+ *   user_summary  top-3 navigatedNeighborhood node SCALARS (id / type / score)
+ *                 each carrying LOCAL temporal scalars { first_seen, last_updated }
+ *                 read ONLY from the neighborhood scalars (null when the source
+ *                 did not supply them -- never a fabricated value).
+ *   facts         the fired sensor reach_ids + evidence SCALARS
+ *                 (counts / enums / phase ids only -- makeReach already froze
+ *                 evidence to a flat primitive bag), each with the same LOCAL
+ *                 temporal scalars (null when the reach did not supply them).
+ *   decision_grounding  which fact (reach_id | 'brain_verb' | 'wicked' |
+ *                 'neighborhood' | null) justified the chosen fire_skill.
+ *
+ * Part 8 (HARD): user_summary + facts carry SCALARS / SLUGS / ENUMS only. No
+ * node body, no reach.signal user value, no reach.dispatch beyond the handle, no
+ * prose. This mirrors buildNavigatedNeighborhood's projection discipline.
+ */
+function buildContextAssembly(navigatedNeighborhood, sensorReaches, decisionGrounding) {
+  const userSummary = [];
+  if (navigatedNeighborhood && Array.isArray(navigatedNeighborhood.ranked)) {
+    const top = navigatedNeighborhood.ranked.slice(0, 3);
+    for (const n of top) {
+      if (!n || typeof n !== 'object') continue;
+      userSummary.push({
+        id: typeof n.id === 'string' ? n.id : null,
+        type: typeof n.type === 'string' ? n.type : null,
+        score: typeof n.score === 'number' ? n.score : null,
+        // LOCAL temporal scalars read ONLY from the neighborhood scalars; null
+        // when the source did not supply them (never fabricated).
+        first_seen: typeof n.first_seen === 'number' || typeof n.first_seen === 'string' ? n.first_seen : null,
+        last_updated: typeof n.last_updated === 'number' || typeof n.last_updated === 'string' ? n.last_updated : null,
+      });
+    }
+  }
+
+  const facts = [];
+  if (Array.isArray(sensorReaches)) {
+    for (const r of sensorReaches) {
+      if (!r || typeof r !== 'object') continue;
+      const ev = (r.evidence && typeof r.evidence === 'object') ? r.evidence : {};
+      // evidence is already a frozen flat scalar bag (makeReach guarantees it).
+      facts.push({
+        reach_id: typeof r.reach_id === 'string' ? r.reach_id : null,
+        posture: typeof r.posture === 'string' ? r.posture : null,
+        evidence: ev,
+        // LOCAL temporal scalars sourced ONLY from the reach evidence scalars;
+        // null when the reach did not supply them (never fabricated).
+        first_seen: typeof ev.first_seen === 'number' || typeof ev.first_seen === 'string' ? ev.first_seen : null,
+        last_updated: typeof ev.last_updated === 'number' || typeof ev.last_updated === 'string' ? ev.last_updated : null,
+      });
+    }
+  }
+
+  return {
+    user_summary: userSummary,
+    facts: facts,
+    decision_grounding: decisionGrounding !== undefined ? decisionGrounding : null,
+  };
+}
+
+/**
+ * groundingForFireSkill(fireSkill, sensorReaches, brain, navigatedNeighborhood)
+ *   -> reach_id | 'wicked' | 'brain_verb' | 'neighborhood' | null
+ *
+ * Phase 144 (Task 2): name which LOCAL fact justified the chosen fire_skill, for
+ * trace.context_assembly.decision_grounding. Mirrors the resolveFireSkill
+ * precedence order (wicked > sensor reach > brain verb > fallback). Pure scalar
+ * reasoning; reads no prose.
+ */
+function groundingForFireSkill(fireSkill, sensorReaches, brain, navigatedNeighborhood) {
+  if (fireSkill === null || fireSkill === undefined) return null;
+  // Wicked soft-systems return is the only family-slug the sensor/verb paths
+  // never produce, so a 'soft-systems' fire_skill is grounded in wicked.
+  if (fireSkill === 'soft-systems') return 'wicked';
+  // A fired sensor reach grounds the canonical-verb fire_skill: the top reach
+  // (canonical REACH_IDS order) drove it per resolveFireSkill step (2).
+  if (Array.isArray(sensorReaches) && sensorReaches.length > 0) {
+    const top = sensorReaches[0];
+    if (top && typeof top === 'object' && reachIdToSkillFamily(top.reach_id) === fireSkill) {
+      return typeof top.reach_id === 'string' ? top.reach_id : null;
+    }
+  }
+  // Otherwise the BRAIN.md pattern_matches / context-engine path set it.
+  if (brain && brain.sections) return 'brain_verb';
+  // Last resort: a neighborhood-only contribution.
+  if (navigatedNeighborhood !== null) return 'neighborhood';
+  return null;
+}
+
+// ---------- Section consumption ----------
+
+function consumedSections(brain) {
+  if (!brain || typeof brain !== 'object' || !brain.sections) return [];
+  const out = [];
+  // Required first (canonical order), then optional.
+  for (const k of REQUIRED_SECTION_KEYS) {
+    const v = brain.sections[k];
+    if (v !== null && v !== undefined && typeof v === 'object' && typeof v.body === 'string' && v.body.length > 0 && !/^\s*\(no signal\)\s*$/i.test(v.body)) {
+      out.push(k);
+    }
+  }
+  for (const k of OPTIONAL_SECTION_KEYS) {
+    const v = brain.sections[k];
+    if (v !== null && v !== undefined && typeof v === 'object' && typeof v.body === 'string' && v.body.length > 0 && !/^\s*\(no signal\)\s*$/i.test(v.body)) {
+      out.push(k);
+    }
+  }
+  return out;
+}
+
+// ---------- Five-signal composition (rule-based per D-02) ----------
+
+/**
+ * reachIdToSkillFamily(reachId) -> canonical-verb | null
+ *
+ * Phase 144 (NAV-01): map the 5 FROZEN reach_ids
+ * (lib/core/sensors/sensor-types.cjs REACH_IDS) to one of the closed 10
+ * CANONICAL_VERBS. This is the keystone of the legacy -> engine flip: the router
+ * (lib/core/skill-activation-router.cjs::validateVerb) validates fire_skill
+ * against CANONICAL_VERBS, NOT against skill-family slugs. So a sensor reach must
+ * resolve to a CANONICAL VERB for the router's Precedence Rule 1
+ * (validateVerb===true) to flip routing_source to engine. (Mirrors the
+ * verbToSkillFamily switch shape; the table is closed -- a non-member key returns
+ * null and the engine never invents a reach.)
+ *
+ *   context_block   -> 'Run Methodology'
+ *   contradiction   -> "Devil's Advocate"
+ *   cross_room      -> 'Navigate Graph'
+ *   brain_consult   -> 'Run Methodology'
+ *   deep_research   -> 'Spawn Sub-Agent'
+ *   hats            -> 'Synthesize'
+ *
+ * The `hats` case (Phase 172-05) fills the missing engine MAPPING for the
+ * already-frozen 6th reach (minted by the Phase-148 D-09 amendment). A fired
+ * `hats` reach runs a Six Thinking Hats / Blue-Hat synthesis pass, so it maps
+ * to the EXISTING canonical verb 'Synthesize' (verbToSkillFamily resolves it to
+ * the 'blue-hat' skill family -- the Blue Hat that wraps a hats pass). NO new
+ * canonical verb is minted; 'Synthesize' is a member of the frozen Part-3
+ * CANONICAL_VERBS set. This is purely the engine-side mapping so a fired hats
+ * reach can flip routing_source legacy->engine; it does NOT add a sensor that
+ * FIRES `hats` at runtime (that is out of scope, deferred by the projection
+ * contract).
+ */
+function reachIdToSkillFamily(reachId) {
+  switch (reachId) {
+    case 'context_block': return 'Run Methodology';
+    case 'contradiction': return "Devil's Advocate";
+    case 'cross_room': return 'Navigate Graph';
+    case 'brain_consult': return 'Run Methodology';
+    case 'deep_research': return 'Spawn Sub-Agent';
+    case 'hats': return 'Synthesize';
+    default: return null;
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Phase 177-09 -- the SEAM 3 behavioral-channel modifier (BCH-07 / BCH-08).
+//
+// WIRED-BUT-SHADOWED contract (read this first):
+//   The modifier is inserted INSIDE resolveFireSkill(), between the sensor-reach
+//   branch (ends :466) and the mode_a Brain-verb path (:468). It is GATED by
+//   BEHAVIORAL_CHANNEL_ARMED (default LOCKED). When the flag is false -- which is
+//   ALWAYS, today, because no real calibration PASS has armed it -- the modifier
+//   returns null UNCONDITIONALLY, the legacy Brain path runs unchanged, and the
+//   decide() output is BYTE-IDENTICAL to the pre-seam engine (D-177-BCH-07). The
+//   seam fires nothing semantic when unarmed; routing_source stays legacy.
+//   This dormant-when-unarmed property is the SAFETY INVARIANT the test asserts.
+//
+//   The LIVE flip (a SIGNAL-tier cue actually outranking the graph) happens ONLY
+//   when a real calibration PASS arms the flag. That arming is DATA-GATED (it
+//   needs ground-truth-labeled shadow data, deferred per 177-08 fit.method
+//   'deferred') AND requires a Canon Custodian / brain-boundary review before any
+//   live arming (Canon Part 8 PR gate). We build the mechanism dormant; we do
+//   NOT fire it. Frozen sets untouched (6 reaches, 3 postures); no 7th reach.
+//
+// Canon Part 8: the modifier ROUTES (it picks a verb from a fired reach via the
+// existing reachIdToSkillFamily); it adds ZERO new Brain egress. It reads only
+// enum/scalar handles (reach_id, posture, tier, confidence) and the ARMED
+// boolean -- NEVER user prose, an artifact body, or a meeting transcript.
+
+/**
+ * Compare two calibrated cues by TRIGGER_TIERS precedence first (signal above
+ * keyword, read from sensor-types -- never a re-typed literal order), confidence
+ * second. Pure helper, separately testable so BCH-08 can assert tier precedence
+ * WITHOUT arming the live flag. Returns the winning cue (the one that should be
+ * preferred). Defensive; never throws.
+ *
+ * @param {object} cueA -- { tier, confidence }
+ * @param {object} cueB -- { tier, confidence }
+ * @returns {object|null} the preferred cue, or null when neither is usable
+ */
+function rankBehavioralCue(cueA, cueB) {
+  const { TRIGGER_TIERS } = require('./sensors/sensor-types.cjs');
+  // Lower index = stronger tier (signal=0, context=1, keyword=2).
+  const rankOf = (cue) => {
+    if (!cue || typeof cue !== 'object') return Infinity;
+    const idx = TRIGGER_TIERS.indexOf(cue.tier);
+    return idx === -1 ? Infinity : idx;
+  };
+  const a = rankOf(cueA);
+  const b = rankOf(cueB);
+  if (a === Infinity && b === Infinity) return null;
+  if (a < b) return cueA;
+  if (b < a) return cueB;
+  // Same tier: higher confidence wins (defensive on non-number).
+  const confA = (cueA && typeof cueA.confidence === 'number') ? cueA.confidence : -1;
+  const confB = (cueB && typeof cueB.confidence === 'number') ? cueB.confidence : -1;
+  return confA >= confB ? cueA : cueB;
+}
+
+/**
+ * Pure predicate: does a protected-band cue LOSE to a Brain RECOMMENDED
+ * confidence? A cue inside the protected band [BEHAVIORAL_CHANNEL_FLOOR,
+ * BEHAVIORAL_CHANNEL_CEILING] CANNOT override a Brain RECOMMENDED confidence
+ * >= RECOMMENDED_CONFIDENCE_FLOOR (0.70). All three thresholds are READ via the
+ * require idiom (floor/ceiling from f-selector-ranker, the Brain floor from this
+ * module's own const) -- NO threshold literal lands here (BCH-05). The
+ * f-selector-ranker require is LAZY (inside the fn) to avoid the import cycle
+ * (f-selector-ranker requires this module for RECOMMENDED_CONFIDENCE_FLOOR).
+ *
+ * @param {object} cue -- { tier, confidence }
+ * @param {number} brainConfidence -- the Brain RECOMMENDED confidence (or null)
+ * @returns {boolean} true when the band cue loses to Brain >= 0.70
+ */
+function bandCueLosesToBrain(cue, brainConfidence) {
+  if (!cue || typeof cue !== 'object' || typeof cue.confidence !== 'number') return false;
+  if (typeof brainConfidence !== 'number') return false;
+  const {
+    BEHAVIORAL_CHANNEL_FLOOR,
+    BEHAVIORAL_CHANNEL_CEILING,
+  } = require('../workflow/f-selector-ranker.cjs');
+  const inBand =
+    cue.confidence >= BEHAVIORAL_CHANNEL_FLOOR &&
+    cue.confidence < BEHAVIORAL_CHANNEL_CEILING;
+  if (!inBand) return false;
+  // A band cue loses when Brain meets or exceeds the RECOMMENDED floor.
+  return brainConfidence >= RECOMMENDED_CONFIDENCE_FLOOR;
+}
+
+/**
+ * The SEAM 3 modifier. Takes the fired top reach object (sensorReaches[0], or
+ * null), the calibrated cue (a scalar { tier, confidence } the sensor spine
+ * already surfaced -- NEVER user prose; Part 8), and the armed boolean.
+ *
+ * FIRST statement is the constitutional kill-switch: if `armed` is not strictly
+ * true, return null. When BEHAVIORAL_CHANNEL_ARMED is false the function is
+ * provably inert, so resolveFireSkill falls through to the existing :468 Brain
+ * path byte-identical.
+ *
+ * The armed-true body (which NEVER runs today) maps the fired reach to its
+ * canonical verb via reachIdToSkillFamily and returns { verb, confidence,
+ * reason }; when the reach does not map, or the cue is below
+ * BEHAVIORAL_CHANNEL_FLOOR, it returns null (defer to Brain).
+ *
+ * @param {object|null} topReach -- the fired top reach { reach_id, posture }
+ * @param {object|null} cue -- the calibrated cue { tier, confidence }
+ * @param {boolean} armed -- BEHAVIORAL_CHANNEL_ARMED (false today)
+ * @returns {{verb:string, confidence:number, reason:string}|null}
+ */
+function behavioralChannelModifier(topReach, cue, armed) {
+  // Constitutional kill-switch (D-177-BCH-07). Unarmed -> provably inert.
+  if (armed !== true) return null;
+
+  // --- armed-true body: NEVER reached today (the flag is false). ---
+  if (!topReach || typeof topReach !== 'object') return null;
+  if (!cue || typeof cue !== 'object' || typeof cue.confidence !== 'number') return null;
+
+  // Below the protected-band floor -> defer to Brain (read the floor, never type it).
+  const { BEHAVIORAL_CHANNEL_FLOOR } = require('../workflow/f-selector-ranker.cjs');
+  if (cue.confidence < BEHAVIORAL_CHANNEL_FLOOR) return null;
+
+  // Map the fired reach to its canonical verb; a non-mapping reach defers.
+  const verb = reachIdToSkillFamily(topReach.reach_id);
+  if (verb === null) return null;
+
+  return {
+    verb: verb,
+    confidence: cue.confidence,
+    reason: 'behavioral_channel:' + topReach.reach_id + '/' + topReach.posture,
+  };
+}
+
+/**
+ * Resolve the best fire_skill given Brain signal state, the wicked escalation
+ * rule, and the Phase 143 sensor candidate reaches.
+ *
+ * Multi-signal precedence (Phase 144 -- the core design decision, documented):
+ *
+ *   1. wicked_escalation   wicked_score >= WICKED_ESCALATION_THRESHOLD (8)
+ *                          returns the soft-systems family FIRST (Canon
+ *                          Appendix E rule R4). Outranks everything.
+ *   2. sensor reach        the TOP reach by canonical REACH_IDS order maps via
+ *                          reachIdToSkillFamily to a CANONICAL VERB and fires in
+ *                          ANY TIER (tier_0 / mode_b included). This is the
+ *                          legacy -> engine flip: a non-null canonical verb makes
+ *                          the router validate true. NOT gated on mode_a.
+ *   3. BRAIN.md verb       the existing mode_a pattern_matches top verb path
+ *                          (unchanged; family-slug return, pre-144 behavior).
+ *   4. weightApplied>=0.9  the existing context-engine fresh-Brain fallback
+ *                          (unchanged).
+ *
+ * The sensor branch runs BEFORE the mode_a BRAIN gate so sensors fire even in
+ * tier_0/mode_b (the legacy-on-every-turn behavior is gone when a reach fires).
+ * dispatchSensors returning [] (or throwing -> caught upstream -> []) degrades
+ * resolveFireSkill EXACTLY to the pre-144 engine: no sensor contribution.
+ *
+ * Returns null when no signal warrants firing.
+ */
+function resolveFireSkill(brain, weightApplied, tierMode, sensorReaches) {
+  // (1) Wicked escalation takes precedence over normal routing (Canon
+  // Appendix E rule R4) -- it even outranks a fired sensor reach.
+  const wickedSection = brain && brain.sections && brain.sections.wicked_indicators;
+  if (wickedSection && wickedSection.body) {
+    const score = extractWickedScore(wickedSection.body);
+    if (score >= WICKED_ESCALATION_THRESHOLD) {
+      return 'soft-systems';
+    }
+  }
+
+  // (2) Sensor reach -> canonical verb (ANY tier). The top reach is already in
+  // canonical REACH_IDS order: dispatchSensors iterates SENSOR_REGISTRY in
+  // canonical order and the registry mirrors REACH_IDS, so sensorReaches[0] is
+  // the top reach by the documented tie-break. Map it to a canonical verb; a
+  // non-mapping reach falls through to the BRAIN path.
+  if (Array.isArray(sensorReaches) && sensorReaches.length > 0) {
+    const top = sensorReaches[0];
+    if (top && typeof top === 'object') {
+      const verb = reachIdToSkillFamily(top.reach_id);
+      if (verb !== null) {
+        return verb;
+      }
+    }
+  }
+
+  // (2.5) SEAM 3 -- the behavioral-channel modifier, WIRED-BUT-SHADOWED (BCH-07).
+  // Inserted between the sensor-reach branch (above) and the mode_a Brain path
+  // (below). The cue is a scalar { tier, confidence } drawn from the top reach
+  // the sensor spine already surfaced (enum/scalar handles ONLY -- never user
+  // prose; Part 8). BEHAVIORAL_CHANNEL_ARMED is false today, so the modifier
+  // returns null UNCONDITIONALLY and this branch is byte-identical to the
+  // pre-seam fall-through (the seam is dormant). When a non-null seam EVER
+  // returns (only after a Canon-Custodian-reviewed, data-gated arming), it
+  // routes to seam.verb. Today it cannot.
+  if (Array.isArray(sensorReaches) && sensorReaches.length > 0) {
+    const top = sensorReaches[0];
+    const cue = (top && typeof top === 'object')
+      ? { tier: top.tier, confidence: top.confidence }
+      : null;
+    const seam = behavioralChannelModifier(top, cue, readBehavioralChannelArmed());
+    if (seam !== null) {
+      return seam.verb;
+    }
+  }
+
+  // (3) Mode A + non-zero weight + parseable verb -> fire that verb's skill.
+  if (tierMode === 'mode_a' && weightApplied > 0) {
+    const pm = brain && brain.sections && brain.sections.pattern_matches;
+    if (pm && pm.body) {
+      const verb = extractTopCandidateVerb(pm.body);
+      if (verb !== null) {
+        // Map canonical verb to skill family. v1 ships a minimal mapping
+        // sufficient to satisfy "fire_skill non-null when fresh" (Test 20).
+        return verbToSkillFamily(verb);
+      }
+    }
+    // (4) Fresh Brain but no pattern_matches verb: fall back to context
+    // skill so Test 20 still has a non-null fire_skill.
+    if (weightApplied >= 0.9) {
+      return 'context-engine';
+    }
+  }
+  return null;
+}
+
+function verbToSkillFamily(verb) {
+  // Closed mapping. Future plans can extend the table; the engine
+  // never invents skill names from prose.
+  switch (verb) {
+    case 'Run Methodology': return 'methodology-router';
+    case 'Reformulate': return 'beautiful-question';
+    case 'Spawn Sub-Agent': return 'subagent-dispatcher';
+    case 'Navigate Graph': return 'graph-navigator';
+    case "Devil's Advocate": return 'red-team';
+    case 'Scenario Plan': return 'scenario-planner';
+    case 'Synthesize': return 'blue-hat';
+    case 'Bank Opportunity': return 'opportunity-bank';
+    case 'Defer': return 'defer';
+    case 'Free-Text': return 'larry-default';
+    default: return null;
+  }
+}
+
+/**
+ * Compose suppress_skills based on intent and persona signals. v1
+ * ships an empty array; future plans will populate from intent
+ * heuristics. Returning [] keeps Test 31 honoring the array invariant.
+ */
+function resolveSuppressSkills(_context, _brain) {
+  return [];
+}
+
+/**
+ * persona_updates: returns null when no threshold crossed; otherwise an
+ * object with allowed keys archetype/problem_type/venture_stage.
+ *
+ * v1 ships null (no auto-update). Plan 91-01 owns the durability layer.
+ */
+function resolvePersonaUpdates(_context, _brain) {
+  return null;
+}
+
+/**
+ * offer_next_step: returns null OR a six-key offer
+ * { command, framework, jtbd, confidence, reason, scope }. Max 1 per turn
+ * is enforced trivially (we never return more than one).
+ *
+ * v1 behavior is now LIVE (Phase 135): delegates to the abstention-gated
+ * resolver in lib/core/navigation-engine-offer.cjs, which returns exactly one
+ * calibrated offer or null computed from real local memory (operator-state x
+ * confidence-margin x rejection-backoff, plus SC2 graph neighborhood / SC4
+ * FEYNMAN temporal + MINTO governing thought + active JTBD). The resolver is
+ * LOCAL-ONLY and SYNCHRONOUS (A3): no awaited Brain packet here; _brain stays
+ * unused. Wrapped in try/catch so the engine never throws and the
+ * framework-chain-composer fallback (which fires only when offer_next_step is
+ * null) still applies on any failure.
+ */
+function resolveOfferNextStep(context, _brain) {
+  try {
+    const offerHelper = require('./navigation-engine-offer.cjs');
+    return offerHelper.resolveOffer(context);
+  } catch (_e) {
+    return null;
+  }
+}
+
+/**
+ * Phase 148-05 Task 1 (IRW-05): render the offer-resolver output through the ONE
+ * F.1 host. The offer is computed by resolveOfferNextStep (the calibration that
+ * stays untouched); the RENDER door is selector-dispatcher.pickShape via
+ * navigation-engine-offer.renderOfferThroughHost -- the IDENTICAL host the
+ * suggest-next surface uses. There is no second bespoke renderer on the engine
+ * side; both surfaces converge on pickShape. Never throws.
+ */
+function renderOfferNextStep(context, _brain) {
+  try {
+    const offerHelper = require('./navigation-engine-offer.cjs');
+    const offer = resolveOfferNextStep(context, _brain);
+    return offerHelper.renderOfferThroughHost(offer, context);
+  } catch (_e) {
+    return { shape: 'error', rendered: { error: 'render_offer_next_step_failed' } };
+  }
+}
+
+/**
+ * Phase 148-05 Task 1 (IRW-05): the suggest-next surface, unified onto the SAME
+ * F.1 host as the offer-resolver. Delegates to
+ * navigation-engine-offer.suggestNext so suggest-next and offer-resolver share
+ * one ranking and one render path (no divergent renderer). Never throws.
+ */
+function suggestNext(context, _brain) {
+  try {
+    const offerHelper = require('./navigation-engine-offer.cjs');
+    return offerHelper.suggestNext(context);
+  } catch (_e) {
+    return { shape: 'error', rendered: { error: 'suggest_next_failed' } };
+  }
+}
+
+/**
+ * Phase 191 (191-03, D-01/D-02/D-06): apply the lift module's candidate onto
+ * an ALREADY-COMPUTED decision + trace, in place. Fills decision.fire_skill
+ * ONLY when no higher-precedence signal (resolveFireSkill's step ladder, or
+ * the problem-type / chain-composer post-hoc contributions on the main path)
+ * has already set it -- the lift never overrides an existing verb. This
+ * keeps the wicked-escalation (step 1) and sensor-reach (step 2) precedence
+ * inside resolveFireSkill completely untouched (191-IFACE.md section 5), and
+ * treats the projection-sourced candidate as the LOWEST-precedence source in
+ * the ladder: it only fires when every other signal stayed silent this turn.
+ *
+ * Enriches trace.projection_offer IN PLACE with the lift's lifted_option +
+ * command_recommendation. B2 (R3): no new top-level decision or trace key is
+ * added here -- only the existing projection_offer object's own fields
+ * change, and only the existing fire_skill field's VALUE (never its
+ * existence or type contract, string-or-null).
+ *
+ * A null liftResult (a caught lift fault, R7) is a full no-op: fire_skill
+ * and projection_offer stay exactly what the pre-191 engine produced.
+ */
+function applyProjectionLift(decision, trace, liftResult) {
+  if (!liftResult || typeof liftResult !== 'object') return;
+  if (trace.projection_offer && typeof trace.projection_offer === 'object') {
+    trace.projection_offer.lifted_candidate = liftResult.lifted_option;
+    trace.projection_offer.command_recommendation = liftResult.command_recommendation;
+  }
+  if (decision.fire_skill === null && typeof liftResult.fire_skill_verb === 'string') {
+    decision.fire_skill = liftResult.fire_skill_verb;
+    // Part 8: name reach_id + command_slug + confidence ONLY -- never user
+    // prose, mirroring the existing sensor-reach rationale style.
+    const reachId = (liftResult.lifted_option && typeof liftResult.lifted_option.reach_id === 'string')
+      ? liftResult.lifted_option.reach_id : 'unknown';
+    const cr = liftResult.command_recommendation;
+    const commandSlug = (cr && typeof cr.command_slug === 'string') ? cr.command_slug : 'unknown';
+    const conf = (typeof liftResult.confidence === 'number') ? liftResult.confidence.toFixed(2) : '0.00';
+    const existing = trace.chosen_rationale || '';
+    const sep = existing.length > 0 ? ' ' : '';
+    trace.chosen_rationale = existing + sep +
+      'Projection candidate lifted: ' + reachId + '/' + commandSlug + ' (confidence ' + conf + ').';
+  }
+}
+
+// ---------- decide() main entry ----------
+
+/**
+ * decide(turn, context) -> decision
+ *
+ * Inputs:
+ *   turn = { userText, sectionPath, sessionId }
+ *   context = {
+ *     quadruple,        // pre-read quadruple (caller responsibility)
+ *     brainAvailable,   // boolean from brain-client.isAvailable()
+ *     userPersona,      // { archetype, problem_type, venture_stage }
+ *     intentSignal      // { intent, confidence }
+ *   }
+ *
+ * Returns: typed decision struct with complete trace.
+ *
+ * Never throws. Every internal failure falls through to a safe
+ * emptyDecision() with chosen_rationale noting the fault.
+ *
+ * Per-turn cache: a single quadruple lookup may be cached within this
+ * call only. This implementation expects context.quadruple to be
+ * pre-read (the standard Phase 91-02 hook integration pattern), so the
+ * cache is implicit. If context.quadruple is absent, decide() falls
+ * back to readQuadruple(turn.sectionPath) once and caches the result
+ * for the duration of THIS call only. Cross-call caching FORBIDDEN per
+ * Section 2.4.
+ */
+function decide(turn, context) {
+  // Phase 144 (Task 2): LOCAL monotonic clock for the decide()-body latency
+  // budget telemetry. Date.now() is LOCAL and never transmitted; the resulting
+  // _meta.latencies_ms is LOCAL trace JSON only (guards the 1200ms NAV budget).
+  const decideStartMs = Date.now();
+
+  // Safe defaults.
+  const ctx = context || {};
+  const t = turn || {};
+  const decision = shared.emptyDecision();
+  const trace = decision.decision_trace;
+
+  // Phase 144 (NAV-01): consume the Phase 143 sensors ONCE, before any return
+  // path, so EVERY decision (tier_0 / legacy / mode_a) sees the same candidate
+  // reaches and the legacy-on-every-turn behavior is gone when a reach fires.
+  // tuple = the /mos:diagnose classification { problem_type, complexity, stage }
+  // read from context (defaulting to {} per field when absent). ctx handles are
+  // LOCAL ({ roomDir, lowFillSections }). dispatchSensors is pure / sync /
+  // LOCAL-first; wrapped in try/catch so a sensor fault degrades to [] and
+  // resolveFireSkill behaves exactly like the pre-144 engine. Never throws.
+  const sensorTuple = {
+    problem_type: ctx.problem_type !== undefined ? ctx.problem_type : undefined,
+    complexity: ctx.complexity !== undefined ? ctx.complexity : undefined,
+    stage: ctx.stage !== undefined ? ctx.stage : undefined,
+  };
+  const sensorCtx = {
+    roomDir: typeof ctx.roomDir === 'string' ? ctx.roomDir : null,
+    lowFillSections: Array.isArray(ctx.lowFillSections) ? ctx.lowFillSections : null,
+  };
+  // MED-01 producer: derive the two LOCAL cortex ctx scalars sensorMemoryCortex
+  // reads (staleGoverningThought / freshContradictions) from the projected cortex
+  // the caller threaded onto context.roomContext.cortexNodes (legD). Presence +
+  // closed-enum reads ONLY -- a 'stale' governing_thought freshness enum or a
+  // 'contradiction' decision kind enum; never prose (Canon Part 8). Inlined here
+  // (rather than importing the hmi adapter) to keep the engine's dependency graph
+  // and Part-8 require-surface minimal. Caller-threaded ctx.* overrides win when
+  // explicitly provided (the test seam). Defensive; never throws.
+  {
+    const cortexNodes = (ctx.roomContext && Array.isArray(ctx.roomContext.cortexNodes))
+      ? ctx.roomContext.cortexNodes
+      : [];
+    let stale = false;
+    let fresh = 0;
+    for (const node of cortexNodes) {
+      if (!node || typeof node !== 'object') continue;
+      const props = (node.properties && typeof node.properties === 'object') ? node.properties : null;
+      if (!props) continue;
+      if (node.type === 'governing_thought' && props.freshness === 'stale') stale = true;
+      else if (node.type === 'decision' && props.kind === 'contradiction') fresh += 1;
+    }
+    sensorCtx.staleGoverningThought = (ctx.staleGoverningThought === true) || stale;
+    sensorCtx.freshContradictions =
+      (typeof ctx.freshContradictions === 'number' && ctx.freshContradictions > 0)
+        ? ctx.freshContradictions
+        : fresh;
+  }
+  // Phase 203-03 producer (SENS-11): derive the reusable-expert scalars
+  // sensorExpertSkill reads (reusableExpertCandidate / reusableExpertInvocationMax /
+  // reusableExpertTierSignal / reusableExpertCandidateCount) from CONFIRMED
+  // SyntheticExpert nodes in room.db, mirroring the MED-01 cortex producer above.
+  // The db read runs HERE (ctx-assembly), NEVER inside the pure sensor (Part 8/9).
+  // The engine does not OPEN room.db; it uses a caller-threaded handle (ctx.roomDb)
+  // when present, exactly as MED-01 consumes the caller-threaded cortex projection.
+  // The LOCAL candidate list (node ids) is threaded onto sensorCtx for the gate
+  // APPROVE handler ONLY -- it is never ridden on the reach. Caller-threaded ctx.*
+  // scalar overrides win (the test seam). Defensive; never throws.
+  {
+    let reusable = false;
+    let invocationMax = 0;
+    let tierSignal = false;
+    let candidateCount = 0;
+    const roomDb = (ctx.roomDb && typeof ctx.roomDb.prepare === 'function') ? ctx.roomDb : null;
+    if (roomDb) {
+      try {
+        const threshold = (typeof ctx.reusableExpertThreshold === 'number')
+          ? ctx.reusableExpertThreshold
+          : undefined;
+        const sig = detectExpertSkillCandidates(roomDb, threshold !== undefined ? { threshold: threshold } : {});
+        if (sig && sig.reusable === true) {
+          reusable = true;
+          invocationMax = sig.invocationMax;
+          tierSignal = sig.tierSignal;
+          candidateCount = Array.isArray(sig.candidates) ? sig.candidates.length : 0;
+          // LOCAL candidate list for the gate APPROVE handler only (never on the reach).
+          sensorCtx.reusableExpertCandidates = Array.isArray(sig.candidates) ? sig.candidates : [];
+        }
+      } catch (_e) {
+        // soft-fail: degrade to no reusable-expert signal
+      }
+    }
+    sensorCtx.reusableExpertCandidate = (ctx.reusableExpertCandidate === true) || reusable;
+    sensorCtx.reusableExpertInvocationMax =
+      (typeof ctx.reusableExpertInvocationMax === 'number' && ctx.reusableExpertInvocationMax > 0)
+        ? ctx.reusableExpertInvocationMax
+        : invocationMax;
+    sensorCtx.reusableExpertTierSignal = (ctx.reusableExpertTierSignal === true) || tierSignal;
+    sensorCtx.reusableExpertCandidateCount =
+      (typeof ctx.reusableExpertCandidateCount === 'number' && ctx.reusableExpertCandidateCount > 0)
+        ? ctx.reusableExpertCandidateCount
+        : candidateCount;
+  }
+  // Phase 244 Plan 05 producer (SENS-16, TRIG-01): derive the content-relevance
+  // scalars sensorContentRelevance reads (contentHitCount / contentCoverage /
+  // contentIndexState / contentCandidates) by querying the ALREADY-SHIPPED
+  // tri-modal-index.lexicalSearch (bm25 FTS5) against room.db, mirroring the
+  // MED-01 cortex producer and the SENS-11 expert producer above. The db read
+  // runs HERE (ctx-assembly), NEVER inside the pure sensor (Part 8/9). The
+  // engine does not OPEN room.db; it uses a caller-threaded handle (ctx.roomDb)
+  // when present, exactly as the SENS-11 producer consumes ctx.roomDb. The
+  // LOCAL candidate list (node ids) is threaded onto sensorCtx for the gate
+  // handler ONLY -- it is never ridden on the reach. Caller-threaded ctx.*
+  // scalar overrides win (the test seam that lets a unit test fire the
+  // sensor with no db, no FTS index, and no fixture at all). Defensive; never
+  // throws. lexicalSearch is sync and measured at 0-1 ms on a warm index, so
+  // this block is free against the 1200ms NAV budget; the expensive
+  // indexNodes call (async, reads artifact bodies off disk) NEVER runs here
+  // -- the producer only ENQUEUES a build request and spawns a detached
+  // drain, both sync (lib/core/eureka/fts-index-lifecycle.cjs).
+  {
+    let contentHitCount = 0;
+    let contentCoverage = 0;
+    let contentIndexState = 'unavailable';
+    const roomDb = (ctx.roomDb && typeof ctx.roomDb.prepare === 'function') ? ctx.roomDb : null;
+    if (roomDb) {
+      try {
+        // Full canonical text precedence (insight-sensors.cjs:347-352). The
+        // 244-RESEARCH.md example read only t.userText, which sees empty text
+        // on every caller that already normalizes -- corrected here.
+        const turnText = (typeof t.text === 'string' && t.text)
+          ? t.text
+          : (typeof t.utterance === 'string' && t.utterance)
+            ? t.utterance
+            : (typeof t.userText === 'string' && t.userText)
+              ? t.userText
+              : '';
+        const roomDirForContent = sensorCtx.roomDir
+          || (ctx.roomState && typeof ctx.roomState.roomDir === 'string' ? ctx.roomState.roomDir : null);
+        const sig = detectContentRelevance(roomDb, {
+          turnText: turnText,
+          roomDir: roomDirForContent || undefined,
+        });
+        if (sig && typeof sig === 'object') {
+          contentHitCount = (typeof sig.hits === 'number' && Number.isFinite(sig.hits)) ? sig.hits : 0;
+          contentCoverage = (typeof sig.coverage === 'number' && Number.isFinite(sig.coverage)) ? sig.coverage : 0;
+          contentIndexState = (typeof sig.state === 'string' && sig.state) ? sig.state : 'unavailable';
+          // LOCAL candidate list for the gate handler only (never on the reach).
+          sensorCtx.contentCandidates = Array.isArray(sig.candidates) ? sig.candidates : [];
+        }
+      } catch (_e) {
+        // soft-fail: degrade to no content-relevance signal
+      }
+    }
+    sensorCtx.contentHitCount =
+      (typeof ctx.contentHitCount === 'number' && ctx.contentHitCount >= 0)
+        ? ctx.contentHitCount
+        : contentHitCount;
+    sensorCtx.contentCoverage =
+      (typeof ctx.contentCoverage === 'number' && ctx.contentCoverage >= 0)
+        ? ctx.contentCoverage
+        : contentCoverage;
+    sensorCtx.contentIndexState =
+      (typeof ctx.contentIndexState === 'string' && ctx.contentIndexState)
+        ? ctx.contentIndexState
+        : contentIndexState;
+    if (!Array.isArray(sensorCtx.contentCandidates)) sensorCtx.contentCandidates = [];
+  }
+  let sensorReaches = [];
+  let dispatchSensorsMs = 0;
+  {
+    const dispatchStartMs = Date.now();
+    try {
+      const reaches = dispatchSensors(t, sensorTuple, sensorCtx);
+      sensorReaches = Array.isArray(reaches) ? reaches : [];
+    } catch (_e) {
+      sensorReaches = []; // soft-fail: degrade to pre-144 behavior
+    }
+    dispatchSensorsMs = Date.now() - dispatchStartMs;
+  }
+
+  // Phase 222 (D-01): re-order the fired subset by the shared scored selection
+  // BEFORE any [0] read - resolveFireSkill, the trace, and the rationale clause
+  // all read [0] and now transparently see the scored winner. Lazy-require inside
+  // decide() (the :510 f-selector-ranker idiom) keeps this engine-to-workflow
+  // require function-local, avoiding module-load cycles. Wrapped in try/catch that
+  // leaves sensorReaches untouched on any error (belt-and-suspenders on the hot
+  // path; the ranker also self-guards and passes 0/1-candidate arrays through by
+  // reference). The engine never OPENS room.db - it threads the caller's ctx.roomDb
+  // handle (may be null; the ranker then degrades to equal weights, no event).
+  try {
+    const reachHedgeRanker = require('../workflow/reach-hedge-ranker.cjs');
+    sensorReaches = reachHedgeRanker.rankFiredCandidates(sensorReaches, {
+      cortexNodes: (ctx.roomContext && Array.isArray(ctx.roomContext.cortexNodes))
+        ? ctx.roomContext.cortexNodes
+        : [],
+      db: ctx.roomDb || null,
+    });
+  } catch (_e) {
+    // soft-fail: leave the unranked fired subset in place
+  }
+
+  // CASC-02 (Phase 142): reflect the navigated neighborhood (getRoomContext Leg C,
+  // getNeighborhood-ranked) the caller threaded onto context.roomContext through the
+  // navigation.cjs chokepoint. Computed once, before any return path, so EVERY
+  // decision (Tier 0 / legacy / mode_a) carries it when present and the safe-default
+  // null when absent. Scalars/slugs/scores only (Part 8). Never throws.
+  let navigatedNeighborhood = null;
+  try {
+    navigatedNeighborhood = buildNavigatedNeighborhood(ctx.roomContext);
+  } catch (_e) {
+    navigatedNeighborhood = null;
+  }
+  // Reference the chokepoint export so the Part-8 anchor require is load-bearing
+  // (the navigated neighborhood ARRIVES through navigation.cjs's getRoomContext leg).
+  void navigation.getRoomContext;
+
+  // Phase 184 (READER-01..04): the decide-time projection OFFER. The THIRD READER --
+  // it RANKS the LOCAL orchestration-projection capabilities for this navigator
+  // context and SURFACES them as Shape F Decision-Gate OPTION CONTENT. It calls
+  // NEITHER runChain NOR any act-command: a reader, never a firer (READER-04, proved
+  // structurally by tests/test-reader-r4-structural-184.cjs). Computed ONCE before
+  // any return path so EVERY decision (tier_0 / legacy / mode_a) carries the same
+  // offer (or the null safe-default when skipped). The R2 gate inside the reader
+  // validates the projection BEFORE the read and FAILS CLOSED on a malformed
+  // projection (the read is skipped, decide() still returns). Part 8: the projection
+  // is a LOCAL derived machinery cache (a local file read); ZERO Brain read/write,
+  // ZERO network. Never throws -- a fault degrades to a null offer.
+  let projectionOffer = null;
+  try {
+    const projectionReader = require('./reader/decide-projection-reader.cjs');
+    projectionOffer = projectionReader.offerProjectionCapabilities(
+      {
+        reach_id: (sensorReaches[0] && typeof sensorReaches[0] === 'object'
+          ? sensorReaches[0].reach_id : null) || (typeof ctx.reach_id === 'string' ? ctx.reach_id : null),
+        firingSensors: Array.isArray(ctx.firingSensors) ? ctx.firingSensors : null,
+        sensorReaches: sensorReaches,
+      },
+      { projection: ctx.projectionOverride, projectionPath: ctx.projectionPath }
+    );
+  } catch (_e) {
+    projectionOffer = null;
+  }
+
+  // Phase 191 (191-03, D-01/D-02/D-06): lift the firing candidate from the
+  // projection offer computed above. Computed ONCE, before any return path
+  // (mirrors projectionOffer itself, above). liftFiringCandidate (Wave 2,
+  // orchestration-candidate-lift.cjs) is pure/sync/LOCAL (R2): it reads only
+  // the already-loaded LOCAL projectionOffer + the LOCAL command subgraph via
+  // its own rankFn seam (default f-selector-ranker.rankForSelector). D-01:
+  // the Phase-184 reader above is NOT touched by this wire; the lift consumes
+  // its output only. D-02: the gate is the lift module's own imported
+  // RECOMMENDED_CONFIDENCE_FLOOR -- never re-typed here. R7: never throws --
+  // a fault degrades to a null liftResult, which applyProjectionLift() below
+  // treats as a full no-op (pre-191 fire_skill + projection_offer untouched).
+  let liftResult = null;
+  try {
+    const candidateLift = require('./orchestration-candidate-lift.cjs');
+    liftResult = candidateLift.liftFiringCandidate({
+      projectionOffer: projectionOffer,
+      sensorReaches: sensorReaches,
+      context: {
+        roomState: ctx.roomState,
+        packetOptional: ctx.packetOptional,
+        problem_type: ctx.problem_type,
+      },
+    });
+  } catch (_e) {
+    liftResult = null;
+  }
+
+  try {
+    // Per-turn cache scope: read quadruple if not provided. The local
+    // `quadruple` binding is the cache; it never escapes this function.
+    let quadruple = ctx.quadruple;
+    if (!quadruple) {
+      const sectionPath = t.sectionPath;
+      if (typeof sectionPath === 'string' && sectionPath.length > 0) {
+        quadruple = folderMemory.readQuadruple(sectionPath);
+      }
+    }
+
+    // If we still have no quadruple, return empty decision in tier_0.
+    if (!quadruple) {
+      trace.brain_md_tier_mode = 'tier_0';
+      trace.brain_md_staleness = 'absent';
+      trace.brain_md_weight_applied = 0.0;
+      trace.icm_scope = buildIcmScope(null, t);
+      trace.sql_signals = buildSqlSignals(null, t);
+      trace.minto_reasoning = buildMintoReasoning(null);
+      trace.intent_persona = buildIntentPersona(ctx);
+      trace.navigated_neighborhood = navigatedNeighborhood;
+      // Phase 245 (245-05): the Brain pattern_matches observation rides the tier_0
+      // early-return path too, so EVERY decision carries both fields. There is no
+      // quadruple here, hence no brain, hence the explicit null observation. Set
+      // explicitly rather than left to the emptyDecisionTrace() default, so this
+      // path states its answer instead of inheriting one.
+      {
+        const bpv = brainPatternVerbObservation(null);
+        trace.brain_pattern_verb = bpv.verb;
+        trace.brain_pattern_verb_confidence = bpv.confidence;
+      }
+      // Phase 184 (READER): the decide-time projection offer rides the tier_0 path too.
+      trace.projection_offer = projectionOffer;
+      trace.chosen_rationale = 'Tier 0 fallback: no quadruple available for sectionPath.';
+      if (navigatedNeighborhood !== null) {
+        trace.chosen_rationale +=
+          ' Routing reflected the navigated neighborhood (' +
+          navigatedNeighborhood.count + ' ranked neighbor' +
+          (navigatedNeighborhood.count === 1 ? '' : 's') + ', top ' +
+          navigatedNeighborhood.top_node_id + ').';
+      }
+
+      // Phase 144 (NAV-01): sensors fire in ANY tier -- including this tier_0
+      // early-return path. A fired reach -> a CANONICAL VERB so the router flips
+      // routing_source to engine even with no quadruple.
+      decision.fire_skill = resolveFireSkill(null, 0.0, 'tier_0', sensorReaches);
+      if (sensorReaches.length > 0) {
+        const top = sensorReaches[0];
+        if (top && typeof top === 'object' && reachIdToSkillFamily(top.reach_id) !== null) {
+          // Part 8: name reach_id + posture ONLY. Never reach.signal's
+          // user-derived value, evidence bodies, or the dispatch handle.
+          trace.chosen_rationale +=
+            ' Sensor reach fired: ' + top.reach_id + '/' + top.posture + '.';
+        }
+      }
+
+      // Phase 191 (191-03): the tier_0 early-return path rides the same lift
+      // wire as the main path (191-IFACE.md section 5). A fault-safe no-op
+      // when liftResult is null (R7).
+      applyProjectionLift(decision, trace, liftResult);
+
+      // Phase 144 (Task 2): LOCAL trace.context_assembly + latency telemetry on
+      // the tier_0 early-return path too (both return paths carry the shape).
+      trace.context_assembly = buildContextAssembly(
+        navigatedNeighborhood,
+        sensorReaches,
+        groundingForFireSkill(decision.fire_skill, sensorReaches, null, navigatedNeighborhood)
+      );
+      trace._meta = {
+        latencies_ms: {
+          dispatchSensors_ms: dispatchSensorsMs,
+          decide_total_ms: Date.now() - decideStartMs,
+        },
+      };
+      return decision;
+    }
+
+    const brainAvailable = ctx.brainAvailable === true;
+    const brain = quadruple.brain || null;
+
+    // Section 5 tier-mode resolver (re-evaluated every turn; no cache).
+    const tierMode = shared.resolveTierMode(quadruple, brainAvailable);
+    trace.brain_md_tier_mode = tierMode;
+
+    // Section 4 staleness multiplier (BRAIN.md contribution scaler).
+    // The result is later assigned to trace.brain_md_weight_applied per
+    // Section 8.1. Multiplier source rows:
+    //   null brain                                -> 0.0
+    //   staleness:fresh                           -> 1.0
+    //   stale_reason:brain_offline                -> 0.9 (EXEMPT)
+    //   stale_reason:age_exceeded                 -> 0.7
+    //   stale_reason:brain_graph_version_mismatch -> 0.5
+    //   stale_reason:governing_thought_changed    -> 0.3
+    //   stale_reason:derivation_timeout           -> 0.2
+    //   stale_reason:parse_failed                 -> 0.0
+    //   staleness:unavailable                     -> 0.0
+    let weightApplied = shared.applyStalenessMultiplier(brain);
+
+    // Section 3.1 attribution guard. brain.author MUST equal 'brain';
+    // any other value indicates a Canon Part 2 attribution breach and
+    // demotes the BRAIN.md weight to 0.0 with a trace note. The
+    // resulting brain_md_weight_applied is observable in the decision
+    // trace by /mos:explain-decision (Plan 91-05).
+    let attributionBreach = false;
+    if (brain !== null && brain.author !== undefined && brain.author !== 'brain') {
+      attributionBreach = true;
+      weightApplied = 0.0; // brain_md_weight_applied demotion target
+    }
+    // Section 8.1 trace.brain_md_weight_applied final assignment.
+    trace.brain_md_weight_applied = weightApplied;
+
+    // Section 8 trace fields (always emitted, even with brain null).
+    if (brain === null) {
+      trace.brain_md_version = null;
+      trace.brain_md_staleness = 'absent';
+      trace.brain_md_stale_reason = null;
+    } else {
+      trace.brain_md_version = typeof brain.brain_graph_version === 'number'
+        ? brain.brain_graph_version
+        : null;
+      trace.brain_md_staleness = typeof brain.staleness === 'string'
+        ? brain.staleness
+        : 'absent';
+      trace.brain_md_stale_reason = typeof brain.stale_reason === 'string'
+        ? brain.stale_reason
+        : null;
+    }
+
+    // Section 3.2 sections_consumed list (non-null bodies only).
+    trace.brain_md_sections_consumed = consumedSections(brain);
+
+    // Section 6 RECOMMENDED gate evaluation (record confidence even
+    // when gate fails; render marker only when ALL conditions hold).
+    let recommendedConfidence = null;
+    let recommendedRendered = false;
+    if (brain !== null && brain.sections) {
+      const pm = brain.sections.pattern_matches;
+      if (pm && typeof pm.body === 'string') {
+        recommendedConfidence = extractHighestConfidence(pm.body);
+      }
+    }
+    trace.brain_md_recommended_confidence = recommendedConfidence;
+
+    if (
+      tierMode === 'mode_a' &&
+      !attributionBreach &&
+      recommendedConfidence !== null &&
+      recommendedConfidence >= RECOMMENDED_CONFIDENCE_FLOOR
+    ) {
+      // Section 6 condition 4: candidate verb must match a Canon Part 3
+      // vocabulary entry. extractTopCandidateVerb returns the resolved
+      // canonical verb or null.
+      const pm = brain.sections.pattern_matches;
+      const verb = extractTopCandidateVerb(pm.body);
+      if (verb !== null) {
+        recommendedRendered = true;
+      }
+    }
+    trace.brain_md_recommended_marker_rendered = recommendedRendered;
+
+    // Phase 245 (245-05, D-25 / 245-RESEARCH.md F-2): the Brain pattern_matches
+    // OBSERVATION on the populated (mode_a / mode_b / legacy) return path.
+    //
+    // Placed here, NOT inside the RECOMMENDED gate directly above, and NOT inside
+    // resolveFireSkill. The gate above discards its verb after using it as a
+    // boolean and only runs at all when tierMode === 'mode_a' && !attributionBreach
+    // && confidence >= 0.70. resolveFireSkill's step (3) is unreachable whenever a
+    // sensor fired, because step (2) returns early -- which is essentially every
+    // observed turn (D-24: the Brain verb is STARVED). This computation is
+    // deliberately ungated so Requirement 1's third fusion input EXISTS at all.
+    //
+    // resolveFireSkill is byte-unchanged by this plan. This adds an observation; it
+    // does not reorder routing. decision.fire_skill and
+    // trace.context_assembly.decision_grounding stay exactly as they were.
+    {
+      const bpv = brainPatternVerbObservation(brain);
+      trace.brain_pattern_verb = bpv.verb;
+      trace.brain_pattern_verb_confidence = bpv.confidence;
+    }
+
+    // Five-signal triangulation (structural; rule-based per D-02).
+    trace.icm_scope = buildIcmScope(quadruple, t);
+    trace.sql_signals = buildSqlSignals(quadruple, t);
+    trace.minto_reasoning = buildMintoReasoning(quadruple);
+    trace.intent_persona = buildIntentPersona(ctx);
+
+    // CASC-02: reflect the navigated neighborhood (computed once above). When a
+    // non-empty getNeighborhood-ranked neighborhood arrived via context.roomContext,
+    // the spine NAVIGATED -- the field carries the ranked neighbor scalars and the
+    // rationale notes it (below). routing_source is NOT assigned here; flipped at
+    // the router layer when fire_skill is non-null (Phase 144).
+    trace.navigated_neighborhood = navigatedNeighborhood;
+
+    // Phase 184 (READER-01..04): surface the decide-time projection offer (ranked
+    // LOCAL orchestration-projection capabilities as Shape F OPTION CONTENT). READER
+    // is a third READER, never a firer: it does not touch fire_skill, suppress_skills,
+    // or any frozen Part 3 contract -- it only populates this trace field.
+    trace.projection_offer = projectionOffer;
+
+    // Compose chosen_rationale from active rules.
+    const rationale = [];
+    if (tierMode === 'tier_0') {
+      if (brain === null) {
+        rationale.push('Tier 0 fallback: BRAIN.md absent.');
+      } else if (brain.parse_failed === true) {
+        rationale.push('brain_parse_failed fell through to tier_0.');
+      } else if (brain.staleness === 'unavailable') {
+        rationale.push('Tier 0 fallback: BRAIN.md staleness unavailable.');
+      } else {
+        rationale.push('Tier 0 fallback: brain unreachable and not offline-exempt.');
+      }
+    } else if (tierMode === 'mode_b') {
+      rationale.push('Mode B local-only: brain unreachable; brain_offline exemption applied (weight 0.9).');
+      rationale.push('mode_b suppresses RECOMMENDED per Canon Part 3.');
+    } else {
+      // mode_a
+      if (recommendedRendered) {
+        rationale.push('Mode A: RECOMMENDED rendered (confidence >= 0.7).');
+      } else if (recommendedConfidence !== null) {
+        rationale.push('Mode A: pattern_matches confidence ' + recommendedConfidence + ' below 0.7 floor; RECOMMENDED suppressed.');
+      } else {
+        rationale.push('Mode A: no pattern_matches candidate available.');
+      }
+    }
+    if (attributionBreach) {
+      rationale.unshift('canon_part_2_attribution_breach: brain.author is not "brain"; weight demoted to 0.0.');
+    }
+
+    // Wicked escalation note (added to rationale when triggered).
+    const wickedSection = brain && brain.sections && brain.sections.wicked_indicators;
+    if (wickedSection && wickedSection.body) {
+      const score = extractWickedScore(wickedSection.body);
+      if (score >= WICKED_ESCALATION_THRESHOLD) {
+        rationale.push('wicked_escalation: wicked_score >= ' + WICKED_ESCALATION_THRESHOLD + ' triggers soft-systems family.');
+      }
+    }
+
+    // CASC-02: append the navigated-neighborhood clause so /mos:explain-decision
+    // surfaces that routing reflected the graph walk (not just stored edges).
+    if (navigatedNeighborhood !== null) {
+      rationale.push(
+        'Routing reflected the navigated neighborhood (' +
+        navigatedNeighborhood.count + ' ranked neighbor' +
+        (navigatedNeighborhood.count === 1 ? '' : 's') + ', top ' +
+        navigatedNeighborhood.top_node_id + ').'
+      );
+    }
+
+    // Phase 144 (NAV-01): when a sensor reach drives fire_skill, append a clause
+    // naming reach_id + posture ONLY (Part 8: never reach.signal's user-derived
+    // value, evidence bodies, the dispatch handle, or any user content).
+    if (sensorReaches.length > 0) {
+      const topReach = sensorReaches[0];
+      if (topReach && typeof topReach === 'object' && reachIdToSkillFamily(topReach.reach_id) !== null) {
+        rationale.push('Sensor reach fired: ' + topReach.reach_id + '/' + topReach.posture + '.');
+      }
+    }
+
+    trace.chosen_rationale = rationale.join(' ');
+
+    // Compose decision outputs (rule-based, not weighted score). Phase 144: the
+    // sensor reaches are threaded in so a fired reach -> a CANONICAL VERB (ANY
+    // tier), flipping routing_source legacy -> engine at the router as a
+    // consequence.
+    decision.fire_skill = resolveFireSkill(brain, weightApplied, tierMode, sensorReaches);
+    decision.suppress_skills = resolveSuppressSkills(ctx, brain);
+    decision.persona_updates = resolvePersonaUpdates(ctx, brain);
+    decision.offer_next_step = resolveOfferNextStep(ctx, brain);
+
+    // Phase 91-07: problem-type routing contribution. Lazy-require so a
+    // missing module does not crash the engine -- problem-type routing
+    // BIASES the rationale (and may inform fire_skill when no
+    // higher-priority signal has set one), but is never load-bearing.
+    // Per locked decision D-08 + Canon Appendix E rule R4.
+    try {
+      if (brain !== null && brain.sections) {
+        const ptRouter = require('./problem-type-router.cjs');
+        const ptSection = brain.sections.problemtype_classification;
+        const wkSection = brain.sections.wicked_indicators;
+        const parsed = ptRouter.parseProblemTypeSection(ptSection);
+        let routing = null;
+        if (parsed !== null) {
+          routing = ptRouter.routeByProblemType(parsed.type, parsed.confidence);
+        }
+        const wickedEscalate = ptRouter.detectWickedEscalation(wkSection);
+        if (wickedEscalate) {
+          // Wicked override applies regardless of base type. If we have
+          // no base routing, synthesize a stub so applyWickedOverride
+          // has a baseline to overlay.
+          if (routing === null) {
+            routing = {
+              recommended_skills: [],
+              suppressed_skills: [],
+              reason: '',
+              wicked_override: false,
+            };
+          }
+          routing = ptRouter.applyWickedOverride(routing);
+        }
+
+        if (routing !== null && routing.reason && routing.reason.length > 0) {
+          // Append routing reason to chosen_rationale so /mos:explain-
+          // decision can surface the contribution.
+          const existing = trace.chosen_rationale || '';
+          const sep = existing.length > 0 ? ' ' : '';
+          trace.chosen_rationale = existing + sep + 'ProblemType routing: ' + routing.reason;
+
+          // If no higher-priority signal set fire_skill AND routing
+          // recommends a primary skill AND confidence is usable, set
+          // fire_skill to the primary recommendation. Wicked override
+          // always wins when present (its first skill is the soft-
+          // systems entry point).
+          const conf = parsed && typeof parsed.confidence === 'number' ? parsed.confidence : null;
+          const confOk = routing.wicked_override || (typeof conf === 'number' && conf >= 0.5);
+          if (
+            decision.fire_skill === null &&
+            confOk &&
+            Array.isArray(routing.recommended_skills) &&
+            routing.recommended_skills.length > 0
+          ) {
+            decision.fire_skill = routing.recommended_skills[0];
+          }
+        }
+      }
+    } catch (_e) {
+      // Routing is biasing-only; never block the decision on a routing
+      // module fault. Leave decision + trace untouched.
+    }
+
+    // Phase 91-08: framework-chain-composer contribution. Lazy-require
+    // so a missing module never crashes the engine -- chain composition
+    // is one signal among five and biases offer_next_step when no higher-
+    // priority signal has populated it. Per locked decision D-09 + Canon
+    // Part 2 Engine 1 + Appendix E. Reads BRAIN.md framework_chain_
+    // predictions section (FEEDS_INTO edges authored by Phase 90-01).
+    // Canon Part 8: zero Brain queries at engine time -- composer reads
+    // only the LOCAL section body passed in via quadruple.
+    try {
+      if (brain !== null && brain.sections) {
+        const chainComposer = require('./framework-chain-composer.cjs');
+        const chainSection = brain.sections.framework_chain_predictions;
+        if (chainSection !== null && chainSection !== undefined) {
+          const edges = chainComposer.parseFrameworkChainSection(chainSection);
+          if (edges.length > 0) {
+            const completed = chainComposer.detectCompletedFramework(
+              null, // roomDir is not needed when reasoning carries the signal
+              t.sectionPath || null,
+              quadruple.reasoning || null
+            );
+            if (completed !== null) {
+              const proposal = chainComposer.proposeNextFramework(completed, edges);
+              if (proposal !== null) {
+                // User override detection: if lastTurnOffer was the chain
+                // proposal AND current turn invokes a DIFFERENT /mos:
+                // command, record REJECTED chain suggestion as graph data
+                // per Canon Part 4. The override check runs BEFORE we
+                // overwrite offer_next_step so the user's intent (in
+                // userText) gets recorded against the prior offer.
+                const lastOffer = ctx.lastTurnOffer;
+                const userText = typeof t.userText === 'string' ? t.userText : '';
+                if (
+                  lastOffer &&
+                  typeof lastOffer === 'object' &&
+                  typeof lastOffer.command === 'string' &&
+                  lastOffer.command.length > 0 &&
+                  userText.length > 0
+                ) {
+                  // Match a /mos: command in the user text.
+                  const userCmdMatch = /\/mos:[a-z][a-z0-9-]*/i.exec(userText);
+                  if (userCmdMatch !== null) {
+                    const userCmd = userCmdMatch[0];
+                    if (userCmd !== lastOffer.command) {
+                      // Override recorded as graph data per Canon Part 4.
+                      const existing = trace.chosen_rationale || '';
+                      const sep = existing.length > 0 ? ' ' : '';
+                      trace.chosen_rationale =
+                        existing + sep +
+                        'REJECTED chain suggestion: user invoked ' + userCmd +
+                        ' instead of ' + lastOffer.command + '.';
+                      trace.chain_override_recorded = true;
+                    }
+                  }
+                }
+
+                // Set offer_next_step ONLY when no higher-priority signal
+                // has already populated it. Chain composition is one
+                // signal among five; it never overwrites a stronger one.
+                if (decision.offer_next_step === null) {
+                  decision.offer_next_step = {
+                    command: proposal.command,
+                    reason: proposal.reason,
+                  };
+                  // Surface chain-specific eligibility flag in trace so
+                  // offer-presenter (Plan 91-04) can render the
+                  // RECOMMENDED marker per Canon Part 3 Section 6 when
+                  // Mode A holds AND chain confidence >= 0.7.
+                  trace.chain_recommended_eligible = proposal.recommended_eligible === true;
+                  // When the engine's RECOMMENDED gate already fired
+                  // (e.g. via pattern_matches), the chain proposal piggy-
+                  // backs on the existing marker. When pattern_matches
+                  // had no candidate (the common case in Wave 3 fixtures)
+                  // and Mode A holds and the chain proposal is eligible,
+                  // promote the marker so the presenter renders it.
+                  if (
+                    tierMode === 'mode_a' &&
+                    proposal.recommended_eligible === true &&
+                    trace.brain_md_recommended_marker_rendered === false
+                  ) {
+                    trace.brain_md_recommended_marker_rendered = true;
+                    trace.brain_md_recommended_confidence =
+                      typeof trace.brain_md_recommended_confidence === 'number' &&
+                      trace.brain_md_recommended_confidence > proposal.confidence
+                        ? trace.brain_md_recommended_confidence
+                        : proposal.confidence;
+                  }
+                  // Append chain rationale for /mos:explain-decision.
+                  const existing2 = trace.chosen_rationale || '';
+                  const sep2 = existing2.length > 0 ? ' ' : '';
+                  trace.chosen_rationale =
+                    existing2 + sep2 +
+                    'Framework chain: ' + completed + ' FEEDS_INTO ' + proposal.next +
+                    ' (Brain confidence ' + proposal.confidence.toFixed(2) + ').';
+                }
+              }
+            }
+          }
+        }
+      }
+    } catch (_e) {
+      // Chain composition is biasing-only; never block the decision on
+      // a composer fault. Leave decision + trace untouched.
+    }
+
+    // Phase 191 (191-03): the main-path lift application, placed AFTER every
+    // existing fire_skill contribution (sensor / brain / problem-type /
+    // chain) so the projection-sourced candidate is the LOWEST-precedence
+    // source in the ladder -- it only fires when every other signal stayed
+    // silent this turn (191-IFACE.md section 5). A fault-safe no-op when
+    // liftResult is null (R7).
+    applyProjectionLift(decision, trace, liftResult);
+
+    // Phase 144 (Task 2): the Zep-shaped LOCAL trace.context_assembly + LOCAL
+    // latency telemetry on the populated return path. Computed AFTER all
+    // fire_skill contributions (sensor / brain / problem-type / chain) so
+    // decision_grounding names the FINAL justification. LOCAL trace JSON only;
+    // nothing reaches the Brain. No await / I/O (pure scalar shaping) -- the
+    // 1200ms NAV budget is untouched.
+    trace.context_assembly = buildContextAssembly(
+      navigatedNeighborhood,
+      sensorReaches,
+      groundingForFireSkill(decision.fire_skill, sensorReaches, brain, navigatedNeighborhood)
+    );
+    trace._meta = {
+      latencies_ms: {
+        dispatchSensors_ms: dispatchSensorsMs,
+        decide_total_ms: Date.now() - decideStartMs,
+      },
+    };
+
+    return decision;
+  } catch (err) {
+    // Graceful degradation: never throw out of decide().
+    const safe = shared.emptyDecision();
+    safe.decision_trace.chosen_rationale =
+      'engine_fault: ' + (err && err.message ? err.message : String(err));
+    safe.decision_trace.brain_md_tier_mode = 'tier_0';
+    safe.decision_trace.brain_md_staleness = 'absent';
+    return safe;
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Phase 205 (item 7) -- the frozen-six drift-guard READ-ONLY accessor.
+//
+// The whole 205 phase rides on ONE floor invariant: GRILL / REFRAME / FUSION are
+// nav-dial POSITIONS and existing-reach routing, they mint NO new reach_id. This
+// accessor gives the drift guard a single read point ON THE decide() SPINE MODULE
+// to assert the frozen constitutional set holds after the 205 machinery runs. It
+// is strictly READ-ONLY: it re-reads the canonical values from their source
+// modules and returns a FROZEN snapshot, changing no value and adding no reach.
+//   - REACH_IDS         : the six frozen reach_ids (sensor-types.cjs)
+//   - MAX_K             : the chooser-row cap = 3 (f-selector-ranker.cjs)
+//   - DIAL_REACH_K      : the reach-bank size = 6 (dial-reach-orchestrator.cjs)
+//   - RECOMMEND_FLOOR   : the 0.70 recommend detent (dial-reach-orchestrator.cjs)
+//   - MARGIN_THRESHOLD  : the 0.15 margin detent  (dial-reach-orchestrator.cjs)
+// A require fault degrades to a null field (the guard then fails loudly, which is
+// the correct behavior for a tamper/floor test).
+function frozenConstitutionalSet() {
+  let reachIds = null;
+  let maxK = null;
+  let dialReachK = null;
+  let recommendFloor = null;
+  let marginThreshold = null;
+  try {
+    const st = require('./sensors/sensor-types.cjs');
+    reachIds = Array.isArray(st.REACH_IDS) ? Object.freeze(st.REACH_IDS.slice()) : null;
+  } catch (_e) { reachIds = null; }
+  try {
+    const ranker = require('../workflow/f-selector-ranker.cjs');
+    maxK = ranker.MAX_K;
+  } catch (_e) { maxK = null; }
+  try {
+    const orch = require('../hmi/dial-reach-orchestrator.cjs');
+    dialReachK = orch.DIAL_REACH_K;
+    recommendFloor = orch.RECOMMEND_FLOOR;
+    marginThreshold = orch.MARGIN_THRESHOLD;
+  } catch (_e) { /* leave nulls */ }
+  return Object.freeze({
+    REACH_IDS: reachIds,
+    MAX_K: maxK,
+    DIAL_REACH_K: dialReachK,
+    RECOMMEND_FLOOR: recommendFloor,
+    MARGIN_THRESHOLD: marginThreshold,
+  });
+}
+
+module.exports = {
+  decide: decide,
+  // Phase 205 (item 7): the read-only frozen-six drift-guard accessor.
+  frozenConstitutionalSet: frozenConstitutionalSet,
+  // Phase 177-01 (BCH-06): the Brain RECOMMENDED-gate confidence floor (0.70),
+  // exported so downstream engine-side consumers can COMPUTE thresholds relative
+  // to it rather than hand-typing the number. The behavioral-channel ceiling in
+  // lib/workflow/f-selector-ranker.cjs derives from this const (ceiling = floor +
+  // 0.15 == 0.85), so the >= 15-point invariant is computed, not duplicated.
+  RECOMMENDED_CONFIDENCE_FLOOR: RECOMMENDED_CONFIDENCE_FLOOR,
+  // Phase 148-05 (IRW-05): offer-resolver + suggest-next unified onto the ONE
+  // selector-dispatcher.pickShape F.1 host (single render path, no bespoke
+  // renderer).
+  resolveOfferNextStep: resolveOfferNextStep,
+  renderOfferNextStep: renderOfferNextStep,
+  suggestNext: suggestNext,
+  // Internal helpers exposed for downstream Phase 91 plans + testing:
+  reachIdToSkillFamily: reachIdToSkillFamily,
+  // Phase 172-05: exposed so the hats engine-flip contract test can exercise the
+  // sensor-reach -> canonical-verb resolution directly (resolveFireSkill step 2).
+  resolveFireSkill: resolveFireSkill,
+  // Phase 177-09 (BCH-07 / BCH-08): the SEAM 3 behavioral-channel modifier + its
+  // two pure helpers, exported so the BCH-07 dormant-byte-identical suite and the
+  // BCH-08 tier-ordering suite can assert the contract WITHOUT arming the flag.
+  behavioralChannelModifier: behavioralChannelModifier,
+  rankBehavioralCue: rankBehavioralCue,
+  bandCueLosesToBrain: bandCueLosesToBrain,
+  resolveTierMode: shared.resolveTierMode,
+  applyStalenessMultiplier: shared.applyStalenessMultiplier,
+  evaluateRecommendedGate: function (decision, quadruple) {
+    // Convenience accessor matching plan frontmatter contract.
+    return decision &&
+      decision.decision_trace &&
+      decision.decision_trace.brain_md_recommended_marker_rendered === true;
+  },
+};