@inerrata-corporation/errata 2.0.22-dev.4638 → 2.0.22-dev.4652

package/consolidate-worker.mjs

@@ -17596,8 +17596,10 @@ function column(c) {
 function compileNodeWhere(q) {
   if (q.labels.length === 0) throw new Error("selectNodes: no labels");
   const params = [...q.labels];
-  const conds = [`label IN (${q.labels.map(() => "?").join(", ")})`];
-  if (!q.includeClosed) conds.push("valid_to IS NULL");
+  const byId = (q.where ?? []).some((w) => "idIn" in w);
+  const unindexed = byId ? "+" : "";
+  const conds = [`${unindexed}label IN (${q.labels.map(() => "?").join(", ")})`];
+  if (!q.includeClosed) conds.push(`${unindexed}valid_to IS NULL`);
   for (const w of q.where ?? []) {
     if ("embedding" in w) {
       conds.push(`${EMBEDDING_DIM_SQL} ${w.embedding === "present" ? ">" : "="} 0`);
@@ -17649,6 +17651,14 @@ function compileNodeWhere(q) {
 function attrValueSql(p) {
   return `CASE json_type(attrs_json, ${p}) WHEN 'integer' THEN CAST(json_extract(attrs_json, ${p}) AS REAL) ELSE json_extract(attrs_json, ${p}) END`;
 }
+function nodeSelectFrom(q) {
+  if (q.indexedBy) return `nodes INDEXED BY ${selectIndex(q.indexedBy)}`;
+  const where = q.where ?? [];
+  if (!q.includeClosed && where.some((w) => "descriptionIn" in w && w.descriptionIn.length > 0) && !where.some((w) => "idIn" in w)) {
+    return "nodes INDEXED BY nodes_live_description";
+  }
+  return "nodes";
+}
 function compileNodeSelect(q) {
   const { where, params } = compileNodeWhere(q);
   const attrs = q.attrs ?? [];
@@ -17666,7 +17676,7 @@ function compileNodeSelect(q) {
   }
   const orderParams = !q.orderBy && q.labels.length > 1 ? [...q.labels] : [];
   const limit = q.limit !== void 0 ? ` LIMIT ${Math.max(0, Math.floor(q.limit))}` : "";
-  const sql = `SELECT id, label, description, created_at, last_updated_at, page_rank, cumulative_hits, extraction_confidence, extraction_source, community, created_at_seq, last_reinforced_at_seq, valid_to, logical_id, version, ${EMBEDDING_DIM_SQL} AS embedding_dim${q.withEmbedding ? ", embedding_blob, embedding_json" : ""}${attrCols} FROM nodes${q.indexedBy ? ` INDEXED BY ${selectIndex(q.indexedBy)}` : ""} WHERE ${where} ORDER BY ${order}${limit}`;
+  const sql = `SELECT id, label, description, created_at, last_updated_at, page_rank, cumulative_hits, extraction_confidence, extraction_source, community, created_at_seq, last_reinforced_at_seq, valid_to, logical_id, version, ${EMBEDDING_DIM_SQL} AS embedding_dim${q.withEmbedding ? ", embedding_blob, embedding_json" : ""}${attrCols} FROM ${nodeSelectFrom(q)} WHERE ${where} ORDER BY ${order}${limit}`;
   return { sql, params: [...params, ...orderParams] };
 }
 function decodeAttr(value, type) {
@@ -17903,6 +17913,18 @@ CREATE INDEX IF NOT EXISTS edges_live_prior_tag ON edges(created_at) WHERE valid
 -- incremental reindex can fetch just the changed files' symbols instead of
 -- scanning every versioned node. Partial: only live rows are ever queried.
 CREATE INDEX IF NOT EXISTS nodes_relpath ON nodes(json_extract(attrs_json, '$.relPath')) WHERE valid_to IS NULL;
+-- Symbols by qname and doc comments by the symbol they attach to: the drain's
+-- symbol lexicon resolves the names in the text it ships one at a time, instead
+-- of reading every Function/Method/Class/Interface/Const and Comment row on each
+-- live push (~300 MB of a 1.1 GB store per push, 2026-10-20).
+CREATE INDEX IF NOT EXISTS nodes_live_qname ON nodes(json_extract(attrs_json, '$.qname')) WHERE valid_to IS NULL;
+CREATE INDEX IF NOT EXISTS nodes_live_attached_to ON nodes(json_extract(attrs_json, '$.attachedTo')) WHERE valid_to IS NULL AND label = 'Comment';
+-- The attrs the cloud drain gates every instance row on, so it reads them from
+-- the index alone: pulled cloud rows are most of these labels (7,105 of 7,708
+-- Solutions on the live store) at ~10 KB of attrs each, and five full scans for
+-- two attrs read ~220 MB per push to find the few local rows that can ship.
+-- created_at leads so the index order is findNodeFieldsByLabel's.
+CREATE INDEX IF NOT EXISTS nodes_live_instance_gate ON nodes(label, created_at, json_extract(attrs_json, '$.source'), json_extract(attrs_json, '$.privateFinding'), json_extract(attrs_json, '$.resolvedAs'), id) WHERE valid_to IS NULL AND label IN ('Problem', 'Vulnerability', 'Solution', 'RootCause');
 -- Closed rows by relPath, for reviveReobserved (every incremental reindex): the
 -- live-only index above cannot serve it, so it scanned the whole table (~275 MB
 -- of a 1.1 GB store per reindex, 2026-10-20).
@@ -18253,6 +18275,52 @@ var SqliteGraphStore = class {
         AND (t.id IS NULL OR t.valid_to IS NOT NULL)
     `).all();
   }
+  strandedSymbolsForFiles(fileIds) {
+    const out = [];
+    for (let i = 0; i < fileIds.length; i += 500) {
+      const slice = fileIds.slice(i, i + 500);
+      if (slice.length === 0) continue;
+      out.push(...this.db.prepare(`
+        SELECT f.id AS fileId, e.to_id AS symbolId
+        FROM nodes f CROSS JOIN edges e ON e.from_id = f.id
+        LEFT JOIN nodes t ON t.id = e.to_id
+        WHERE f.id IN (${slice.map(() => "?").join(",")}) AND f.label = 'File' AND f.valid_to IS NULL AND e.valid_to IS NULL
+          AND e.type IN ('DEFINES', 'CONTAINS')
+          AND (t.id IS NULL OR t.valid_to IS NOT NULL)
+      `).all(...slice));
+    }
+    return out;
+  }
+  firstLiveByNameOrQname(labels, name) {
+    if (labels.length === 0) return null;
+    const inLabels = labels.map(() => "?").join(",");
+    const probe = (index, match) => this.db.prepare(
+      `SELECT rowid AS rid, created_at AS at, label, description, json_extract(attrs_json, '$.qname') AS qname FROM nodes INDEXED BY ${index}
+          WHERE ${match} = ? AND label IN (${inLabels}) AND valid_to IS NULL`
+    ).all(name, ...labels);
+    const rank = (r) => [labels.indexOf(r.label), r.at, r.rid];
+    const before = (a, b) => a[0] < b[0] || a[0] === b[0] && (a[1] < b[1] || a[1] === b[1] && a[2] < b[2]);
+    let best;
+    for (const r of [...probe("nodes_live_description", "description"), ...probe("nodes_live_qname", "json_extract(attrs_json, '$.qname')")]) {
+      if (!best || before(rank(r), rank(best))) best = r;
+    }
+    return best ? { label: best.label, description: best.description, qname: typeof best.qname === "string" ? best.qname : null } : null;
+  }
+  instanceGateRows(label) {
+    return this.db.prepare(
+      `SELECT id, json_extract(attrs_json, '$.source') AS source, json_extract(attrs_json, '$.privateFinding') AS privateFinding,
+              json_extract(attrs_json, '$.resolvedAs') AS resolvedAs
+         FROM nodes INDEXED BY nodes_live_instance_gate
+        WHERE label IN ('Problem', 'Vulnerability', 'Solution', 'RootCause') AND label = ? AND valid_to IS NULL
+        ORDER BY label, created_at, rowid`
+    ).all(label);
+  }
+  liveCommentsAttachedTo(name) {
+    return this.db.prepare(
+      `SELECT json_extract(attrs_json, '$.kind') AS kind, json_extract(attrs_json, '$.text') AS text, description FROM nodes INDEXED BY nodes_live_attached_to
+        WHERE label = 'Comment' AND valid_to IS NULL AND json_extract(attrs_json, '$.attachedTo') = ? ORDER BY created_at, rowid`
+    ).all(name);
+  }
   strandedEdges() {
     return countStrandedEdges(this.db);
   }
@@ -18431,7 +18499,11 @@ var SqliteGraphStore = class {
     if (!stmt2) {
       const cols = attrKeys.map((k, i) => `, ${attrValueSql(`'$.${k}'`)} AS a${i}, json_type(attrs_json, '$.${k}') AS t${i}`).join("");
       stmt2 = this.db.prepare(
-        `SELECT id, description, community, extraction_confidence${cols} FROM nodes WHERE label = ? AND valid_to IS NULL`
+        // Ordered, not left to the planner: readers take "first row wins" (the
+        // symbol lexicon's name winner and doc pick), and the order otherwise
+        // followed whichever partial index the planner chose. nodes_live_created
+        // serves this order without a sort.
+        `SELECT id, description, community, extraction_confidence${cols} FROM nodes WHERE label = ? AND valid_to IS NULL ORDER BY created_at, rowid`
       );
       this.fieldStmts.set(cacheKey, stmt2);
     }
@@ -18457,7 +18529,7 @@ var SqliteGraphStore = class {
   }
   countNodes(q) {
     const { where, params } = compileNodeWhere(q);
-    return Number(this.db.prepare(`SELECT COUNT(*) AS n FROM nodes WHERE ${where}`).get(...params).n);
+    return Number(this.db.prepare(`SELECT COUNT(*) AS n FROM ${nodeSelectFrom(q)} WHERE ${where}`).get(...params).n);
   }
   setCommunity(id, community) {
     this.stmts.setCommunity.run({ id, v: community });
@@ -20877,6 +20949,62 @@ function buildSymbolLexicon(store) {
   }
   return lex;
 }
+function lazySymbolLexicon(store) {
+  if (!store.firstLiveByNameOrQname || !store.liveCommentsAttachedTo) return null;
+  const labels = SYMBOL_LABELS.map(([label]) => label);
+  const kindOf = new Map(SYMBOL_LABELS.map(([label, kind]) => [label, kind]));
+  const members = /* @__PURE__ */ new Map();
+  const docs = /* @__PURE__ */ new Map();
+  const entries = /* @__PURE__ */ new Map();
+  const member = (name) => {
+    if (!members.has(name)) {
+      const hit = isDistinctiveIdentifier(name) ? store.firstLiveByNameOrQname(labels, name) : null;
+      members.set(name, hit ? { kind: kindOf.get(hit.label), description: hit.description, qname: hit.qname } : null);
+    }
+    return members.get(name);
+  };
+  const docFor = (name) => {
+    if (!docs.has(name)) {
+      let found;
+      for (const c of store.liveCommentsAttachedTo(name)) {
+        if (typeof c.kind !== "string" || !DOC_COMMENT_KINDS.has(c.kind)) continue;
+        const text = typeof c.text === "string" ? c.text : c.description;
+        if (!text?.trim()) continue;
+        found = text.trim().slice(0, MAX_DOC_CHARS);
+        break;
+      }
+      docs.set(name, found);
+    }
+    return docs.get(name);
+  };
+  const isKnown = (tok) => member(tok) !== null;
+  let anySymbol;
+  return {
+    has: isKnown,
+    get(name) {
+      if (!entries.has(name)) {
+        const m = member(name);
+        let entry = null;
+        if (m) {
+          entry = { kind: m.kind };
+          const doc = (m.description ? docFor(m.description) : void 0) ?? (m.qname ? docFor(m.qname) : void 0);
+          for (const phrase of summaryCandidates(name, m.kind, doc)) {
+            if (summaryRejectionReason(phrase, isKnown) === null) {
+              entry.summary = phrase;
+              break;
+            }
+          }
+        }
+        entries.set(name, entry);
+      }
+      return entries.get(name) ?? void 0;
+    },
+    get size() {
+      anySymbol ??= labels.some((label) => store.countNodesByLabel(label) > 0);
+      return anySymbol ? 1 : 0;
+    }
+  };
+}
 var LEXICON_SYMBOL_LABELS = ["Function", "Method", "Class", "Interface", "Const", "Comment"];
 function codeGraphKey(store) {
   const files = store.labelFingerprint?.(["File"]);
@@ -20888,7 +21016,7 @@ function lexiconsFor(store) {
   const key = codeGraphKey(store);
   const hit = key !== null ? lexiconCache.get(store) : void 0;
   if (hit && hit.key === key) return hit;
-  const built = { symbols: buildSymbolLexicon(store), files: buildFileLexicon(store) };
+  const built = { symbols: lazySymbolLexicon(store) ?? buildSymbolLexicon(store), files: buildFileLexicon(store) };
   if (key !== null) lexiconCache.set(store, { key, ...built });
   return built;
 }
@@ -21127,6 +21255,38 @@ function stripCodebaseScope(scope) {
   delete s["codebase"];
   return s;
 }
+function instanceGateRows(store, label) {
+  return store.instanceGateRows?.(label) ?? store.findNodeFieldsByLabel(label, ["source", "privateFinding", "resolvedAs"]).map((r) => ({
+    id: r.id,
+    source: r.attrs["source"],
+    privateFinding: r.attrs["privateFinding"],
+    resolvedAs: r.attrs["resolvedAs"]
+  }));
+}
+var GATE_VIEW_ATTRS = [
+  "privateFinding",
+  "source",
+  "sources",
+  "resolvedAs",
+  "instanceReceipts",
+  "cloudWithdrawn",
+  "withdrawnAtCorrectionSeq",
+  "contributedAtSeq",
+  "cloudNodeId"
+];
+var GATE_VIEW_CHUNK = 500;
+var gateView = /* @__PURE__ */ new WeakSet();
+function gateViews(store, label, ids) {
+  const byId = /* @__PURE__ */ new Map();
+  for (let i = 0; i < ids.length; i += GATE_VIEW_CHUNK) {
+    for (const r of store.selectNodes({ labels: [label], attrs: GATE_VIEW_ATTRS, where: [{ idIn: ids.slice(i, i + GATE_VIEW_CHUNK) }] })) {
+      const view = { ...r, embedding: [] };
+      gateView.add(view);
+      byId.set(r.id, view);
+    }
+  }
+  return ids.flatMap((id) => byId.get(id) ?? []);
+}
 function contributedWireId(n, destinationKey) {
   const cloudId = instanceReceipt(n, destinationKey)["cloudNodeId"];
   return typeof cloudId === "string" && cloudId ? cloudId : n.id;
@@ -21285,7 +21445,7 @@ function buildInstanceIngest(store, profile, daemonVersion, ignorePatterns = [],
     edgesHeldDropped: 0
   };
   const sharedCauses = /* @__PURE__ */ new Map();
-  const causalQueue = store.findNodeFieldsByLabel("Problem", ["source", "resolvedAs"]).filter((r) => r.attrs["source"] !== "cloud" && r.attrs["resolvedAs"] !== PROBLEM_RESOLUTION.FALSE_POSITIVE).map((r) => r.id).filter((id) => ignorePatterns.length === 0 || !ignored(store.getNode(id)?.description ?? ""));
+  const causalQueue = instanceGateRows(store, "Problem").filter((r) => r.source !== "cloud" && r.resolvedAs !== PROBLEM_RESOLUTION.FALSE_POSITIVE).map((r) => r.id).filter((id) => ignorePatterns.length === 0 || !ignored(store.getNode(id)?.description ?? ""));
   const causalSeen = new Set(causalQueue);
   for (let i = 0; i < causalQueue.length; i++) {
     const id = causalQueue[i];
@@ -21301,21 +21461,21 @@ function buildInstanceIngest(store, profile, daemonVersion, ignorePatterns = [],
   for (const label of INSTANCE_LABELS) {
     const ref = [];
     includedByLabel.set(label, ref);
-    const local = [];
-    for (const row of store.findNodeFieldsByLabel(label, ["source", "privateFinding"])) {
-      if (row.attrs["source"] !== "cloud") {
-        const n = store.getNode(row.id);
-        if (n) local.push(n);
+    const localIds = [];
+    for (const row of instanceGateRows(store, label)) {
+      if (row.source !== "cloud") {
+        localIds.push(row.id);
         continue;
       }
       if (seen.has(row.id) || walked.has(row.id)) continue;
       walked.add(row.id);
       composition.candidates++;
-      if ((row.attrs["privateFinding"] || label === "Vulnerability") && !opts.project) composition.privateTargetMissing++;
+      if ((row.privateFinding || label === "Vulnerability") && !opts.project) composition.privateTargetMissing++;
       else composition.cloudSourced++;
     }
+    const local = gateViews(store, label, localIds);
     const population = label === "RootCause" && opts.sharedStore ? [...local, ...sharedCauses.values()] : local;
-    for (const n of population) {
+    for (let n of population) {
       if (seen.has(n.id)) continue;
       if (walked.has(n.id)) continue;
       walked.add(n.id);
@@ -21351,6 +21511,11 @@ function buildInstanceIngest(store, profile, daemonVersion, ignorePatterns = [],
         semanticEndpoints.set(n.id, contributedWireId(n, opts.destinationKey));
         continue;
       }
+      if (gateView.has(n)) {
+        const full = store.getNode(n.id);
+        if (!full) continue;
+        n = full;
+      }
       if (isEditPlaceholderSolution(n)) {
         composition.autoMint++;
         continue;
@@ -21754,7 +21919,7 @@ function runPass(msg) {
       return revisitContradictedPrinciples(shared, msg.ts ?? Date.now());
     case "strandedCount": {
       if (!msg.castaliaPath) throw new Error("strandedCount: castaliaPath required");
-      return findStrandedSymbols(storeAt(msg.castaliaPath)).length;
+      return findStrandedSymbols(storeAt(msg.castaliaPath));
     }
     case "drainContext": {
       if (!msg.castaliaPath) throw new Error("drainContext: castaliaPath required");

package/errata.mjs

@@ -20393,8 +20393,10 @@ function column(c) {
 function compileNodeWhere(q) {
   if (q.labels.length === 0) throw new Error("selectNodes: no labels");
   const params = [...q.labels];
-  const conds = [`label IN (${q.labels.map(() => "?").join(", ")})`];
-  if (!q.includeClosed) conds.push("valid_to IS NULL");
+  const byId = (q.where ?? []).some((w) => "idIn" in w);
+  const unindexed = byId ? "+" : "";
+  const conds = [`${unindexed}label IN (${q.labels.map(() => "?").join(", ")})`];
+  if (!q.includeClosed) conds.push(`${unindexed}valid_to IS NULL`);
   for (const w of q.where ?? []) {
     if ("embedding" in w) {
       conds.push(`${EMBEDDING_DIM_SQL} ${w.embedding === "present" ? ">" : "="} 0`);
@@ -20446,6 +20448,14 @@ function compileNodeWhere(q) {
 function attrValueSql(p) {
   return `CASE json_type(attrs_json, ${p}) WHEN 'integer' THEN CAST(json_extract(attrs_json, ${p}) AS REAL) ELSE json_extract(attrs_json, ${p}) END`;
 }
+function nodeSelectFrom(q) {
+  if (q.indexedBy) return `nodes INDEXED BY ${selectIndex(q.indexedBy)}`;
+  const where = q.where ?? [];
+  if (!q.includeClosed && where.some((w) => "descriptionIn" in w && w.descriptionIn.length > 0) && !where.some((w) => "idIn" in w)) {
+    return "nodes INDEXED BY nodes_live_description";
+  }
+  return "nodes";
+}
 function compileNodeSelect(q) {
   const { where, params } = compileNodeWhere(q);
   const attrs = q.attrs ?? [];
@@ -20463,7 +20473,7 @@ function compileNodeSelect(q) {
   }
   const orderParams = !q.orderBy && q.labels.length > 1 ? [...q.labels] : [];
   const limit = q.limit !== void 0 ? ` LIMIT ${Math.max(0, Math.floor(q.limit))}` : "";
-  const sql = `SELECT id, label, description, created_at, last_updated_at, page_rank, cumulative_hits, extraction_confidence, extraction_source, community, created_at_seq, last_reinforced_at_seq, valid_to, logical_id, version, ${EMBEDDING_DIM_SQL} AS embedding_dim${q.withEmbedding ? ", embedding_blob, embedding_json" : ""}${attrCols} FROM nodes${q.indexedBy ? ` INDEXED BY ${selectIndex(q.indexedBy)}` : ""} WHERE ${where} ORDER BY ${order}${limit}`;
+  const sql = `SELECT id, label, description, created_at, last_updated_at, page_rank, cumulative_hits, extraction_confidence, extraction_source, community, created_at_seq, last_reinforced_at_seq, valid_to, logical_id, version, ${EMBEDDING_DIM_SQL} AS embedding_dim${q.withEmbedding ? ", embedding_blob, embedding_json" : ""}${attrCols} FROM ${nodeSelectFrom(q)} WHERE ${where} ORDER BY ${order}${limit}`;
   return { sql, params: [...params, ...orderParams] };
 }
 function decodeAttr(value, type) {
@@ -20812,6 +20822,18 @@ CREATE INDEX IF NOT EXISTS edges_live_prior_tag ON edges(created_at) WHERE valid
 -- incremental reindex can fetch just the changed files' symbols instead of
 -- scanning every versioned node. Partial: only live rows are ever queried.
 CREATE INDEX IF NOT EXISTS nodes_relpath ON nodes(json_extract(attrs_json, '$.relPath')) WHERE valid_to IS NULL;
+-- Symbols by qname and doc comments by the symbol they attach to: the drain's
+-- symbol lexicon resolves the names in the text it ships one at a time, instead
+-- of reading every Function/Method/Class/Interface/Const and Comment row on each
+-- live push (~300 MB of a 1.1 GB store per push, 2026-10-20).
+CREATE INDEX IF NOT EXISTS nodes_live_qname ON nodes(json_extract(attrs_json, '$.qname')) WHERE valid_to IS NULL;
+CREATE INDEX IF NOT EXISTS nodes_live_attached_to ON nodes(json_extract(attrs_json, '$.attachedTo')) WHERE valid_to IS NULL AND label = 'Comment';
+-- The attrs the cloud drain gates every instance row on, so it reads them from
+-- the index alone: pulled cloud rows are most of these labels (7,105 of 7,708
+-- Solutions on the live store) at ~10 KB of attrs each, and five full scans for
+-- two attrs read ~220 MB per push to find the few local rows that can ship.
+-- created_at leads so the index order is findNodeFieldsByLabel's.
+CREATE INDEX IF NOT EXISTS nodes_live_instance_gate ON nodes(label, created_at, json_extract(attrs_json, '$.source'), json_extract(attrs_json, '$.privateFinding'), json_extract(attrs_json, '$.resolvedAs'), id) WHERE valid_to IS NULL AND label IN ('Problem', 'Vulnerability', 'Solution', 'RootCause');
 -- Closed rows by relPath, for reviveReobserved (every incremental reindex): the
 -- live-only index above cannot serve it, so it scanned the whole table (~275 MB
 -- of a 1.1 GB store per reindex, 2026-10-20).
@@ -21105,6 +21127,52 @@ CREATE INDEX IF NOT EXISTS nodes_revisit_true ON nodes(valid_from) WHERE json_ty
         AND (t.id IS NULL OR t.valid_to IS NOT NULL)
     `).all();
       }
+      strandedSymbolsForFiles(fileIds) {
+        const out2 = [];
+        for (let i2 = 0; i2 < fileIds.length; i2 += 500) {
+          const slice = fileIds.slice(i2, i2 + 500);
+          if (slice.length === 0) continue;
+          out2.push(...this.db.prepare(`
+        SELECT f.id AS fileId, e.to_id AS symbolId
+        FROM nodes f CROSS JOIN edges e ON e.from_id = f.id
+        LEFT JOIN nodes t ON t.id = e.to_id
+        WHERE f.id IN (${slice.map(() => "?").join(",")}) AND f.label = 'File' AND f.valid_to IS NULL AND e.valid_to IS NULL
+          AND e.type IN ('DEFINES', 'CONTAINS')
+          AND (t.id IS NULL OR t.valid_to IS NOT NULL)
+      `).all(...slice));
+        }
+        return out2;
+      }
+      firstLiveByNameOrQname(labels, name2) {
+        if (labels.length === 0) return null;
+        const inLabels = labels.map(() => "?").join(",");
+        const probe = (index, match2) => this.db.prepare(
+          `SELECT rowid AS rid, created_at AS at, label, description, json_extract(attrs_json, '$.qname') AS qname FROM nodes INDEXED BY ${index}
+          WHERE ${match2} = ? AND label IN (${inLabels}) AND valid_to IS NULL`
+        ).all(name2, ...labels);
+        const rank = (r) => [labels.indexOf(r.label), r.at, r.rid];
+        const before = (a, b) => a[0] < b[0] || a[0] === b[0] && (a[1] < b[1] || a[1] === b[1] && a[2] < b[2]);
+        let best;
+        for (const r of [...probe("nodes_live_description", "description"), ...probe("nodes_live_qname", "json_extract(attrs_json, '$.qname')")]) {
+          if (!best || before(rank(r), rank(best))) best = r;
+        }
+        return best ? { label: best.label, description: best.description, qname: typeof best.qname === "string" ? best.qname : null } : null;
+      }
+      instanceGateRows(label) {
+        return this.db.prepare(
+          `SELECT id, json_extract(attrs_json, '$.source') AS source, json_extract(attrs_json, '$.privateFinding') AS privateFinding,
+              json_extract(attrs_json, '$.resolvedAs') AS resolvedAs
+         FROM nodes INDEXED BY nodes_live_instance_gate
+        WHERE label IN ('Problem', 'Vulnerability', 'Solution', 'RootCause') AND label = ? AND valid_to IS NULL
+        ORDER BY label, created_at, rowid`
+        ).all(label);
+      }
+      liveCommentsAttachedTo(name2) {
+        return this.db.prepare(
+          `SELECT json_extract(attrs_json, '$.kind') AS kind, json_extract(attrs_json, '$.text') AS text, description FROM nodes INDEXED BY nodes_live_attached_to
+        WHERE label = 'Comment' AND valid_to IS NULL AND json_extract(attrs_json, '$.attachedTo') = ? ORDER BY created_at, rowid`
+        ).all(name2);
+      }
       strandedEdges() {
         return countStrandedEdges(this.db);
       }
@@ -21283,7 +21351,11 @@ CREATE INDEX IF NOT EXISTS nodes_revisit_true ON nodes(valid_from) WHERE json_ty
         if (!stmt2) {
           const cols = attrKeys.map((k, i2) => `, ${attrValueSql(`'$.${k}'`)} AS a${i2}, json_type(attrs_json, '$.${k}') AS t${i2}`).join("");
           stmt2 = this.db.prepare(
-            `SELECT id, description, community, extraction_confidence${cols} FROM nodes WHERE label = ? AND valid_to IS NULL`
+            // Ordered, not left to the planner: readers take "first row wins" (the
+            // symbol lexicon's name winner and doc pick), and the order otherwise
+            // followed whichever partial index the planner chose. nodes_live_created
+            // serves this order without a sort.
+            `SELECT id, description, community, extraction_confidence${cols} FROM nodes WHERE label = ? AND valid_to IS NULL ORDER BY created_at, rowid`
           );
           this.fieldStmts.set(cacheKey2, stmt2);
         }
@@ -21309,7 +21381,7 @@ CREATE INDEX IF NOT EXISTS nodes_revisit_true ON nodes(valid_from) WHERE json_ty
       }
       countNodes(q) {
         const { where, params } = compileNodeWhere(q);
-        return Number(this.db.prepare(`SELECT COUNT(*) AS n FROM nodes WHERE ${where}`).get(...params).n);
+        return Number(this.db.prepare(`SELECT COUNT(*) AS n FROM ${nodeSelectFrom(q)} WHERE ${where}`).get(...params).n);
       }
       setCommunity(id, community) {
         this.stmts.setCommunity.run({ id, v: community });
@@ -37286,6 +37358,62 @@ function buildSymbolLexicon(store) {
   }
   return lex;
 }
+function lazySymbolLexicon(store) {
+  if (!store.firstLiveByNameOrQname || !store.liveCommentsAttachedTo) return null;
+  const labels = SYMBOL_LABELS2.map(([label]) => label);
+  const kindOf = new Map(SYMBOL_LABELS2.map(([label, kind]) => [label, kind]));
+  const members = /* @__PURE__ */ new Map();
+  const docs = /* @__PURE__ */ new Map();
+  const entries = /* @__PURE__ */ new Map();
+  const member = (name2) => {
+    if (!members.has(name2)) {
+      const hit = isDistinctiveIdentifier(name2) ? store.firstLiveByNameOrQname(labels, name2) : null;
+      members.set(name2, hit ? { kind: kindOf.get(hit.label), description: hit.description, qname: hit.qname } : null);
+    }
+    return members.get(name2);
+  };
+  const docFor = (name2) => {
+    if (!docs.has(name2)) {
+      let found;
+      for (const c of store.liveCommentsAttachedTo(name2)) {
+        if (typeof c.kind !== "string" || !DOC_COMMENT_KINDS.has(c.kind)) continue;
+        const text = typeof c.text === "string" ? c.text : c.description;
+        if (!text?.trim()) continue;
+        found = text.trim().slice(0, MAX_DOC_CHARS);
+        break;
+      }
+      docs.set(name2, found);
+    }
+    return docs.get(name2);
+  };
+  const isKnown = (tok) => member(tok) !== null;
+  let anySymbol;
+  return {
+    has: isKnown,
+    get(name2) {
+      if (!entries.has(name2)) {
+        const m = member(name2);
+        let entry = null;
+        if (m) {
+          entry = { kind: m.kind };
+          const doc = (m.description ? docFor(m.description) : void 0) ?? (m.qname ? docFor(m.qname) : void 0);
+          for (const phrase2 of summaryCandidates(name2, m.kind, doc)) {
+            if (summaryRejectionReason(phrase2, isKnown) === null) {
+              entry.summary = phrase2;
+              break;
+            }
+          }
+        }
+        entries.set(name2, entry);
+      }
+      return entries.get(name2) ?? void 0;
+    },
+    get size() {
+      anySymbol ??= labels.some((label) => store.countNodesByLabel(label) > 0);
+      return anySymbol ? 1 : 0;
+    }
+  };
+}
 function codeGraphKey(store) {
   const files = store.labelFingerprint?.(["File"]);
   if (files === void 0) return null;
@@ -37295,7 +37423,7 @@ function lexiconsFor(store) {
   const key = codeGraphKey(store);
   const hit = key !== null ? lexiconCache.get(store) : void 0;
   if (hit && hit.key === key) return hit;
-  const built = { symbols: buildSymbolLexicon(store), files: buildFileLexicon(store) };
+  const built = { symbols: lazySymbolLexicon(store) ?? buildSymbolLexicon(store), files: buildFileLexicon(store) };
   if (key !== null) lexiconCache.set(store, { key, ...built });
   return built;
 }
@@ -47122,14 +47250,31 @@ function standingStrandedCount(store) {
   try {
     const last = JSON.parse(store.getMeta?.(STRANDED_KEY) ?? "null");
     if (!last) return null;
-    if ((store.changeSeq?.() ?? Infinity) - last.seq > STRANDED_WINDOW_MAX) return null;
-    return store.lineageChangedFiles?.(last.seq)?.length === 0 ? last.count : null;
+    const seq = store.changeSeq?.();
+    if ((seq ?? Infinity) - last.seq > STRANDED_WINDOW_MAX) return null;
+    const changed = store.lineageChangedFiles?.(last.seq);
+    if (!changed) return null;
+    if (changed.length === 0) return last.count;
+    if (!last.byFile || !store.strandedSymbolsForFiles || seq === void 0) return null;
+    const byFile = { ...last.byFile };
+    for (const f of changed) delete byFile[f];
+    for (const s of store.strandedSymbolsForFiles(changed)) byFile[s.fileId] = (byFile[s.fileId] ?? 0) + 1;
+    const count = Object.values(byFile).reduce((a, b) => a + b, 0);
+    store.setMeta?.(STRANDED_KEY, JSON.stringify({ seq, count, byFile }));
+    return count;
   } catch {
     return null;
   }
 }
-function recordStrandedCount(store, seq, count) {
-  if (seq !== void 0) store.setMeta?.(STRANDED_KEY, JSON.stringify({ seq, count }));
+function recordStrandedCount(store, seq, stranded) {
+  if (seq === void 0) return;
+  if (typeof stranded === "number") {
+    store.setMeta?.(STRANDED_KEY, JSON.stringify({ seq, count: stranded }));
+    return;
+  }
+  const byFile = {};
+  for (const s of stranded) byFile[s.fileId] = (byFile[s.fileId] ?? 0) + 1;
+  store.setMeta?.(STRANDED_KEY, JSON.stringify({ seq, count: stranded.length, byFile }));
 }
 function reapOrphanedFiles(store, rootPath, workspaceId2, ignoreDirs = []) {
   const ignores = /* @__PURE__ */ new Set([...DEFAULT_IGNORES, ...ignoreDirs]);
@@ -80199,7 +80344,7 @@ import { readFileSync as readFileSync18, renameSync as renameSync3, statSync as
 import { fileURLToPath as fileURLToPath2 } from "node:url";
 import { dirname as dirname11, join as join22 } from "node:path";
 function snapshotReusable(held, want) {
-  return held !== null && held.key === want.key && held.atCommit === want.atCommit && held.workingFile === want.workingFile;
+  return held !== null && held.key === want.key && held.atCommit === want.atCommit;
 }
 var codeIdentityMemo;
 function snapshotCodeIdentity(version3) {
@@ -86861,7 +87006,7 @@ function createWatchBreaker(deps) {
 }
 
 // src/engine.ts
-var DAEMON_VERSION = true ? "2.0.22-dev.4638" : "2.0.0-alpha.0";
+var DAEMON_VERSION = true ? "2.0.22-dev.4652" : "2.0.0-alpha.0";
 var IGNORED_PATH = /[\\/](?:\.git|node_modules|\.errata|\.claude|\.codex|\.cursor|\.turbo|\.next|dist|coverage|test-results|playwright-report|__pycache__)(?:[\\/]|$)/;
 var IGNORED_NOISE = /castalia\.db|eventlog\.sqlite|turn-cursors/;
 var GIT_OP_MUTE_MS = 4e3;
@@ -87374,10 +87519,18 @@ function createWorkspaceEngine(opts) {
     };
     const changes = store.changeSeq ? store.changeSeq() - store.changeSeq([EXPOSURE_SCOPE]) : store.changeStamp();
     const { id, name: name2, stack, languages, domains, goals } = snapOpts.profile;
-    const snapshotKey = `${snapshotCodeIdentity(DAEMON_VERSION)}|${changes}|${JSON.stringify({ ...snapOpts, profile: { id, name: name2, stack, languages, domains, goals } })}`;
+    const { workingFile: _workingFile, ...keyedOpts } = snapOpts;
+    const snapshotKey = `${snapshotCodeIdentity(DAEMON_VERSION)}|${changes}|${JSON.stringify({ ...keyedOpts, profile: { id, name: name2, stack, languages, domains, goals } })}`;
     lastWorkerSnapshot ??= loadHeldSnapshot(heldSnapshotPath(paths.configDir));
     const held = lastWorkerSnapshot;
-    const prebuilt = snapshotReusable(held, { key: snapshotKey, atCommit: snapOpts.atCommit, workingFile: snapOpts.workingFile }) ? structuredClone(held.snapshot) : passWorker ? await passWorker.snapshot(snapOpts).then((snapshot2) => {
+    const withWorkingFile = (snap, builtFor, want, atCommit) => {
+      if (builtFor === want) return snap;
+      const slice = want ? sliceForFile(store, want, atCommit) : void 0;
+      if (slice) snap.workingFile = slice;
+      else delete snap.workingFile;
+      return snap;
+    };
+    const prebuilt = snapshotReusable(held, { key: snapshotKey, atCommit: snapOpts.atCommit }) ? withWorkingFile(structuredClone(held.snapshot), held.workingFile, snapOpts.workingFile, snapOpts.atCommit) : passWorker ? await passWorker.snapshot(snapOpts).then((snapshot2) => {
       lastWorkerSnapshot = { key: snapshotKey, at: Date.now(), atCommit: snapOpts.atCommit, workingFile: snapOpts.workingFile, snapshot: structuredClone(snapshot2) };
       saveHeldSnapshot(heldSnapshotPath(paths.configDir), lastWorkerSnapshot);
       return snapshot2;
@@ -90691,6 +90844,38 @@ function compositionCounters(c) {
     ...Object.fromEntries(buckets.map((b) => [`sidecarDrop_${b}`, sidecarDropReasons[b] ?? 0]))
   };
 }
+function instanceGateRows(store, label) {
+  return store.instanceGateRows?.(label) ?? store.findNodeFieldsByLabel(label, ["source", "privateFinding", "resolvedAs"]).map((r) => ({
+    id: r.id,
+    source: r.attrs["source"],
+    privateFinding: r.attrs["privateFinding"],
+    resolvedAs: r.attrs["resolvedAs"]
+  }));
+}
+var GATE_VIEW_ATTRS = [
+  "privateFinding",
+  "source",
+  "sources",
+  "resolvedAs",
+  "instanceReceipts",
+  "cloudWithdrawn",
+  "withdrawnAtCorrectionSeq",
+  "contributedAtSeq",
+  "cloudNodeId"
+];
+var GATE_VIEW_CHUNK = 500;
+var gateView = /* @__PURE__ */ new WeakSet();
+function gateViews(store, label, ids) {
+  const byId = /* @__PURE__ */ new Map();
+  for (let i2 = 0; i2 < ids.length; i2 += GATE_VIEW_CHUNK) {
+    for (const r of store.selectNodes({ labels: [label], attrs: GATE_VIEW_ATTRS, where: [{ idIn: ids.slice(i2, i2 + GATE_VIEW_CHUNK) }] })) {
+      const view = { ...r, embedding: [] };
+      gateView.add(view);
+      byId.set(r.id, view);
+    }
+  }
+  return ids.flatMap((id) => byId.get(id) ?? []);
+}
 function contributedWireId(n, destinationKey) {
   const cloudId = instanceReceipt(n, destinationKey)["cloudNodeId"];
   return typeof cloudId === "string" && cloudId ? cloudId : n.id;
@@ -90849,7 +91034,7 @@ function buildInstanceIngest(store, profile, daemonVersion, ignorePatterns = [],
     edgesHeldDropped: 0
   };
   const sharedCauses = /* @__PURE__ */ new Map();
-  const causalQueue = store.findNodeFieldsByLabel("Problem", ["source", "resolvedAs"]).filter((r) => r.attrs["source"] !== "cloud" && r.attrs["resolvedAs"] !== PROBLEM_RESOLUTION.FALSE_POSITIVE).map((r) => r.id).filter((id) => ignorePatterns.length === 0 || !ignored(store.getNode(id)?.description ?? ""));
+  const causalQueue = instanceGateRows(store, "Problem").filter((r) => r.source !== "cloud" && r.resolvedAs !== PROBLEM_RESOLUTION.FALSE_POSITIVE).map((r) => r.id).filter((id) => ignorePatterns.length === 0 || !ignored(store.getNode(id)?.description ?? ""));
   const causalSeen = new Set(causalQueue);
   for (let i2 = 0; i2 < causalQueue.length; i2++) {
     const id = causalQueue[i2];
@@ -90865,21 +91050,21 @@ function buildInstanceIngest(store, profile, daemonVersion, ignorePatterns = [],
   for (const label of INSTANCE_LABELS) {
     const ref = [];
     includedByLabel.set(label, ref);
-    const local = [];
-    for (const row of store.findNodeFieldsByLabel(label, ["source", "privateFinding"])) {
-      if (row.attrs["source"] !== "cloud") {
-        const n = store.getNode(row.id);
-        if (n) local.push(n);
+    const localIds = [];
+    for (const row of instanceGateRows(store, label)) {
+      if (row.source !== "cloud") {
+        localIds.push(row.id);
         continue;
       }
       if (seen.has(row.id) || walked.has(row.id)) continue;
       walked.add(row.id);
       composition.candidates++;
-      if ((row.attrs["privateFinding"] || label === "Vulnerability") && !opts.project) composition.privateTargetMissing++;
+      if ((row.privateFinding || label === "Vulnerability") && !opts.project) composition.privateTargetMissing++;
       else composition.cloudSourced++;
     }
+    const local = gateViews(store, label, localIds);
     const population = label === "RootCause" && opts.sharedStore ? [...local, ...sharedCauses.values()] : local;
-    for (const n of population) {
+    for (let n of population) {
       if (seen.has(n.id)) continue;
       if (walked.has(n.id)) continue;
       walked.add(n.id);
@@ -90915,6 +91100,11 @@ function buildInstanceIngest(store, profile, daemonVersion, ignorePatterns = [],
         semanticEndpoints.set(n.id, contributedWireId(n, opts.destinationKey));
         continue;
       }
+      if (gateView.has(n)) {
+        const full = store.getNode(n.id);
+        if (!full) continue;
+        n = full;
+      }
       if (isEditPlaceholderSolution(n)) {
         composition.autoMint++;
         continue;
@@ -95314,17 +95504,17 @@ async function startMultiDaemon(opts = {}) {
     if (!strandedRefreshing.has(r.id)) {
       strandedRefreshing.add(r.id);
       const seq = store.changeSeq?.();
-      const finish = (count) => {
-        strandedCache.set(r.id, { at: Date.now(), count });
+      const finish = (stranded) => {
+        strandedCache.set(r.id, { at: Date.now(), count: stranded.length });
         strandedRefreshing.delete(r.id);
-        recordStrandedCount(store, seq, count);
+        recordStrandedCount(store, seq, stranded);
       };
       if (consolidateWorker && r.engine.paths.castalia !== ":memory:") {
-        consolidateWorker.run({ kind: "strandedCount", sharedDbPath, castaliaPath: r.engine.paths.castalia }).then((n) => finish(Number(n))).catch(() => {
+        consolidateWorker.run({ kind: "strandedCount", sharedDbPath, castaliaPath: r.engine.paths.castalia }).then((stranded) => finish(Array.isArray(stranded) ? stranded : [])).catch(() => {
           setImmediate(() => {
             const done = markPass(`stranded-scan:${r.entry.name}`);
             try {
-              finish(findStrandedSymbols(r.engine.store).length);
+              finish(findStrandedSymbols(r.engine.store));
             } catch {
               strandedRefreshing.delete(r.id);
             } finally {
@@ -95336,7 +95526,7 @@ async function startMultiDaemon(opts = {}) {
         setImmediate(() => {
           const done = markPass(`stranded-scan:${r.entry.name}`);
           try {
-            finish(findStrandedSymbols(r.engine.store).length);
+            finish(findStrandedSymbols(r.engine.store));
           } catch {
             strandedRefreshing.delete(r.id);
           } finally {

package/generalize-worker.mjs

@@ -15993,8 +15993,10 @@ function column(c) {
 function compileNodeWhere(q) {
   if (q.labels.length === 0) throw new Error("selectNodes: no labels");
   const params = [...q.labels];
-  const conds = [`label IN (${q.labels.map(() => "?").join(", ")})`];
-  if (!q.includeClosed) conds.push("valid_to IS NULL");
+  const byId = (q.where ?? []).some((w) => "idIn" in w);
+  const unindexed = byId ? "+" : "";
+  const conds = [`${unindexed}label IN (${q.labels.map(() => "?").join(", ")})`];
+  if (!q.includeClosed) conds.push(`${unindexed}valid_to IS NULL`);
   for (const w of q.where ?? []) {
     if ("embedding" in w) {
       conds.push(`${EMBEDDING_DIM_SQL} ${w.embedding === "present" ? ">" : "="} 0`);
@@ -16046,6 +16048,14 @@ function compileNodeWhere(q) {
 function attrValueSql(p) {
   return `CASE json_type(attrs_json, ${p}) WHEN 'integer' THEN CAST(json_extract(attrs_json, ${p}) AS REAL) ELSE json_extract(attrs_json, ${p}) END`;
 }
+function nodeSelectFrom(q) {
+  if (q.indexedBy) return `nodes INDEXED BY ${selectIndex(q.indexedBy)}`;
+  const where = q.where ?? [];
+  if (!q.includeClosed && where.some((w) => "descriptionIn" in w && w.descriptionIn.length > 0) && !where.some((w) => "idIn" in w)) {
+    return "nodes INDEXED BY nodes_live_description";
+  }
+  return "nodes";
+}
 function compileNodeSelect(q) {
   const { where, params } = compileNodeWhere(q);
   const attrs = q.attrs ?? [];
@@ -16063,7 +16073,7 @@ function compileNodeSelect(q) {
   }
   const orderParams = !q.orderBy && q.labels.length > 1 ? [...q.labels] : [];
   const limit = q.limit !== void 0 ? ` LIMIT ${Math.max(0, Math.floor(q.limit))}` : "";
-  const sql = `SELECT id, label, description, created_at, last_updated_at, page_rank, cumulative_hits, extraction_confidence, extraction_source, community, created_at_seq, last_reinforced_at_seq, valid_to, logical_id, version, ${EMBEDDING_DIM_SQL} AS embedding_dim${q.withEmbedding ? ", embedding_blob, embedding_json" : ""}${attrCols} FROM nodes${q.indexedBy ? ` INDEXED BY ${selectIndex(q.indexedBy)}` : ""} WHERE ${where} ORDER BY ${order}${limit}`;
+  const sql = `SELECT id, label, description, created_at, last_updated_at, page_rank, cumulative_hits, extraction_confidence, extraction_source, community, created_at_seq, last_reinforced_at_seq, valid_to, logical_id, version, ${EMBEDDING_DIM_SQL} AS embedding_dim${q.withEmbedding ? ", embedding_blob, embedding_json" : ""}${attrCols} FROM ${nodeSelectFrom(q)} WHERE ${where} ORDER BY ${order}${limit}`;
   return { sql, params: [...params, ...orderParams] };
 }
 function decodeAttr(value, type) {
@@ -16300,6 +16310,18 @@ CREATE INDEX IF NOT EXISTS edges_live_prior_tag ON edges(created_at) WHERE valid
 -- incremental reindex can fetch just the changed files' symbols instead of
 -- scanning every versioned node. Partial: only live rows are ever queried.
 CREATE INDEX IF NOT EXISTS nodes_relpath ON nodes(json_extract(attrs_json, '$.relPath')) WHERE valid_to IS NULL;
+-- Symbols by qname and doc comments by the symbol they attach to: the drain's
+-- symbol lexicon resolves the names in the text it ships one at a time, instead
+-- of reading every Function/Method/Class/Interface/Const and Comment row on each
+-- live push (~300 MB of a 1.1 GB store per push, 2026-10-20).
+CREATE INDEX IF NOT EXISTS nodes_live_qname ON nodes(json_extract(attrs_json, '$.qname')) WHERE valid_to IS NULL;
+CREATE INDEX IF NOT EXISTS nodes_live_attached_to ON nodes(json_extract(attrs_json, '$.attachedTo')) WHERE valid_to IS NULL AND label = 'Comment';
+-- The attrs the cloud drain gates every instance row on, so it reads them from
+-- the index alone: pulled cloud rows are most of these labels (7,105 of 7,708
+-- Solutions on the live store) at ~10 KB of attrs each, and five full scans for
+-- two attrs read ~220 MB per push to find the few local rows that can ship.
+-- created_at leads so the index order is findNodeFieldsByLabel's.
+CREATE INDEX IF NOT EXISTS nodes_live_instance_gate ON nodes(label, created_at, json_extract(attrs_json, '$.source'), json_extract(attrs_json, '$.privateFinding'), json_extract(attrs_json, '$.resolvedAs'), id) WHERE valid_to IS NULL AND label IN ('Problem', 'Vulnerability', 'Solution', 'RootCause');
 -- Closed rows by relPath, for reviveReobserved (every incremental reindex): the
 -- live-only index above cannot serve it, so it scanned the whole table (~275 MB
 -- of a 1.1 GB store per reindex, 2026-10-20).
@@ -16650,6 +16672,52 @@ var SqliteGraphStore = class {
         AND (t.id IS NULL OR t.valid_to IS NOT NULL)
     `).all();
   }
+  strandedSymbolsForFiles(fileIds) {
+    const out = [];
+    for (let i = 0; i < fileIds.length; i += 500) {
+      const slice = fileIds.slice(i, i + 500);
+      if (slice.length === 0) continue;
+      out.push(...this.db.prepare(`
+        SELECT f.id AS fileId, e.to_id AS symbolId
+        FROM nodes f CROSS JOIN edges e ON e.from_id = f.id
+        LEFT JOIN nodes t ON t.id = e.to_id
+        WHERE f.id IN (${slice.map(() => "?").join(",")}) AND f.label = 'File' AND f.valid_to IS NULL AND e.valid_to IS NULL
+          AND e.type IN ('DEFINES', 'CONTAINS')
+          AND (t.id IS NULL OR t.valid_to IS NOT NULL)
+      `).all(...slice));
+    }
+    return out;
+  }
+  firstLiveByNameOrQname(labels, name) {
+    if (labels.length === 0) return null;
+    const inLabels = labels.map(() => "?").join(",");
+    const probe = (index, match) => this.db.prepare(
+      `SELECT rowid AS rid, created_at AS at, label, description, json_extract(attrs_json, '$.qname') AS qname FROM nodes INDEXED BY ${index}
+          WHERE ${match} = ? AND label IN (${inLabels}) AND valid_to IS NULL`
+    ).all(name, ...labels);
+    const rank = (r) => [labels.indexOf(r.label), r.at, r.rid];
+    const before = (a, b) => a[0] < b[0] || a[0] === b[0] && (a[1] < b[1] || a[1] === b[1] && a[2] < b[2]);
+    let best;
+    for (const r of [...probe("nodes_live_description", "description"), ...probe("nodes_live_qname", "json_extract(attrs_json, '$.qname')")]) {
+      if (!best || before(rank(r), rank(best))) best = r;
+    }
+    return best ? { label: best.label, description: best.description, qname: typeof best.qname === "string" ? best.qname : null } : null;
+  }
+  instanceGateRows(label) {
+    return this.db.prepare(
+      `SELECT id, json_extract(attrs_json, '$.source') AS source, json_extract(attrs_json, '$.privateFinding') AS privateFinding,
+              json_extract(attrs_json, '$.resolvedAs') AS resolvedAs
+         FROM nodes INDEXED BY nodes_live_instance_gate
+        WHERE label IN ('Problem', 'Vulnerability', 'Solution', 'RootCause') AND label = ? AND valid_to IS NULL
+        ORDER BY label, created_at, rowid`
+    ).all(label);
+  }
+  liveCommentsAttachedTo(name) {
+    return this.db.prepare(
+      `SELECT json_extract(attrs_json, '$.kind') AS kind, json_extract(attrs_json, '$.text') AS text, description FROM nodes INDEXED BY nodes_live_attached_to
+        WHERE label = 'Comment' AND valid_to IS NULL AND json_extract(attrs_json, '$.attachedTo') = ? ORDER BY created_at, rowid`
+    ).all(name);
+  }
   strandedEdges() {
     return countStrandedEdges(this.db);
   }
@@ -16828,7 +16896,11 @@ var SqliteGraphStore = class {
     if (!stmt2) {
       const cols = attrKeys.map((k, i) => `, ${attrValueSql(`'$.${k}'`)} AS a${i}, json_type(attrs_json, '$.${k}') AS t${i}`).join("");
       stmt2 = this.db.prepare(
-        `SELECT id, description, community, extraction_confidence${cols} FROM nodes WHERE label = ? AND valid_to IS NULL`
+        // Ordered, not left to the planner: readers take "first row wins" (the
+        // symbol lexicon's name winner and doc pick), and the order otherwise
+        // followed whichever partial index the planner chose. nodes_live_created
+        // serves this order without a sort.
+        `SELECT id, description, community, extraction_confidence${cols} FROM nodes WHERE label = ? AND valid_to IS NULL ORDER BY created_at, rowid`
       );
       this.fieldStmts.set(cacheKey, stmt2);
     }
@@ -16854,7 +16926,7 @@ var SqliteGraphStore = class {
   }
   countNodes(q) {
     const { where, params } = compileNodeWhere(q);
-    return Number(this.db.prepare(`SELECT COUNT(*) AS n FROM nodes WHERE ${where}`).get(...params).n);
+    return Number(this.db.prepare(`SELECT COUNT(*) AS n FROM ${nodeSelectFrom(q)} WHERE ${where}`).get(...params).n);
   }
   setCommunity(id, community) {
     this.stmts.setCommunity.run({ id, v: community });
@@ -18334,6 +18406,62 @@ function buildSymbolLexicon(store) {
   }
   return lex;
 }
+function lazySymbolLexicon(store) {
+  if (!store.firstLiveByNameOrQname || !store.liveCommentsAttachedTo) return null;
+  const labels = SYMBOL_LABELS.map(([label]) => label);
+  const kindOf = new Map(SYMBOL_LABELS.map(([label, kind]) => [label, kind]));
+  const members = /* @__PURE__ */ new Map();
+  const docs = /* @__PURE__ */ new Map();
+  const entries = /* @__PURE__ */ new Map();
+  const member = (name) => {
+    if (!members.has(name)) {
+      const hit = isDistinctiveIdentifier(name) ? store.firstLiveByNameOrQname(labels, name) : null;
+      members.set(name, hit ? { kind: kindOf.get(hit.label), description: hit.description, qname: hit.qname } : null);
+    }
+    return members.get(name);
+  };
+  const docFor = (name) => {
+    if (!docs.has(name)) {
+      let found;
+      for (const c of store.liveCommentsAttachedTo(name)) {
+        if (typeof c.kind !== "string" || !DOC_COMMENT_KINDS.has(c.kind)) continue;
+        const text = typeof c.text === "string" ? c.text : c.description;
+        if (!text?.trim()) continue;
+        found = text.trim().slice(0, MAX_DOC_CHARS);
+        break;
+      }
+      docs.set(name, found);
+    }
+    return docs.get(name);
+  };
+  const isKnown = (tok) => member(tok) !== null;
+  let anySymbol;
+  return {
+    has: isKnown,
+    get(name) {
+      if (!entries.has(name)) {
+        const m = member(name);
+        let entry = null;
+        if (m) {
+          entry = { kind: m.kind };
+          const doc = (m.description ? docFor(m.description) : void 0) ?? (m.qname ? docFor(m.qname) : void 0);
+          for (const phrase of summaryCandidates(name, m.kind, doc)) {
+            if (summaryRejectionReason(phrase, isKnown) === null) {
+              entry.summary = phrase;
+              break;
+            }
+          }
+        }
+        entries.set(name, entry);
+      }
+      return entries.get(name) ?? void 0;
+    },
+    get size() {
+      anySymbol ??= labels.some((label) => store.countNodesByLabel(label) > 0);
+      return anySymbol ? 1 : 0;
+    }
+  };
+}
 var LEXICON_SYMBOL_LABELS = ["Function", "Method", "Class", "Interface", "Const", "Comment"];
 function codeGraphKey(store) {
   const files = store.labelFingerprint?.(["File"]);
@@ -18345,7 +18473,7 @@ function lexiconsFor(store) {
   const key = codeGraphKey(store);
   const hit = key !== null ? lexiconCache.get(store) : void 0;
   if (hit && hit.key === key) return hit;
-  const built = { symbols: buildSymbolLexicon(store), files: buildFileLexicon(store) };
+  const built = { symbols: lazySymbolLexicon(store) ?? buildSymbolLexicon(store), files: buildFileLexicon(store) };
   if (key !== null) lexiconCache.set(store, { key, ...built });
   return built;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inerrata-corporation/errata",
-  "version": "2.0.22-dev.4638",
+  "version": "2.0.22-dev.4652",
   "description": "errata - local-first observation engine for AI coding agents",
   "type": "module",
   "bin": {

package/pass-worker.mjs

@@ -26871,8 +26871,10 @@ function column(c) {
 function compileNodeWhere(q) {
   if (q.labels.length === 0) throw new Error("selectNodes: no labels");
   const params = [...q.labels];
-  const conds = [`label IN (${q.labels.map(() => "?").join(", ")})`];
-  if (!q.includeClosed) conds.push("valid_to IS NULL");
+  const byId = (q.where ?? []).some((w) => "idIn" in w);
+  const unindexed = byId ? "+" : "";
+  const conds = [`${unindexed}label IN (${q.labels.map(() => "?").join(", ")})`];
+  if (!q.includeClosed) conds.push(`${unindexed}valid_to IS NULL`);
   for (const w of q.where ?? []) {
     if ("embedding" in w) {
       conds.push(`${EMBEDDING_DIM_SQL} ${w.embedding === "present" ? ">" : "="} 0`);
@@ -26924,6 +26926,14 @@ function compileNodeWhere(q) {
 function attrValueSql(p) {
   return `CASE json_type(attrs_json, ${p}) WHEN 'integer' THEN CAST(json_extract(attrs_json, ${p}) AS REAL) ELSE json_extract(attrs_json, ${p}) END`;
 }
+function nodeSelectFrom(q) {
+  if (q.indexedBy) return `nodes INDEXED BY ${selectIndex(q.indexedBy)}`;
+  const where = q.where ?? [];
+  if (!q.includeClosed && where.some((w) => "descriptionIn" in w && w.descriptionIn.length > 0) && !where.some((w) => "idIn" in w)) {
+    return "nodes INDEXED BY nodes_live_description";
+  }
+  return "nodes";
+}
 function compileNodeSelect(q) {
   const { where, params } = compileNodeWhere(q);
   const attrs = q.attrs ?? [];
@@ -26941,7 +26951,7 @@ function compileNodeSelect(q) {
   }
   const orderParams = !q.orderBy && q.labels.length > 1 ? [...q.labels] : [];
   const limit = q.limit !== void 0 ? ` LIMIT ${Math.max(0, Math.floor(q.limit))}` : "";
-  const sql = `SELECT id, label, description, created_at, last_updated_at, page_rank, cumulative_hits, extraction_confidence, extraction_source, community, created_at_seq, last_reinforced_at_seq, valid_to, logical_id, version, ${EMBEDDING_DIM_SQL} AS embedding_dim${q.withEmbedding ? ", embedding_blob, embedding_json" : ""}${attrCols} FROM nodes${q.indexedBy ? ` INDEXED BY ${selectIndex(q.indexedBy)}` : ""} WHERE ${where} ORDER BY ${order}${limit}`;
+  const sql = `SELECT id, label, description, created_at, last_updated_at, page_rank, cumulative_hits, extraction_confidence, extraction_source, community, created_at_seq, last_reinforced_at_seq, valid_to, logical_id, version, ${EMBEDDING_DIM_SQL} AS embedding_dim${q.withEmbedding ? ", embedding_blob, embedding_json" : ""}${attrCols} FROM ${nodeSelectFrom(q)} WHERE ${where} ORDER BY ${order}${limit}`;
   return { sql, params: [...params, ...orderParams] };
 }
 function decodeAttr(value, type) {
@@ -27178,6 +27188,18 @@ CREATE INDEX IF NOT EXISTS edges_live_prior_tag ON edges(created_at) WHERE valid
 -- incremental reindex can fetch just the changed files' symbols instead of
 -- scanning every versioned node. Partial: only live rows are ever queried.
 CREATE INDEX IF NOT EXISTS nodes_relpath ON nodes(json_extract(attrs_json, '$.relPath')) WHERE valid_to IS NULL;
+-- Symbols by qname and doc comments by the symbol they attach to: the drain's
+-- symbol lexicon resolves the names in the text it ships one at a time, instead
+-- of reading every Function/Method/Class/Interface/Const and Comment row on each
+-- live push (~300 MB of a 1.1 GB store per push, 2026-10-20).
+CREATE INDEX IF NOT EXISTS nodes_live_qname ON nodes(json_extract(attrs_json, '$.qname')) WHERE valid_to IS NULL;
+CREATE INDEX IF NOT EXISTS nodes_live_attached_to ON nodes(json_extract(attrs_json, '$.attachedTo')) WHERE valid_to IS NULL AND label = 'Comment';
+-- The attrs the cloud drain gates every instance row on, so it reads them from
+-- the index alone: pulled cloud rows are most of these labels (7,105 of 7,708
+-- Solutions on the live store) at ~10 KB of attrs each, and five full scans for
+-- two attrs read ~220 MB per push to find the few local rows that can ship.
+-- created_at leads so the index order is findNodeFieldsByLabel's.
+CREATE INDEX IF NOT EXISTS nodes_live_instance_gate ON nodes(label, created_at, json_extract(attrs_json, '$.source'), json_extract(attrs_json, '$.privateFinding'), json_extract(attrs_json, '$.resolvedAs'), id) WHERE valid_to IS NULL AND label IN ('Problem', 'Vulnerability', 'Solution', 'RootCause');
 -- Closed rows by relPath, for reviveReobserved (every incremental reindex): the
 -- live-only index above cannot serve it, so it scanned the whole table (~275 MB
 -- of a 1.1 GB store per reindex, 2026-10-20).
@@ -27528,6 +27550,52 @@ var SqliteGraphStore = class {
         AND (t.id IS NULL OR t.valid_to IS NOT NULL)
     `).all();
   }
+  strandedSymbolsForFiles(fileIds) {
+    const out2 = [];
+    for (let i2 = 0; i2 < fileIds.length; i2 += 500) {
+      const slice = fileIds.slice(i2, i2 + 500);
+      if (slice.length === 0) continue;
+      out2.push(...this.db.prepare(`
+        SELECT f.id AS fileId, e.to_id AS symbolId
+        FROM nodes f CROSS JOIN edges e ON e.from_id = f.id
+        LEFT JOIN nodes t ON t.id = e.to_id
+        WHERE f.id IN (${slice.map(() => "?").join(",")}) AND f.label = 'File' AND f.valid_to IS NULL AND e.valid_to IS NULL
+          AND e.type IN ('DEFINES', 'CONTAINS')
+          AND (t.id IS NULL OR t.valid_to IS NOT NULL)
+      `).all(...slice));
+    }
+    return out2;
+  }
+  firstLiveByNameOrQname(labels, name2) {
+    if (labels.length === 0) return null;
+    const inLabels = labels.map(() => "?").join(",");
+    const probe = (index, match) => this.db.prepare(
+      `SELECT rowid AS rid, created_at AS at, label, description, json_extract(attrs_json, '$.qname') AS qname FROM nodes INDEXED BY ${index}
+          WHERE ${match} = ? AND label IN (${inLabels}) AND valid_to IS NULL`
+    ).all(name2, ...labels);
+    const rank = (r) => [labels.indexOf(r.label), r.at, r.rid];
+    const before = (a, b) => a[0] < b[0] || a[0] === b[0] && (a[1] < b[1] || a[1] === b[1] && a[2] < b[2]);
+    let best;
+    for (const r of [...probe("nodes_live_description", "description"), ...probe("nodes_live_qname", "json_extract(attrs_json, '$.qname')")]) {
+      if (!best || before(rank(r), rank(best))) best = r;
+    }
+    return best ? { label: best.label, description: best.description, qname: typeof best.qname === "string" ? best.qname : null } : null;
+  }
+  instanceGateRows(label) {
+    return this.db.prepare(
+      `SELECT id, json_extract(attrs_json, '$.source') AS source, json_extract(attrs_json, '$.privateFinding') AS privateFinding,
+              json_extract(attrs_json, '$.resolvedAs') AS resolvedAs
+         FROM nodes INDEXED BY nodes_live_instance_gate
+        WHERE label IN ('Problem', 'Vulnerability', 'Solution', 'RootCause') AND label = ? AND valid_to IS NULL
+        ORDER BY label, created_at, rowid`
+    ).all(label);
+  }
+  liveCommentsAttachedTo(name2) {
+    return this.db.prepare(
+      `SELECT json_extract(attrs_json, '$.kind') AS kind, json_extract(attrs_json, '$.text') AS text, description FROM nodes INDEXED BY nodes_live_attached_to
+        WHERE label = 'Comment' AND valid_to IS NULL AND json_extract(attrs_json, '$.attachedTo') = ? ORDER BY created_at, rowid`
+    ).all(name2);
+  }
   strandedEdges() {
     return countStrandedEdges(this.db);
   }
@@ -27706,7 +27774,11 @@ var SqliteGraphStore = class {
     if (!stmt2) {
       const cols = attrKeys.map((k, i2) => `, ${attrValueSql(`'$.${k}'`)} AS a${i2}, json_type(attrs_json, '$.${k}') AS t${i2}`).join("");
       stmt2 = this.db.prepare(
-        `SELECT id, description, community, extraction_confidence${cols} FROM nodes WHERE label = ? AND valid_to IS NULL`
+        // Ordered, not left to the planner: readers take "first row wins" (the
+        // symbol lexicon's name winner and doc pick), and the order otherwise
+        // followed whichever partial index the planner chose. nodes_live_created
+        // serves this order without a sort.
+        `SELECT id, description, community, extraction_confidence${cols} FROM nodes WHERE label = ? AND valid_to IS NULL ORDER BY created_at, rowid`
       );
       this.fieldStmts.set(cacheKey, stmt2);
     }
@@ -27732,7 +27804,7 @@ var SqliteGraphStore = class {
   }
   countNodes(q) {
     const { where, params } = compileNodeWhere(q);
-    return Number(this.db.prepare(`SELECT COUNT(*) AS n FROM nodes WHERE ${where}`).get(...params).n);
+    return Number(this.db.prepare(`SELECT COUNT(*) AS n FROM ${nodeSelectFrom(q)} WHERE ${where}`).get(...params).n);
   }
   setCommunity(id, community) {
     this.stmts.setCommunity.run({ id, v: community });

package/resources/context-plugin/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "errata-context",
-  "version": "2.0.22-dev.4638",
+  "version": "2.0.22-dev.4652",
   "description": "A minimal 3D view of loaded Errata priors and recent graph bursts.",
   "author": {
     "name": "inErrata"

package/resources/context-plugin/SHA256SUMS

@@ -1,4 +1,4 @@
-16485de5505d71b8dbb138984d557208bdfde9d8bf52baff33f96a477df65145  .codex-plugin/plugin.json
+ba1065832fc364902b70117e8568ae7f0d76cd3af48ed4d0929f03d9d3d55c7f  .codex-plugin/plugin.json
 a7bebae9fb01c838d6c75fc784129e28be571233fa66ff5ee8039b574f35ccf7  assets/icon.png
 fc38b64e4df1e0b72c35764bef6c4c05430d97c9009ad0ecc8a9e75b4deab232  configure-local.mjs
 73dc6028b3b070257c7c31f252d11e55a9e096f4aed0f187f10cb21716e1e6d7  dist/client.js