@hviana/sema 0.4.2 → 0.4.4

package/dist/src/mind/traverse.js

@@ -0,0 +1,650 @@
+// traverse.ts — Traverse primitives + disambiguation (Sections 1 & 6 of the mind).
+//
+//   Traverse — node → nodes   (edgeAncestors, nextOf, prevOf, contains,
+//                               guidedNext, chooseNext, chooseAmong, hubCap)
+//
+// The PROJECTIONS built on these walks (follow, conceptHop, reverseContext,
+// project) live in match.ts — the elementary match-and-project operation.
+import { cosine } from "../vec.js";
+import { gistOf, read } from "./primitives.js";
+const structCaches = new WeakMap();
+// ── The shared ancestor-reach memo ──────────────────────────────────────
+//
+// `edgeAncestors` is a pure function of (node, N) over a read-only store —
+// asking never writes — so its result is reusable for as long as the store
+// is not written to.  There used to be TWO memos and they never met: the
+// climb built a private one per call (computeAttention), while
+// `Precomputed.reachMemo` — documented as "one response-scoped memo serves
+// every mechanism that prices commonality" — was reached only by confluence.
+// The climb is by far the biggest consumer.
+//
+// Keyed by the Mind's structural lifecycle identity: ordinary and
+// conversational asks share it, and every ingest invalidates it. A real
+// battery repeatedly reaches the same corpus scaffolding even when its
+// surface questions differ.
+//
+// Budgeted, not unbounded (AGENTS §2.12): past the cap the whole map is
+// dropped and re-derived, costing a cold climb and never a wrong answer.
+const REACH_MEMO_MAX = 100_000;
+const reachCaches = new WeakMap();
+/** The reach memo this ask should use — see the note above.
+ *
+ *  A TRACED response always gets a fresh, empty one.  `AncestorReach`'s
+ *  `visited`/`maxDepth`/`saturation` fields are populated only when a trace
+ *  is attached, so an entry deposited by an untraced earlier turn would
+ *  silently black out the reach detail of a later traced one; and the trace's
+ *  reach payload is serialised by ITERATING this map, which must therefore
+ *  hold what THIS climb consulted, not the whole conversation's history.
+ *  Consistent with AGENTS §2.11: a traced response is a different machine —
+ *  never benchmark with a trace attached. */
+export function sharedReachMemo(ctx) {
+    if (ctx.trace !== null || ctx.climbMemo === null)
+        return new Map();
+    let m = reachCaches.get(ctx._structMemoKey);
+    if (m === undefined)
+        reachCaches.set(ctx._structMemoKey, m = new Map());
+    else if (m.size >= REACH_MEMO_MAX)
+        m.clear();
+    return m;
+}
+function getStructCache(ctx) {
+    if (ctx.climbMemo === null)
+        return null;
+    let c = structCaches.get(ctx._structMemoKey);
+    if (c === undefined) {
+        structCaches.set(ctx._structMemoKey, c = {
+            hasNext: new Map(),
+            prevCount: new Map(),
+            hasParents: new Map(),
+        });
+    }
+    return c;
+}
+/** Invalidate every session-lifetime structural read after a write. */
+export function invalidateStructuralCaches(ctx) {
+    reachCaches.delete(ctx._structMemoKey);
+    structCaches.delete(ctx._structMemoKey);
+}
+/** Cached {@link Store.hasNext} — pure during one respond(). */
+function cachedHasNext(ctx, id, cache) {
+    if (cache === null)
+        return ctx.store.hasNext(id);
+    let v = cache.hasNext.get(id);
+    if (v === undefined) {
+        v = ctx.store.hasNext(id);
+        cache.hasNext.set(id, v);
+    }
+    return v;
+}
+/** Cached {@link Store.prevCount} — pure during one respond(). */
+function cachedPrevCount(ctx, id, cache) {
+    if (cache === null)
+        return ctx.store.prevCount(id);
+    let v = cache.prevCount.get(id);
+    if (v === undefined) {
+        v = ctx.store.prevCount(id);
+        cache.prevCount.set(id, v);
+    }
+    return v;
+}
+/** Cached {@link Store.hasParents} — pure during one respond(). */
+function cachedHasParents(ctx, id, cache) {
+    if (cache === null)
+        return ctx.store.hasParents(id);
+    let v = cache.hasParents.get(id);
+    if (v === undefined) {
+        v = ctx.store.hasParents(id);
+        cache.hasParents.set(id, v);
+    }
+    return v;
+}
+// ── Graph climbing ───────────────────────────────────────────────────────
+/** Climb the structural DAG from a node to its edge-bearing ancestor contexts.
+ *  Ascent stops at hub nodes (parents > √N) — their reach is non-discriminative.
+ *  When the start node has no structural parents, climbs from containment parents
+ *  (sub-span flat branches inheriting their chunks' context).
+ *
+ *  `memo`, when given, caches whole climbs by start id for the duration of ONE
+ *  query (the store is read-only while a query is in flight, so a climb is a
+ *  pure function of the id).  The consensus pipeline climbs the SAME anchors
+ *  repeatedly — regions sharing a chunk, and canonicalChunkId probing each
+ *  chunk's prefixes — so without the memo every repeat re-pays the full
+ *  fan-out reads. */
+export function edgeAncestors(ctx, id, contextCount, memo) {
+    const hit = memo?.get(id);
+    if (hit !== undefined)
+        return hit;
+    // BYTE-ATOM COMMONALITY.  A single-byte leaf (implicit negative id) has no
+    // structural parents BY CONSTRUCTION — atoms are never linked into the kid
+    // or contain tables — so this climb cannot observe its containment at all.
+    // The walk below would see only the atom's own direct edges and report
+    // contextsReached ≈ 1, turning the MOST common content in the store into
+    // the MOST discriminative voter (observed on a 325K-context store: every
+    // recognised single-letter site voted full ln N for the one fact whose
+    // continuation is that letter, and their pooled sum out-voted every
+    // genuine anchor).  An unmeasurable containment must not default to
+    // "maximally rare": it is bounded below by the uniform expectation over
+    // the byte alphabet — N contexts, each at least one chunk of up to W of
+    // the 256 possible atoms, reach ≥ N·W/256 contexts per atom on average
+    // (see {@link atomReach}).  When that floor itself exceeds the hub bound
+    // √N the atom is a hub at this corpus scale and the climb abstains
+    // (saturated) — the atom's own edges remain fully traversable (tier-0
+    // exact recall, chooseNext, project); only its say as a consensus voter
+    // is withdrawn.  On a small store the floor stays ≤ √N and the atom
+    // climbs exactly as before, so single-letter facts keep working.
+    if (id < 0 && atomIsHub(ctx, contextCount)) {
+        const bound0 = boundFor(contextCount);
+        const reach = {
+            roots: [],
+            contextsReached: 0,
+            saturated: true,
+            ...(ctx.trace
+                ? {
+                    saturation: {
+                        reason: "byte-atom-commonality",
+                        node: id,
+                        observed: atomReach(ctx, contextCount),
+                        limit: bound0,
+                    },
+                    visited: 0,
+                    maxDepth: 0,
+                }
+                : {}),
+        };
+        memo?.set(id, reach);
+        return reach;
+    }
+    const bound = boundFor(contextCount);
+    const roots = [];
+    const seen = new Set([id]);
+    const ctxSeen = new Set();
+    let saturated = false;
+    // Provenance of the FIRST decision that saturated this climb — allocated
+    // only when a trace is requested (see AncestorReach.saturation's doc); the
+    // climb itself never reads it back.
+    let satStop;
+    // EXPAND-UNTIL-DECIDED: a reach is consumed either as a VOTE (which needs
+    // contextsReached exactly, and only while ≤ √N — beyond that the region is
+    // non-discriminative) or as an ABSTENTION (saturated — whose roots and
+    // counts no consumer reads).  So the climb may STOP the moment the answer
+    // is decided:
+    //   • a node whose prev fan-in alone exceeds √N decides it (its
+    //     predecessors are √N+ distinct contexts) — no read needed, prevCount
+    //     is an indexed O(1);
+    //   • distinct contexts crossing √N decides it;
+    //   • a node with more than √N parents decides its own expansion (the
+    //     classic hub guard; the walk aborts rather than continue, which no
+    //     consumer can distinguish — saturated reaches are never voted).
+    // Below every decision threshold the walk is EXACT — identical roots and
+    // contexts to the unbounded climb — because prevFirst(√N) IS the full prev
+    // list and parentsFirst(√N+1) IS the full parent list whenever they do not
+    // decide.  Work is bounded by √N contexts × the climb's local structure,
+    // never by the corpus.
+    const structCache = getStructCache(ctx);
+    // LATERAL-BRANCH ACCOUNT — the cumulative dual of the per-node hub guard.
+    // Within one deposit the ascent is a CHAIN (each node's first parent);
+    // every parent BEYOND a node's first is an entry into another containing
+    // structure (hash-consing: a shared subtree's extra parents are other
+    // deposits' chunks).  The per-node guard already declares a node with more
+    // than √N parents non-discriminative; a climb whose ACCUMULATED lateral
+    // entries exceed √N has spread across just as many distinct containing
+    // structures — the same commonness, distributed along the cone instead of
+    // concentrated at one node — and is decided: saturated.  A deep chain in
+    // ONE structure accrues no laterals, so legitimate deep scaffolding (a
+    // fragment far down a long cumulative context) still climbs to its root
+    // at any depth; what dies is the cross-structure drift that visited tens
+    // of thousands of edge-free interiors (profiled on a 17.7M-node store:
+    // ~20K distinct nodes per climb family, >95% unique — not memoisable)
+    // while the context account never decided.
+    let lateral = 0;
+    // CLIMB READ-OUT (pure instrumentation, same contract as satStop): the
+    // parallel `depths` stack mirrors every push/pop of `stack`, so a node's
+    // ascent distance is known at its pop.  Allocated only when a trace is
+    // requested; the climb itself never reads any of these back.
+    const depths = ctx.trace ? [] : null;
+    let curDepth = 0;
+    let visitedCount = 0;
+    let maxDepth = 0;
+    const visit = (x) => {
+        if (ctx.meter)
+            ctx.meter.ancestorVisits++;
+        if (depths) {
+            visitedCount++;
+            if (curDepth > maxDepth)
+                maxDepth = curDepth;
+        }
+        const hasNx = cachedHasNext(ctx, x, structCache);
+        const pc = cachedPrevCount(ctx, x, structCache);
+        if (hasNx || pc > 0) {
+            roots.push(x);
+            if (hasNx)
+                ctxSeen.add(x);
+            if (pc > bound) {
+                // decided: ≥ pc > √N distinct contexts
+                if (ctx.trace) {
+                    satStop = {
+                        reason: "predecessor-fan-in",
+                        node: x,
+                        observed: pc,
+                        limit: bound,
+                    };
+                }
+                return false;
+            }
+            for (const p of ctx.store.prevFirst(x, bound))
+                ctxSeen.add(p);
+            if (ctxSeen.size > bound) {
+                // decided
+                if (ctx.trace) {
+                    satStop = {
+                        reason: "distinct-context-limit",
+                        node: x,
+                        observed: ctxSeen.size,
+                        limit: bound,
+                    };
+                }
+                return false;
+            }
+        }
+        const parents = ctx.store.parentsFirst(x, bound + 1);
+        if (parents.length > bound) {
+            // decided: hub
+            if (ctx.trace) {
+                satStop = {
+                    reason: "parent-fan-out",
+                    node: x,
+                    observed: parents.length,
+                    limit: bound,
+                };
+            }
+            return false;
+        }
+        let fresh = 0;
+        for (const p of parents) {
+            if (!seen.has(p)) {
+                seen.add(p);
+                stack.push(p);
+                depths?.push(curDepth + 1);
+                fresh++;
+            }
+        }
+        if (fresh > 1) {
+            lateral += fresh - 1;
+            if (lateral > bound) {
+                // decided: cone-wide hub
+                if (ctx.trace) {
+                    satStop = {
+                        reason: "lateral-cone-limit",
+                        node: x,
+                        observed: lateral,
+                        limit: bound,
+                    };
+                }
+                return false;
+            }
+        }
+        return true;
+    };
+    const stack = [];
+    const containment = !cachedHasParents(ctx, id, structCache);
+    if (!containment) {
+        stack.push(id);
+        depths?.push(0);
+    }
+    // The containment seed is STREAMED in pages of √N: a distinctive window's
+    // containers (which converge on one or two contexts, however many chunks
+    // of one deposit repeat it) are walked IN FULL — exact — while a common
+    // window's corpus-sized container list is abandoned at the first decision
+    // above, after O(√N) pages at most (each page adds containers whose climbs
+    // add contexts; √N distinct contexts decide).
+    let containerOff = 0;
+    let containersExhausted = !containment;
+    climb: for (;;) {
+        if (stack.length === 0) {
+            if (containersExhausted)
+                break;
+            const page = ctx.store.containersSlice(id, containerOff, bound);
+            containerOff += page.length;
+            if (page.length < bound)
+                containersExhausted = true;
+            for (const c of page) {
+                if (!seen.has(c)) {
+                    seen.add(c);
+                    stack.push(c);
+                    depths?.push(1);
+                }
+            }
+            if (stack.length === 0) {
+                if (containerOff === 0) {
+                    stack.push(id); // no containers at all
+                    depths?.push(0);
+                }
+                else
+                    break;
+            }
+        }
+        while (stack.length > 0) {
+            let x = stack.pop();
+            if (depths)
+                curDepth = depths.pop();
+            // TRANSPARENT-CHAIN HOP: a node with no edges in or out and exactly one
+            // parent contributes nothing here — no root, no context, no lateral
+            // entry — so the run to its first non-transparent ancestor is skipped
+            // in ONE store read (Store.chainRun) instead of three probes per node.
+            // The interior nodes still enter `seen`, exactly as a node-at-a-time
+            // ascent would have recorded them at push time, so sibling entries into
+            // the same chain keep identical fresh/lateral accounting; and if the
+            // terminal was already seen (another chain merged into this one first),
+            // it is not visited twice — the same dedup the push-time seen-check
+            // used to provide.
+            const run = ctx.store.chainRun(x);
+            if (run.length > 1) {
+                const top = run[run.length - 1];
+                const dup = seen.has(top);
+                for (let i = 1; i < run.length; i++)
+                    seen.add(run[i]);
+                if (dup)
+                    continue;
+                x = top;
+                // The chain's interior hops are part of the terminal's ascent
+                // distance — count them exactly as a node-at-a-time ascent would.
+                if (depths)
+                    curDepth += run.length - 1;
+            }
+            if (!visit(x)) {
+                saturated = true;
+                break climb;
+            }
+        }
+    }
+    const reach = {
+        roots,
+        contextsReached: ctxSeen.size,
+        saturated,
+        ...(saturated && satStop ? { saturation: satStop } : {}),
+        ...(depths ? { visited: visitedCount, maxDepth } : {}),
+    };
+    memo?.set(id, reach);
+    return reach;
+}
+/** Convenience: forward edges of a node. */
+export function nextOf(ctx, id) {
+    return ctx.store.next(id);
+}
+/** Convenience: reverse edges of a node. */
+export function prevOf(ctx, id) {
+    return ctx.store.prev(id);
+}
+/** The uniform-expectation floor on a byte atom's corpus commonality: N
+ *  learnt contexts, each at least one perception chunk of up to W of the 256
+ *  possible byte values, contain a given atom in ≥ N·W/256 contexts on
+ *  average.  An atom's TRUE containment is unmeasurable (atoms carry no
+ *  kid/contain links by construction), so this floor is the honest stand-in:
+ *  derived entirely from the corpus scale N, the perception window W, and
+ *  the alphabet size — never tuned. */
+export function atomReach(ctx, contextCount) {
+    return Math.max(1, Math.ceil((contextCount * ctx.space.maxGroup) / 256));
+}
+/** Whether a byte atom is a hub at this corpus scale — its commonality floor
+ *  {@link atomReach} exceeds the hub bound √N.  Below it (small stores) an
+ *  atom votes and is recognised exactly as any stored form; above it the
+ *  alphabet is scaffolding everywhere and abstains. */
+export function atomIsHub(ctx, contextCount) {
+    return atomReach(ctx, contextCount) > boundFor(contextCount);
+}
+/** Whether a node LEADS SOMEWHERE — it bears a continuation edge or a halo.
+ *  The admission predicate recognition filters sites with (HOW_IT_WORKS
+ *  §15.3): a form that leads nowhere contributes nothing to any derivation.
+ *  Runs once per candidate span on the recognition hot path — `hasNext` is
+ *  cached per response (the same flat-branch ids are probed across prefix
+ *  variants by canonicalChunkId).  `hasHalo` is not cached: it's a single
+ *  indexed point probe per candidate, and the candidates that reach this
+ *  check have already been filtered by hasNext above in edgeAncestors. */
+export function leadsSomewhere(ctx, id) {
+    const memo = getStructCache(ctx);
+    if (cachedHasNext(ctx, id, memo))
+        return true;
+    return ctx.store.hasHalo(id);
+}
+/** The structural IDF read of ONE node: how many distinct learnt contexts
+ *  its containment/edge climb reaches, or Infinity when it reaches none or
+ *  saturates (no usable identity evidence).  The number every
+ *  discriminative-vs-scaffolding decision derives from — paired with the
+ *  half-dominance convention (geometry.dominates(reach, N)): content
+ *  reaching a corpus MINORITY of contexts discriminates (an entity, a
+ *  filler); content reaching a majority is frame scaffolding. */
+export function reachOf(ctx, id, contextCount, memo) {
+    const r = edgeAncestors(ctx, id, contextCount, memo);
+    if (r.saturated || r.roots.length === 0)
+        return Infinity;
+    return Math.max(1, r.contextsReached);
+}
+/** The corpus scale N — the count of DISTINCT learnt contexts, floored at 2
+ *  so its derived readings (ln N in the consensus floor, √N in the hub bound)
+ *  stay meaningful on a near-empty store.  The one definition every consumer
+ *  of "how big is this corpus?" reads. */
+export function corpusN(ctx) {
+    return Math.max(2, ctx.store.edgeSourceCount());
+}
+/** The hub bound √N itself (≥ 2 always, since N is floored at 2) — for
+ *  consumers that pass it to the store's LIMITed reads instead of capping a
+ *  materialised list.  {@link hubCap} is the list-side reading of the same
+ *  convention. */
+export function hubBound(ctx) {
+    return boundFor(corpusN(ctx));
+}
+/** √N for an EXPLICIT context count — the ctx-free reading of {@link
+ *  hubBound}, for the callers inside this module that are handed a count
+ *  rather than a context ({@link edgeAncestors}, {@link atomIsHub}).  The
+ *  floor at 2 matches {@link corpusN}'s, so both readings agree for every
+ *  input: the two used to be spelled out inline, once WITH the floor and
+ *  once without, in the same function. */
+function boundFor(contextCount) {
+    return Math.ceil(Math.sqrt(Math.max(2, contextCount)));
+}
+/** Cap a candidate list at the hub bound √N (insertion order) — the ONE
+ *  fan-out convention every walk and disambiguation uses (see HOW_IT_WORKS
+ *  §8.6).  A node connected to more than √N others is a hub whose individual
+ *  connections carry ~no discriminative information; materialising or scoring
+ *  them all would make single decisions scale with the corpus. */
+export function hubCap(ctx, ids) {
+    const bound = hubBound(ctx);
+    return ids.length > bound ? ids.slice(0, bound) : ids;
+}
+/** Whether `descendant` lies within `ancestor`'s subtree — a structural DAG
+ *  relation read off the hash-consed `kids` lists, by a bounded explicit-stack
+ *  descent.  Used by articulation to keep a voice from revoicing a fragment
+ *  OF that voice. */
+export function contains(ctx, ancestor, descendant) {
+    if (ancestor === descendant)
+        return true;
+    const seen = new Set([ancestor]);
+    const stack = [ancestor];
+    while (stack.length > 0) {
+        const rec = ctx.store.get(stack.pop());
+        if (!rec?.kids)
+            continue;
+        for (const k of rec.kids) {
+            if (k === descendant)
+                return true;
+            if (!seen.has(k)) {
+                seen.add(k);
+                stack.push(k);
+            }
+        }
+    }
+    return false;
+}
+// ── Edge disambiguation (Section 6) ──────────────────────────────────────
+/** The best-scoring item by cosine against `query`, among items scoring at
+ *  or above `threshold` — the shared arg-max every Pattern-A "which of these
+ *  resonates best" decision reduces to.  `strict` picks the tie-break a
+ *  caller needs: `true` keeps the first-seen leader on a tie (`>`), the
+ *  default lets a later equal score take it (`>=`). */
+export function argmaxBy(items, scoreOf, threshold, strict = false) {
+    let best = null;
+    for (const item of items) {
+        const score = scoreOf(item);
+        const bar = best?.score ?? threshold;
+        if (strict ? score > bar : score >= bar)
+            best = { item, score };
+    }
+    return best;
+}
+export function argmaxCosine(query, items, vecOf, threshold, strict = false) {
+    return argmaxBy(items, (item) => {
+        const v = vecOf(item);
+        return v ? cosine(query, v) : -Infinity;
+    }, threshold, strict);
+}
+/** The guided-or-first continuation of a node, as answer-shaped bytes source:
+ *  chooseNext under the response guide, falling back to the FIRST-inserted
+ *  edge — the one no-guide convention chooseNext, project() and the search's
+ *  formRules all share.  undefined when the node has no continuation. */
+export function guidedFirst(ctx, id) {
+    const pick = guidedNext(ctx, id);
+    if (pick !== undefined)
+        return pick;
+    // No guide in flight (or nothing chosen): the first-inserted edge, read
+    // with LIMIT 1 — never the full fan-out.
+    const nx = ctx.store.nextFirst(id, 1);
+    return nx.length > 0 ? nx[0] : undefined;
+}
+export function guidedNext(ctx, node) {
+    if (ctx._edgeGuide === null)
+        return undefined;
+    // The pick memo is BYPASSED while a rationale trace is attached — the same
+    // policy climbMemo and recogniseMemo follow (every mechanism must emit its
+    // own steps; a memo hit would swallow the repeat's `disambiguate` step).
+    // Consistency does not need the memo: chooseNext is a pure function of the
+    // (read-only) store and the guide, so recomputation yields the same pick.
+    if (!ctx.trace) {
+        const memo = ctx._edgeChoice.get(node);
+        if (memo !== undefined)
+            return memo === -1 ? undefined : memo;
+    }
+    const pick = chooseNext(ctx, node, ctx._edgeGuide);
+    if (!ctx.trace)
+        ctx._edgeChoice.set(node, pick ?? -1);
+    return pick;
+}
+/** Disambiguate among a node's learnt continuations by distributional
+ *  support.  NOTE the `guide` contract: its VALUE is deliberately unused —
+ *  only its PRESENCE gates disambiguation (a null guide means no query is in
+ *  flight, so structural walkers keep plain first-edge behaviour).  The
+ *  gist-cosine of short answer candidates against a query guide is dominated
+ *  by accidental byte-pattern correlations, not semantic relatedness, so the
+ *  evidence consulted is structural: each candidate's reverse-edge support
+ *  count (see below).  Contrast {@link chooseAmong}, the REVERSE-direction
+ *  disambiguator, whose candidates are whole learnt contexts — long enough
+ *  that their perceived gists ARE semantically meaningful — and which
+ *  therefore scores by guide cosine.  The two directions consult different
+ *  halves of the evidence on purpose. */
+export function chooseNext(ctx, id, guide) {
+    // CAPPED read: only the first √N continuations are ever candidates (the
+    // documented hub trade), so only they are read — a hub context's full
+    // fan-out is corpus-sized and must never be materialised.  hubBound ≥ 2,
+    // so the single-continuation fast path below stays exact.
+    const nx = ctx.store.nextFirst(id, hubBound(ctx));
+    if (nx.length === 0)
+        return undefined;
+    if (nx.length === 1 || !guide)
+        return nx[0];
+    // Cap candidates at √N — the same bound the original chooseAmong used.
+    // A hub context can accumulate thousands of continuations; the best-fit
+    // one is among the first √N by insertion order (edges are never deleted,
+    // so the oldest are the most established).  A strongly-supported edge
+    // inserted beyond the cap is invisible here — the deliberate trade
+    // against paying O(fan-out) count reads on every disambiguation.
+    const capped = nx; // already the hub-capped prefix, by the read above
+    // Distributional-evidence disambiguation, consulting BOTH read-outs of the
+    // evidence the training poured:
+    //   1. prevCount — how many DISTINCT contexts predict this candidate (one
+    //      indexed COUNT; never a materialisation — a common continuation's
+    //      reverse fan-in is corpus-sized).  Diversity of independent evidence
+    //      is the primary signal: three different formulations agreeing beat
+    //      one formulation repeated.
+    //   2. haloMass — how many episode signatures were poured into the
+    //      candidate's halo (repetition counts).  The tie-break among equally
+    //      diverse candidates: a fact reinforced across many episodes is more
+    //      corroborated than one seen once, and this is the DIRECT measure of
+    //      that — consulting only the structural count would leave poured
+    //      evidence on the table.
+    // When both are equal, first-inserted wins (backward compatible).
+    let best = capped[0];
+    let bestSupport = ctx.store.prevCount(best);
+    let bestMass = ctx.store.haloMass(best);
+    for (let i = 1; i < capped.length; i++) {
+        const support = ctx.store.prevCount(capped[i]);
+        if (support < bestSupport)
+            continue;
+        const mass = ctx.store.haloMass(capped[i]);
+        if (support > bestSupport || mass > bestMass) {
+            best = capped[i];
+            bestSupport = support;
+            bestMass = mass;
+        }
+    }
+    // NO consensusFloor gate here (tried and reverted — see
+    // test/40-choosenext-scale-guard.test.mjs): that floor is calibrated for
+    // POOLED, IDF-weighted CLIMB VOTES (recallByResonance, commitVotes), where
+    // each corroborating region contributes at most ln N and the floor grows
+    // with N exactly as that per-region ceiling does (HOW_IT_WORKS.md §8.6).
+    // `bestSupport` here is a different kind of quantity — a raw prevCount of
+    // how many training contexts predicted ONE destination, bounded by how
+    // often that specific fact was retold, never by corpus size N.  Gating an
+    // N-invariant count against an N-growing threshold guarantees failure
+    // once N is large enough, discarding genuinely, structurally dominant
+    // edges (observed: a fact corroborated 2-to-1-1-1 refused at N≈325K,
+    // falling back to a noisy concept-hop).  The loop above already IS the
+    // "genuinely competing" test: a tie leaves first-inserted as the pick
+    // (test/30's own pinned behaviour); a strict winner is real evidence
+    // regardless of corpus scale.  Matches HOW_IT_WORKS.md §25's own
+    // chooseNext pseudocode, which has no such floor.
+    // Trace is built lazily — the filter + map below only execute when a
+    // trace listener is attached, so the common (no-trace) path pays only
+    // for the prevCount calls in the loop above, never for extra rItemShort
+    // byte-reads.
+    if (ctx.trace) {
+        const others = capped.filter((c) => c !== best);
+        ctx.trace.step("disambiguate", [rItemShort(ctx, best, "halo-evidence", bestSupport)], others.map((c) => rItemShort(ctx, c, "candidate", ctx.store.prevCount(c))), `${capped.length} continuations — distributional evidence selects ` +
+            `the most corroborated (distinct contexts ${bestSupport}, ` +
+            `poured mass ${bestMass})`);
+    }
+    return best;
+}
+/** The perceived gist of a candidate node, through the session gist cache.
+ *  Re-gisting a candidate is a full river fold of its bytes — the measured
+ *  recall bottleneck (a hub context offers up to √N continuations, EACH
+ *  re-perceived per pick).  A node's bytes are immutable and perception is
+ *  pure, so the cached gist is valid for the store's lifetime.  Exported for
+ *  every "score node ids against a guide" decision (chooseAmong here, the
+ *  bridge's junction pick) so they share ONE cache and one convention. */
+export function candidateGist(ctx, c) {
+    const hit = ctx._gistCache.get(c);
+    if (hit !== undefined)
+        return hit;
+    const b = read(ctx, c);
+    if (b.length === 0)
+        return null;
+    const g = gistOf(ctx, b);
+    ctx._gistCache.set(c, g);
+    return g;
+}
+export function chooseAmong(ctx, candidates, guide) {
+    const capped = hubCap(ctx, candidates);
+    const found = argmaxCosine(guide, capped, (c) => candidateGist(ctx, c), -Infinity, true);
+    return found
+        ? { id: found.item, score: found.score }
+        : { id: candidates[0], score: -Infinity };
+}
+// ── Trace shim (used by chooseNext before trace module is loaded) ────────
+import { decodeText } from "./rationale.js";
+function rItemShort(ctx, id, role, score) {
+    return {
+        text: decodeText(read(ctx, id)),
+        node: id,
+        role,
+        score,
+    };
+}