@inerrata-corporation/errata 2.0.22-dev.4617 → 2.0.22-dev.4629

package/consolidate-worker.mjs

@@ -17179,6 +17179,12 @@ function changeLogCovers(db, sinceSeq) {
   const prunedTo = Number(db.prepare("SELECT value FROM store_meta WHERE key = ?").get(PRUNED_TO_KEY)?.["value"] ?? NaN);
   return Number.isFinite(prunedTo) && sinceSeq >= prunedTo;
 }
+function readChangedNodeIds(db, sinceSeq, labels) {
+  if (!changeLogCovers(db, sinceSeq)) return null;
+  if (!labels.length) return [];
+  const rows = db.prepare(`SELECT DISTINCT id FROM change_log WHERE change_seq > ? AND kind = 'node' AND label IN (${labels.map(() => "?").join(",")})`).all(sinceSeq, ...labels);
+  return rows.map((r) => r.id);
+}
 function readLineageChangedFiles(db, sinceSeq) {
   if (!changeLogCovers(db, sinceSeq)) return null;
   const rows = db.prepare(LINEAGE_CHANGED_FILES_SQL).all({ s: sinceSeq });
@@ -17563,6 +17569,7 @@ var DEMAND_HOT = Math.log1p(10);
 var DEMAND_WARM = Math.log1p(5);
 
 // ../../packages/local-graph/src/node-select.ts
+var SELECT_INDEXES = ["nodes_live_local_community", "nodes_live_fold_candidate"];
 var EMBEDDING_DIM_SQL = "(CASE WHEN length(embedding_blob) >= 4 THEN length(embedding_blob) / 4 ELSE COALESCE(json_array_length(embedding_json), 0) END)";
 var KEY = /^[A-Za-z_][A-Za-z0-9_]*(\.[A-Za-z_][A-Za-z0-9_]*)*$/;
 var COLUMNS = /* @__PURE__ */ new Set([
@@ -17578,6 +17585,10 @@ function path(key) {
   if (!KEY.test(key)) throw new Error(`selectNodes: bad attr key ${key}`);
   return `'$.${key}'`;
 }
+function selectIndex(i) {
+  if (!SELECT_INDEXES.includes(i)) throw new Error(`selectNodes: bad index ${i}`);
+  return i;
+}
 function column(c) {
   if (!COLUMNS.has(c)) throw new Error(`selectNodes: bad column ${c}`);
   return c;
@@ -17593,6 +17604,12 @@ function compileNodeWhere(q) {
     } else if ("idPrefix" in w) {
       conds.push("substr(id, 1, ?) = ?");
       params.push(w.idPrefix.length, w.idPrefix);
+    } else if ("idIn" in w) {
+      if (w.idIn.length === 0) conds.push("0");
+      else {
+        conds.push(`id IN (${w.idIn.map(() => "?").join(", ")})`);
+        params.push(...w.idIn);
+      }
     } else if ("descriptionIn" in w) {
       if (w.descriptionIn.length === 0) conds.push("0");
       else {
@@ -17649,7 +17666,7 @@ 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 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" : ""}${attrCols} FROM nodes${q.indexedBy ? ` INDEXED BY ${selectIndex(q.indexedBy)}` : ""} WHERE ${where} ORDER BY ${order}${limit}`;
   return { sql, params: [...params, ...orderParams] };
 }
 function decodeAttr(value, type) {
@@ -17847,6 +17864,19 @@ var TEMPORAL_INDEX_SQL = `
 CREATE INDEX IF NOT EXISTS nodes_live ON nodes(label, valid_to);
 CREATE INDEX IF NOT EXISTS nodes_live_count ON nodes(valid_to);
 CREATE INDEX IF NOT EXISTS nodes_live_recent ON nodes(label, last_updated_at DESC) WHERE valid_to IS NULL;
+-- "Anything created since the watermark?" guards (the fold, regression and
+-- canonical-bind sweeps run one per session boundary). Without it the planner
+-- walks nodes_live and reads every live row of the label to test created_at:
+-- 56 MB per guard on a 1.1 GB store, to learn there was nothing to do. valid_to
+-- is a key column so the planner prefers this over nodes_live (one more term).
+CREATE INDEX IF NOT EXISTS nodes_live_created ON nodes(label, valid_to, created_at) WHERE valid_to IS NULL;
+-- Community members (the skills pull reads them twice per session boundary):
+-- ~15% of semantic nodes carry a localCommunity, and without this every live
+-- semantic row was read to find them (168 MB \u2192 18 MB per read).
+CREATE INDEX IF NOT EXISTS nodes_live_local_community ON nodes(label, valid_to, json_extract(attrs_json, '$.localCommunity')) WHERE valid_to IS NULL AND json_extract(attrs_json, '$.localCommunity') IS NOT NULL;
+-- Problems parked for the fold judge (read every settle boundary): ~1 in 5 open
+-- Problems; without it every live Problem was read to find them (56 MB).
+CREATE INDEX IF NOT EXISTS nodes_live_fold_candidate ON nodes(label, valid_to) WHERE valid_to IS NULL AND json_type(attrs_json, '$.foldCandidate') IS NOT NULL;
 CREATE INDEX IF NOT EXISTS nodes_session_evidence ON nodes(id) WHERE ${SESSION_EVIDENCE_PREDICATE};
 CREATE INDEX IF NOT EXISTS edges_session_evidence ON edges(id) WHERE ${SESSION_EVIDENCE_PREDICATE};
 -- Covering + partial index for labelFingerprint over the instance labels: the
@@ -17861,6 +17891,14 @@ CREATE INDEX IF NOT EXISTS nodes_instance_fingerprint_v2 ON nodes(label, last_up
 CREATE INDEX IF NOT EXISTS nodes_live_page_rank ON nodes(valid_to, page_rank);
 CREATE INDEX IF NOT EXISTS nodes_logical ON nodes(logical_id, version);
 CREATE INDEX IF NOT EXISTS edges_live ON edges(valid_to);
+-- Newest live edges first, so a "recent N matching edges" read stops early instead
+-- of matching and sorting every live edge (the hourly liveness check: 163 MB \u2192
+-- 69 MB). Named by recentEdgeAttrCoverage; the planner keeps edges_live otherwise.
+CREATE INDEX IF NOT EXISTS edges_live_created ON edges(created_at) WHERE valid_to IS NULL;
+-- Live prior-tag edges by recency: the liveness corroboration check reads the
+-- newest 100 of ~300, and walking all live edges newest-first to find them read
+-- 69 MB. Its predicate is the one recentEdgeAttrCoverage uses for that marker.
+CREATE INDEX IF NOT EXISTS edges_live_prior_tag ON edges(created_at) WHERE valid_to IS NULL AND json_extract(attrs_json, '$.priorTag') = 1;
 -- Expression index on the per-node relPath (stored in attrs_json) so the
 -- incremental reindex can fetch just the changed files' symbols instead of
 -- scanning every versioned node. Partial: only live rows are ever queried.
@@ -17951,6 +17989,7 @@ function rowToEdge(r) {
 }
 var FINGERPRINT_INDEXED_LABELS = ["Problem", "Vulnerability", "Solution", "RootCause"];
 var NODE_LABELS_CHUNK = 256;
+var PRIOR_TAG_MARKER = '"priorTag":true';
 var SqliteGraphStore = class {
   db;
   // Typed edge lookups build their SQL from the number of types asked for, and
@@ -18557,6 +18596,27 @@ var SqliteGraphStore = class {
       cloud: r.cloud === 1
     }));
   }
+  findNodeStampsByLabelAttrs(label, attrKeys) {
+    for (const k of attrKeys) if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(k)) throw new Error(`findNodeStampsByLabelAttrs: bad attr key ${k}`);
+    const keys = [...new Set(attrKeys)];
+    const rows = this.db.prepare(
+      `SELECT id, created_at, created_at_seq,
+                ${keys.map((k, i) => `json_type(attrs_json, '$.${k}') IS NOT NULL AS h${i}`).join(", ")}${keys.length ? "," : ""}
+                COALESCE(json_extract(attrs_json, '$.source') = 'cloud', 0) AS cloud
+           FROM nodes WHERE label = ? AND valid_to IS NULL`
+    ).all(label);
+    const out = new Map(keys.map((k) => [k, []]));
+    for (const r of rows) {
+      keys.forEach((k, i) => out.get(k).push({
+        id: r.id,
+        createdAt: r.created_at,
+        ...r.created_at_seq != null ? { createdAtSeq: r.created_at_seq } : {},
+        has: r[`h${i}`] === 1,
+        cloud: r.cloud === 1
+      }));
+    }
+    return out;
+  }
   findNodeAttrsByLabel(label) {
     const rows = this.stmts.attrsByLabel.all(label);
     return rows.map((r) => ({
@@ -18798,6 +18858,9 @@ var SqliteGraphStore = class {
   lineageChangedFiles(sinceSeq) {
     return readLineageChangedFiles(this.db, sinceSeq);
   }
+  changedNodeIds(sinceSeq, labels) {
+    return readChangedNodeIds(this.db, sinceSeq, labels);
+  }
   pruneChangeLog() {
     return pruneChangeLog(this.db);
   }
@@ -18988,11 +19051,26 @@ var SqliteGraphStore = class {
     ).get(...edgeTypes, label, cloudOnly ? 1 : 0);
     return { count: Number(r.count), total: Number(r.total) };
   }
+  liveEdgeTargets(edgeTypes) {
+    if (edgeTypes.length === 0) return [];
+    const rows = this.db.prepare(`SELECT DISTINCT to_id FROM edges INDEXED BY edges_type WHERE type IN (${edgeTypes.map(() => "?").join(",")}) AND valid_to IS NULL`).all(...edgeTypes);
+    return rows.map((r) => r.to_id);
+  }
   recentEdgeAttrCoverage(marker, attr, limit) {
+    if (marker === PRIOR_TAG_MARKER) {
+      const r2 = this.db.prepare(
+        `SELECT COUNT(*) AS total,
+                  COALESCE(SUM(CASE WHEN json_extract(attrs_json, '$.' || ?) = 1 THEN 1 ELSE 0 END), 0) AS count
+             FROM (SELECT attrs_json FROM edges INDEXED BY edges_live_prior_tag
+                    WHERE valid_to IS NULL AND json_extract(attrs_json, '$.priorTag') = 1
+                    ORDER BY created_at DESC LIMIT ?)`
+      ).get(attr, limit);
+      return { count: Number(r2.count), total: Number(r2.total) };
+    }
     const r = this.db.prepare(
       `SELECT COUNT(*) AS total,
                 COALESCE(SUM(CASE WHEN json_extract(attrs_json, '$.' || ?) = 1 THEN 1 ELSE 0 END), 0) AS count
-           FROM (SELECT attrs_json FROM edges
+           FROM (SELECT attrs_json FROM edges INDEXED BY edges_live_created
                   WHERE valid_to IS NULL AND attrs_json LIKE ?
                   ORDER BY created_at DESC LIMIT ?)`
     ).get(attr, `%${marker}%`, limit);
@@ -19140,8 +19218,9 @@ var SqliteGraphStore = class {
    *  so it must be a plain identifier. */
   nodesAsOfWhereTrue(t, attr) {
     if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(attr)) throw new Error(`nodesAsOfWhereTrue: bad attr ${JSON.stringify(attr)}`);
+    const indexed = attr === "revisit" ? " INDEXED BY nodes_revisit_true" : "";
     const rows = this.db.prepare(
-      `SELECT * FROM nodes WHERE valid_from <= :t AND (valid_to IS NULL OR valid_to > :t)
+      `SELECT * FROM nodes${indexed} WHERE valid_from <= :t AND (valid_to IS NULL OR valid_to > :t)
            AND json_type(attrs_json, '$.${attr}') = 'true'
          ORDER BY rowid`
     ).all({ t });

package/errata.mjs

@@ -19243,6 +19243,12 @@ function changeLogCovers(db, sinceSeq) {
   const prunedTo = Number(db.prepare("SELECT value FROM store_meta WHERE key = ?").get(PRUNED_TO_KEY)?.["value"] ?? NaN);
   return Number.isFinite(prunedTo) && sinceSeq >= prunedTo;
 }
+function readChangedNodeIds(db, sinceSeq, labels) {
+  if (!changeLogCovers(db, sinceSeq)) return null;
+  if (!labels.length) return [];
+  const rows = db.prepare(`SELECT DISTINCT id FROM change_log WHERE change_seq > ? AND kind = 'node' AND label IN (${labels.map(() => "?").join(",")})`).all(sinceSeq, ...labels);
+  return rows.map((r) => r.id);
+}
 function readLineageChangedFiles(db, sinceSeq) {
   if (!changeLogCovers(db, sinceSeq)) return null;
   const rows = db.prepare(LINEAGE_CHANGED_FILES_SQL).all({ s: sinceSeq });
@@ -20376,6 +20382,10 @@ function path(key) {
   if (!KEY.test(key)) throw new Error(`selectNodes: bad attr key ${key}`);
   return `'$.${key}'`;
 }
+function selectIndex(i2) {
+  if (!SELECT_INDEXES.includes(i2)) throw new Error(`selectNodes: bad index ${i2}`);
+  return i2;
+}
 function column(c) {
   if (!COLUMNS.has(c)) throw new Error(`selectNodes: bad column ${c}`);
   return c;
@@ -20391,6 +20401,12 @@ function compileNodeWhere(q) {
     } else if ("idPrefix" in w) {
       conds.push("substr(id, 1, ?) = ?");
       params.push(w.idPrefix.length, w.idPrefix);
+    } else if ("idIn" in w) {
+      if (w.idIn.length === 0) conds.push("0");
+      else {
+        conds.push(`id IN (${w.idIn.map(() => "?").join(", ")})`);
+        params.push(...w.idIn);
+      }
     } else if ("descriptionIn" in w) {
       if (w.descriptionIn.length === 0) conds.push("0");
       else {
@@ -20447,7 +20463,7 @@ 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 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" : ""}${attrCols} FROM nodes${q.indexedBy ? ` INDEXED BY ${selectIndex(q.indexedBy)}` : ""} WHERE ${where} ORDER BY ${order}${limit}`;
   return { sql, params: [...params, ...orderParams] };
 }
 function decodeAttr(value, type) {
@@ -20495,10 +20511,11 @@ function decodeSelectedRow(r, attrs, decodeVector) {
     attrs: out2
   };
 }
-var EMBEDDING_DIM_SQL, KEY, COLUMNS;
+var SELECT_INDEXES, EMBEDDING_DIM_SQL, KEY, COLUMNS;
 var init_node_select = __esm({
   "../../packages/local-graph/src/node-select.ts"() {
     "use strict";
+    SELECT_INDEXES = ["nodes_live_local_community", "nodes_live_fold_candidate"];
     EMBEDDING_DIM_SQL = "(CASE WHEN length(embedding_blob) >= 4 THEN length(embedding_blob) / 4 ELSE COALESCE(json_array_length(embedding_json), 0) END)";
     KEY = /^[A-Za-z_][A-Za-z0-9_]*(\.[A-Za-z_][A-Za-z0-9_]*)*$/;
     COLUMNS = /* @__PURE__ */ new Set([
@@ -20611,7 +20628,7 @@ function rowToEdge(r) {
 function openGraphStore(opts) {
   return new SqliteGraphStore(opts);
 }
-var PAGE_RANK_SCALE_KEY, TEXT_SEARCH_COLUMNS, LOCAL_RULE_OVERRIDES, SCHEMA_VERSION, SESSION_REPAIR_KEY, SESSION_EVIDENCE_PREDICATE, SCHEMA_SQL, TEMPORAL_INDEX_SQL, FINGERPRINT_INDEXED_LABELS, NODE_LABELS_CHUNK, SqliteGraphStore;
+var PAGE_RANK_SCALE_KEY, TEXT_SEARCH_COLUMNS, LOCAL_RULE_OVERRIDES, SCHEMA_VERSION, SESSION_REPAIR_KEY, SESSION_EVIDENCE_PREDICATE, SCHEMA_SQL, TEMPORAL_INDEX_SQL, FINGERPRINT_INDEXED_LABELS, NODE_LABELS_CHUNK, PRIOR_TAG_MARKER, SqliteGraphStore;
 var init_store = __esm({
   "../../packages/local-graph/src/store.ts"() {
     "use strict";
@@ -20756,6 +20773,19 @@ CREATE UNIQUE INDEX IF NOT EXISTS edges_triple ON edges(from_id, type, to_id);
 CREATE INDEX IF NOT EXISTS nodes_live ON nodes(label, valid_to);
 CREATE INDEX IF NOT EXISTS nodes_live_count ON nodes(valid_to);
 CREATE INDEX IF NOT EXISTS nodes_live_recent ON nodes(label, last_updated_at DESC) WHERE valid_to IS NULL;
+-- "Anything created since the watermark?" guards (the fold, regression and
+-- canonical-bind sweeps run one per session boundary). Without it the planner
+-- walks nodes_live and reads every live row of the label to test created_at:
+-- 56 MB per guard on a 1.1 GB store, to learn there was nothing to do. valid_to
+-- is a key column so the planner prefers this over nodes_live (one more term).
+CREATE INDEX IF NOT EXISTS nodes_live_created ON nodes(label, valid_to, created_at) WHERE valid_to IS NULL;
+-- Community members (the skills pull reads them twice per session boundary):
+-- ~15% of semantic nodes carry a localCommunity, and without this every live
+-- semantic row was read to find them (168 MB \u2192 18 MB per read).
+CREATE INDEX IF NOT EXISTS nodes_live_local_community ON nodes(label, valid_to, json_extract(attrs_json, '$.localCommunity')) WHERE valid_to IS NULL AND json_extract(attrs_json, '$.localCommunity') IS NOT NULL;
+-- Problems parked for the fold judge (read every settle boundary): ~1 in 5 open
+-- Problems; without it every live Problem was read to find them (56 MB).
+CREATE INDEX IF NOT EXISTS nodes_live_fold_candidate ON nodes(label, valid_to) WHERE valid_to IS NULL AND json_type(attrs_json, '$.foldCandidate') IS NOT NULL;
 CREATE INDEX IF NOT EXISTS nodes_session_evidence ON nodes(id) WHERE ${SESSION_EVIDENCE_PREDICATE};
 CREATE INDEX IF NOT EXISTS edges_session_evidence ON edges(id) WHERE ${SESSION_EVIDENCE_PREDICATE};
 -- Covering + partial index for labelFingerprint over the instance labels: the
@@ -20770,6 +20800,14 @@ CREATE INDEX IF NOT EXISTS nodes_instance_fingerprint_v2 ON nodes(label, last_up
 CREATE INDEX IF NOT EXISTS nodes_live_page_rank ON nodes(valid_to, page_rank);
 CREATE INDEX IF NOT EXISTS nodes_logical ON nodes(logical_id, version);
 CREATE INDEX IF NOT EXISTS edges_live ON edges(valid_to);
+-- Newest live edges first, so a "recent N matching edges" read stops early instead
+-- of matching and sorting every live edge (the hourly liveness check: 163 MB \u2192
+-- 69 MB). Named by recentEdgeAttrCoverage; the planner keeps edges_live otherwise.
+CREATE INDEX IF NOT EXISTS edges_live_created ON edges(created_at) WHERE valid_to IS NULL;
+-- Live prior-tag edges by recency: the liveness corroboration check reads the
+-- newest 100 of ~300, and walking all live edges newest-first to find them read
+-- 69 MB. Its predicate is the one recentEdgeAttrCoverage uses for that marker.
+CREATE INDEX IF NOT EXISTS edges_live_prior_tag ON edges(created_at) WHERE valid_to IS NULL AND json_extract(attrs_json, '$.priorTag') = 1;
 -- Expression index on the per-node relPath (stored in attrs_json) so the
 -- incremental reindex can fetch just the changed files' symbols instead of
 -- scanning every versioned node. Partial: only live rows are ever queried.
@@ -20803,6 +20841,7 @@ CREATE INDEX IF NOT EXISTS nodes_revisit_true ON nodes(valid_from) WHERE json_ty
 `;
     FINGERPRINT_INDEXED_LABELS = ["Problem", "Vulnerability", "Solution", "RootCause"];
     NODE_LABELS_CHUNK = 256;
+    PRIOR_TAG_MARKER = '"priorTag":true';
     SqliteGraphStore = class {
       db;
       // Typed edge lookups build their SQL from the number of types asked for, and
@@ -21409,6 +21448,27 @@ CREATE INDEX IF NOT EXISTS nodes_revisit_true ON nodes(valid_from) WHERE json_ty
           cloud: r.cloud === 1
         }));
       }
+      findNodeStampsByLabelAttrs(label, attrKeys) {
+        for (const k of attrKeys) if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(k)) throw new Error(`findNodeStampsByLabelAttrs: bad attr key ${k}`);
+        const keys = [...new Set(attrKeys)];
+        const rows = this.db.prepare(
+          `SELECT id, created_at, created_at_seq,
+                ${keys.map((k, i2) => `json_type(attrs_json, '$.${k}') IS NOT NULL AS h${i2}`).join(", ")}${keys.length ? "," : ""}
+                COALESCE(json_extract(attrs_json, '$.source') = 'cloud', 0) AS cloud
+           FROM nodes WHERE label = ? AND valid_to IS NULL`
+        ).all(label);
+        const out2 = new Map(keys.map((k) => [k, []]));
+        for (const r of rows) {
+          keys.forEach((k, i2) => out2.get(k).push({
+            id: r.id,
+            createdAt: r.created_at,
+            ...r.created_at_seq != null ? { createdAtSeq: r.created_at_seq } : {},
+            has: r[`h${i2}`] === 1,
+            cloud: r.cloud === 1
+          }));
+        }
+        return out2;
+      }
       findNodeAttrsByLabel(label) {
         const rows = this.stmts.attrsByLabel.all(label);
         return rows.map((r) => ({
@@ -21650,6 +21710,9 @@ CREATE INDEX IF NOT EXISTS nodes_revisit_true ON nodes(valid_from) WHERE json_ty
       lineageChangedFiles(sinceSeq) {
         return readLineageChangedFiles(this.db, sinceSeq);
       }
+      changedNodeIds(sinceSeq, labels) {
+        return readChangedNodeIds(this.db, sinceSeq, labels);
+      }
       pruneChangeLog() {
         return pruneChangeLog(this.db);
       }
@@ -21840,11 +21903,26 @@ CREATE INDEX IF NOT EXISTS nodes_revisit_true ON nodes(valid_from) WHERE json_ty
         ).get(...edgeTypes, label, cloudOnly ? 1 : 0);
         return { count: Number(r.count), total: Number(r.total) };
       }
+      liveEdgeTargets(edgeTypes) {
+        if (edgeTypes.length === 0) return [];
+        const rows = this.db.prepare(`SELECT DISTINCT to_id FROM edges INDEXED BY edges_type WHERE type IN (${edgeTypes.map(() => "?").join(",")}) AND valid_to IS NULL`).all(...edgeTypes);
+        return rows.map((r) => r.to_id);
+      }
       recentEdgeAttrCoverage(marker, attr, limit) {
+        if (marker === PRIOR_TAG_MARKER) {
+          const r2 = this.db.prepare(
+            `SELECT COUNT(*) AS total,
+                  COALESCE(SUM(CASE WHEN json_extract(attrs_json, '$.' || ?) = 1 THEN 1 ELSE 0 END), 0) AS count
+             FROM (SELECT attrs_json FROM edges INDEXED BY edges_live_prior_tag
+                    WHERE valid_to IS NULL AND json_extract(attrs_json, '$.priorTag') = 1
+                    ORDER BY created_at DESC LIMIT ?)`
+          ).get(attr, limit);
+          return { count: Number(r2.count), total: Number(r2.total) };
+        }
         const r = this.db.prepare(
           `SELECT COUNT(*) AS total,
                 COALESCE(SUM(CASE WHEN json_extract(attrs_json, '$.' || ?) = 1 THEN 1 ELSE 0 END), 0) AS count
-           FROM (SELECT attrs_json FROM edges
+           FROM (SELECT attrs_json FROM edges INDEXED BY edges_live_created
                   WHERE valid_to IS NULL AND attrs_json LIKE ?
                   ORDER BY created_at DESC LIMIT ?)`
         ).get(attr, `%${marker}%`, limit);
@@ -21992,8 +22070,9 @@ CREATE INDEX IF NOT EXISTS nodes_revisit_true ON nodes(valid_from) WHERE json_ty
        *  so it must be a plain identifier. */
       nodesAsOfWhereTrue(t, attr) {
         if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(attr)) throw new Error(`nodesAsOfWhereTrue: bad attr ${JSON.stringify(attr)}`);
+        const indexed = attr === "revisit" ? " INDEXED BY nodes_revisit_true" : "";
         const rows = this.db.prepare(
-          `SELECT * FROM nodes WHERE valid_from <= :t AND (valid_to IS NULL OR valid_to > :t)
+          `SELECT * FROM nodes${indexed} WHERE valid_from <= :t AND (valid_to IS NULL OR valid_to > :t)
            AND json_type(attrs_json, '$.${attr}') = 'true'
          ORDER BY rowid`
         ).all({ t });
@@ -22373,19 +22452,16 @@ function conditionOf(node2) {
 }
 function clearAnsweredRevisits(store, ts) {
   const report = { cleared: 0, standing: 0 };
-  for (const label of SEMANTIC_NODE_LABELS) {
-    for (const row of store.findNodeFieldsByLabel(label, ["revisit"])) {
-      if (row.attrs["revisit"] !== 1 && row.attrs["revisit"] !== true) continue;
-      const node2 = store.getNode(row.id);
-      if (!node2 || node2.attrs["revisit"] !== true) continue;
-      const condition = conditionOf(node2);
-      if (condition === void 0 || !isRevisitAnswered(store, node2, condition)) {
-        report.standing++;
-        continue;
-      }
-      clearRevisit(store, node2.id, ts);
-      report.cleared++;
+  const semantic = new Set(SEMANTIC_NODE_LABELS);
+  for (const node2 of store.nodesAsOfWhereTrue(ts, "revisit")) {
+    if (!semantic.has(node2.label) || node2.validTo != null) continue;
+    const condition = conditionOf(node2);
+    if (condition === void 0 || !isRevisitAnswered(store, node2, condition)) {
+      report.standing++;
+      continue;
     }
+    clearRevisit(store, node2.id, ts);
+    report.cleared++;
   }
   return report;
 }
@@ -26253,7 +26329,10 @@ function communityMembers(store) {
   return store.selectNodes({
     labels: JOINT_COMMUNITY_LABELS,
     attrs: ["localCommunity", "cloudNodeId", "source"],
-    where: [{ attr: "localCommunity", notNullish: true }]
+    where: [{ attr: "localCommunity", notNullish: true }],
+    // ~15% of semantic nodes are members; the ordered select otherwise walks
+    // nodes_live and reads every live semantic row (168 MB on a 1.1 GB store).
+    indexedBy: "nodes_live_local_community"
   });
 }
 function communitySeeds(store, opts = {}) {
@@ -26451,7 +26530,10 @@ async function resolveFoldCandidates(store, opts) {
       { attr: "resolvedAt", nullish: true },
       { attr: "mergedInto", absent: true },
       { attr: "foldCandidate", present: true }
-    ]
+    ],
+    // Only parked Problems are in it; the rowid-ordered select otherwise walks
+    // nodes_live and reads every live Problem (56 MB per settle boundary).
+    indexedBy: "nodes_live_fold_candidate"
   });
   const markOf = (n) => n.attrs["foldCandidate"];
   report.pending = pool.length;
@@ -26783,7 +26865,14 @@ var init_principle_sync = __esm({
 });
 
 // ../../packages/local-graph/src/mechanism-liveness.ts
-function evaluateFed(store, d) {
+function evaluateFed(store, d, stamps) {
+  const incomingRows = d.fedIncoming && store.liveEdgeTargets ? stamps?.(d.fedIncoming.label, "source") : void 0;
+  if (d.fedIncoming && incomingRows) {
+    const targets = new Set(store.liveEdgeTargets(d.fedIncoming.edgeTypes));
+    const rows = d.fedIncoming.cloudOnly ? incomingRows.filter((r) => r.cloud) : incomingRows;
+    const count2 = rows.filter((r) => targets.has(r.id)).length;
+    return { count: count2, total: rows.length, fraction: rows.length === 0 ? 0 : count2 / rows.length };
+  }
   if (d.fedIncoming) {
     const { count: count2, total } = store.incomingEdgeCoverage?.(
       d.fedIncoming.label,
@@ -26803,7 +26892,7 @@ function evaluateFed(store, d) {
   if (!d.fed) return { count: 0, total: 0, fraction: 0 };
   const column2 = COLUMN_INPUTS[d.fed.attr];
   const fed = d.fed;
-  const all = fed.labels.flatMap((label) => store.findNodeStampsByLabel ? store.findNodeStampsByLabel(label, fed.attr) : store.findNodeAttrsByLabel(label));
+  const all = fed.labels.flatMap((label) => stamps?.(label, fed.attr) ?? (store.findNodeStampsByLabel ? store.findNodeStampsByLabel(label, fed.attr) : store.findNodeAttrsByLabel(label)));
   const isCloudMirror = (n) => n.cloud ?? n.attrs?.["source"] === "cloud";
   const nodes = fed.localOnly ? all.filter((n) => !isCloudMirror(n)) : all;
   const has = (n) => column2 ? column2(n) : n.has ?? n.attrs?.[fed.attr] !== void 0;
@@ -26826,8 +26915,16 @@ function evaluateMechanisms(store, invocations = /* @__PURE__ */ new Map(), mech
     if (!counts.has(label)) counts.set(label, store.countNodesByLabel(label));
     return counts.get(label);
   };
+  const wanted = /* @__PURE__ */ new Map();
+  for (const d of mechanisms) for (const l of d.fed?.labels ?? []) wanted.set(l, (wanted.get(l) ?? /* @__PURE__ */ new Set()).add(d.fed.attr));
+  for (const d of mechanisms) if (d.fedIncoming) wanted.set(d.fedIncoming.label, (wanted.get(d.fedIncoming.label) ?? /* @__PURE__ */ new Set()).add("source"));
+  const byLabel = /* @__PURE__ */ new Map();
+  const stamps = store.findNodeStampsByLabelAttrs ? (label, attr) => {
+    if (!byLabel.has(label)) byLabel.set(label, store.findNodeStampsByLabelAttrs(label, [...wanted.get(label) ?? [attr]]));
+    return byLabel.get(label).get(attr);
+  } : void 0;
   return mechanisms.map((d) => {
-    const fed = evaluateFed(store, d);
+    const fed = evaluateFed(store, d, stamps);
     const inv = invocations.get(d.pass);
     const effectCount = d.effectIsLevel ? Math.max(
       0,
@@ -34598,7 +34695,10 @@ function embedCodeNodes(store, opts = {}) {
   return embedNodesByLabels(store, CODE_LABELS, opts);
 }
 function embedSemanticNodes(store, opts = {}) {
-  return embedNodesByLabels(store, SEMANTIC_LABELS, opts);
+  return embedNodesByLabels(store, SEMANTIC_LABELS, {
+    incrementalKey: `embed-scan:semantic:${opts.mode === "hash" ? "hash" : "model"}`,
+    ...opts
+  });
 }
 async function ensureNodeEmbedded(store, nodeId, opts = {}) {
   const node2 = store.getNode(nodeId);
@@ -34671,14 +34771,32 @@ async function embedNodesByLabels(store, labels, opts = {}) {
   };
   const hashOnly = opts.mode === "hash";
   const candidates = [];
-  for (const label of labels) {
-    for (const node2 of store.selectNodes({
-      labels: [label],
-      attrs: EMBED_SCAN_ATTRS
-    })) {
+  const watermarkKey = opts.incrementalKey;
+  const seqBefore = watermarkKey ? store.changeSeq?.() : void 0;
+  let held = null;
+  try {
+    held = watermarkKey ? JSON.parse(store.getMeta?.(watermarkKey) ?? "null") : null;
+  } catch {
+    held = null;
+  }
+  const changedIds = !opts.force && held && seqBefore !== void 0 ? store.changedNodeIds?.(held.seq, labels) ?? null : null;
+  const scan2 = function* () {
+    if (changedIds === null) {
+      for (const label of labels) yield* store.selectNodes({ labels: [label], attrs: EMBED_SCAN_ATTRS });
+      return;
+    }
+    for (let i2 = 0; i2 < changedIds.length; i2 += ID_CHUNK) {
+      yield* store.selectNodes({ labels, attrs: EMBED_SCAN_ATTRS, where: [{ idIn: changedIds.slice(i2, i2 + ID_CHUNK) }] });
+    }
+  };
+  const complete = (version3) => {
+    if (watermarkKey && seqBefore !== void 0 && report.failed === 0) store.setMeta?.(watermarkKey, JSON.stringify({ seq: seqBefore, version: version3 }));
+  };
+  {
+    for (const node2 of scan2()) {
       report.scanned++;
       if (!eligibleForEmbedding(store, node2)) {
-        clearWithdrawnVector(store, node2.id);
+        if (node2.embeddingDim > 0 || node2.attrs["embeddingVersion"] !== void 0) clearWithdrawnVector(store, node2.id);
         continue;
       }
       const existingVersion = node2.attrs["embeddingVersion"] ?? null;
@@ -34693,10 +34811,18 @@ async function embedNodesByLabels(store, labels, opts = {}) {
     }
   }
   if (candidates.length === 0) {
+    complete(held?.version ?? (hashOnly ? EMBEDDING_VERSION : null));
     report.durationMs = Date.now() - started2;
     return report;
   }
   const currentVersion = hashOnly ? EMBEDDING_VERSION : (await embedBatchWithModelOrHash(["probe"]))[0]?.version ?? "unknown";
+  if (changedIds !== null && held?.version && held.version !== currentVersion) {
+    const { incrementalKey: _incremental, ...fullOpts } = opts;
+    return embedNodesByLabels(store, labels, fullOpts).then((full) => {
+      if (full.failed === 0 && watermarkKey && seqBefore !== void 0) store.setMeta?.(watermarkKey, JSON.stringify({ seq: seqBefore, version: currentVersion }));
+      return full;
+    });
+  }
   const todo = [];
   for (const { node: node2, existingVersion } of candidates) {
     if (!opts.force && existingVersion === currentVersion && node2.embeddingDim > 0) {
@@ -34732,6 +34858,7 @@ async function embedNodesByLabels(store, labels, opts = {}) {
     });
     if (opts.progress) opts.progress(report.embedded, todo.length);
   }
+  complete(currentVersion);
   report.durationMs = Date.now() - started2;
   return report;
 }
@@ -34752,7 +34879,7 @@ function findDocstring(store, nodeId) {
   }
   return null;
 }
-var CODE_LABELS, SEMANTIC_LABELS, BATCH_SIZE, EMBED_SCAN_ATTRS;
+var CODE_LABELS, SEMANTIC_LABELS, BATCH_SIZE, ID_CHUNK, EMBED_SCAN_ATTRS;
 var init_embed_code = __esm({
   "../../packages/generalizer/src/embed-code.ts"() {
     "use strict";
@@ -34761,6 +34888,7 @@ var init_embed_code = __esm({
     CODE_LABELS = ["Function", "Method", "Class", "Const"];
     SEMANTIC_LABELS = ["Problem", "RootCause", "Solution", "Pattern", "Claim"];
     BATCH_SIZE = 32;
+    ID_CHUNK = 500;
     EMBED_SCAN_ATTRS = [
       "cloudNodeId",
       "source",
@@ -86568,6 +86696,8 @@ function createLivenessWatch(deps) {
 // src/workspace-maintenance.ts
 init_src8();
 init_src5();
+var WM_CALIBRATION_AT_KEY = "wm-calibration-at";
+var WM_CALIBRATION_INTERVAL_MS = 30 * 6e4;
 function runWorkspaceMaintenance(store, workspaceId2) {
   const seqBefore = store.currentIngestSeq();
   const report = runSemanticMaintenance(store);
@@ -86578,9 +86708,13 @@ function runWorkspaceMaintenance(store, workspaceId2) {
   }
   let calibration = null;
   const calibrationStarted = Date.now();
-  try {
-    calibration = wmCalibrationCells(store);
-  } catch {
+  const lastCalibration = Number(store.getMeta?.(WM_CALIBRATION_AT_KEY) ?? 0);
+  if (!(calibrationStarted - lastCalibration < WM_CALIBRATION_INTERVAL_MS)) {
+    try {
+      calibration = wmCalibrationCells(store);
+      store.setMeta?.(WM_CALIBRATION_AT_KEY, String(calibrationStarted));
+    } catch {
+    }
   }
   return {
     report,
@@ -86711,7 +86845,7 @@ function createWatchBreaker(deps) {
 }
 
 // src/engine.ts
-var DAEMON_VERSION = true ? "2.0.22-dev.4617" : "2.0.0-alpha.0";
+var DAEMON_VERSION = true ? "2.0.22-dev.4629" : "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;

package/generalize-worker.mjs

@@ -15753,6 +15753,12 @@ function changeLogCovers(db, sinceSeq) {
   const prunedTo = Number(db.prepare("SELECT value FROM store_meta WHERE key = ?").get(PRUNED_TO_KEY)?.["value"] ?? NaN);
   return Number.isFinite(prunedTo) && sinceSeq >= prunedTo;
 }
+function readChangedNodeIds(db, sinceSeq, labels) {
+  if (!changeLogCovers(db, sinceSeq)) return null;
+  if (!labels.length) return [];
+  const rows = db.prepare(`SELECT DISTINCT id FROM change_log WHERE change_seq > ? AND kind = 'node' AND label IN (${labels.map(() => "?").join(",")})`).all(sinceSeq, ...labels);
+  return rows.map((r) => r.id);
+}
 function readLineageChangedFiles(db, sinceSeq) {
   if (!changeLogCovers(db, sinceSeq)) return null;
   const rows = db.prepare(LINEAGE_CHANGED_FILES_SQL).all({ s: sinceSeq });
@@ -15960,6 +15966,7 @@ var DEMAND_HOT = Math.log1p(10);
 var DEMAND_WARM = Math.log1p(5);
 
 // ../../packages/local-graph/src/node-select.ts
+var SELECT_INDEXES = ["nodes_live_local_community", "nodes_live_fold_candidate"];
 var EMBEDDING_DIM_SQL = "(CASE WHEN length(embedding_blob) >= 4 THEN length(embedding_blob) / 4 ELSE COALESCE(json_array_length(embedding_json), 0) END)";
 var KEY = /^[A-Za-z_][A-Za-z0-9_]*(\.[A-Za-z_][A-Za-z0-9_]*)*$/;
 var COLUMNS = /* @__PURE__ */ new Set([
@@ -15975,6 +15982,10 @@ function path(key) {
   if (!KEY.test(key)) throw new Error(`selectNodes: bad attr key ${key}`);
   return `'$.${key}'`;
 }
+function selectIndex(i) {
+  if (!SELECT_INDEXES.includes(i)) throw new Error(`selectNodes: bad index ${i}`);
+  return i;
+}
 function column(c) {
   if (!COLUMNS.has(c)) throw new Error(`selectNodes: bad column ${c}`);
   return c;
@@ -15990,6 +16001,12 @@ function compileNodeWhere(q) {
     } else if ("idPrefix" in w) {
       conds.push("substr(id, 1, ?) = ?");
       params.push(w.idPrefix.length, w.idPrefix);
+    } else if ("idIn" in w) {
+      if (w.idIn.length === 0) conds.push("0");
+      else {
+        conds.push(`id IN (${w.idIn.map(() => "?").join(", ")})`);
+        params.push(...w.idIn);
+      }
     } else if ("descriptionIn" in w) {
       if (w.descriptionIn.length === 0) conds.push("0");
       else {
@@ -16046,7 +16063,7 @@ 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 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" : ""}${attrCols} FROM nodes${q.indexedBy ? ` INDEXED BY ${selectIndex(q.indexedBy)}` : ""} WHERE ${where} ORDER BY ${order}${limit}`;
   return { sql, params: [...params, ...orderParams] };
 }
 function decodeAttr(value, type) {
@@ -16244,6 +16261,19 @@ var TEMPORAL_INDEX_SQL = `
 CREATE INDEX IF NOT EXISTS nodes_live ON nodes(label, valid_to);
 CREATE INDEX IF NOT EXISTS nodes_live_count ON nodes(valid_to);
 CREATE INDEX IF NOT EXISTS nodes_live_recent ON nodes(label, last_updated_at DESC) WHERE valid_to IS NULL;
+-- "Anything created since the watermark?" guards (the fold, regression and
+-- canonical-bind sweeps run one per session boundary). Without it the planner
+-- walks nodes_live and reads every live row of the label to test created_at:
+-- 56 MB per guard on a 1.1 GB store, to learn there was nothing to do. valid_to
+-- is a key column so the planner prefers this over nodes_live (one more term).
+CREATE INDEX IF NOT EXISTS nodes_live_created ON nodes(label, valid_to, created_at) WHERE valid_to IS NULL;
+-- Community members (the skills pull reads them twice per session boundary):
+-- ~15% of semantic nodes carry a localCommunity, and without this every live
+-- semantic row was read to find them (168 MB \u2192 18 MB per read).
+CREATE INDEX IF NOT EXISTS nodes_live_local_community ON nodes(label, valid_to, json_extract(attrs_json, '$.localCommunity')) WHERE valid_to IS NULL AND json_extract(attrs_json, '$.localCommunity') IS NOT NULL;
+-- Problems parked for the fold judge (read every settle boundary): ~1 in 5 open
+-- Problems; without it every live Problem was read to find them (56 MB).
+CREATE INDEX IF NOT EXISTS nodes_live_fold_candidate ON nodes(label, valid_to) WHERE valid_to IS NULL AND json_type(attrs_json, '$.foldCandidate') IS NOT NULL;
 CREATE INDEX IF NOT EXISTS nodes_session_evidence ON nodes(id) WHERE ${SESSION_EVIDENCE_PREDICATE};
 CREATE INDEX IF NOT EXISTS edges_session_evidence ON edges(id) WHERE ${SESSION_EVIDENCE_PREDICATE};
 -- Covering + partial index for labelFingerprint over the instance labels: the
@@ -16258,6 +16288,14 @@ CREATE INDEX IF NOT EXISTS nodes_instance_fingerprint_v2 ON nodes(label, last_up
 CREATE INDEX IF NOT EXISTS nodes_live_page_rank ON nodes(valid_to, page_rank);
 CREATE INDEX IF NOT EXISTS nodes_logical ON nodes(logical_id, version);
 CREATE INDEX IF NOT EXISTS edges_live ON edges(valid_to);
+-- Newest live edges first, so a "recent N matching edges" read stops early instead
+-- of matching and sorting every live edge (the hourly liveness check: 163 MB \u2192
+-- 69 MB). Named by recentEdgeAttrCoverage; the planner keeps edges_live otherwise.
+CREATE INDEX IF NOT EXISTS edges_live_created ON edges(created_at) WHERE valid_to IS NULL;
+-- Live prior-tag edges by recency: the liveness corroboration check reads the
+-- newest 100 of ~300, and walking all live edges newest-first to find them read
+-- 69 MB. Its predicate is the one recentEdgeAttrCoverage uses for that marker.
+CREATE INDEX IF NOT EXISTS edges_live_prior_tag ON edges(created_at) WHERE valid_to IS NULL AND json_extract(attrs_json, '$.priorTag') = 1;
 -- Expression index on the per-node relPath (stored in attrs_json) so the
 -- incremental reindex can fetch just the changed files' symbols instead of
 -- scanning every versioned node. Partial: only live rows are ever queried.
@@ -16348,6 +16386,7 @@ function rowToEdge(r) {
 }
 var FINGERPRINT_INDEXED_LABELS = ["Problem", "Vulnerability", "Solution", "RootCause"];
 var NODE_LABELS_CHUNK = 256;
+var PRIOR_TAG_MARKER = '"priorTag":true';
 var SqliteGraphStore = class {
   db;
   // Typed edge lookups build their SQL from the number of types asked for, and
@@ -16954,6 +16993,27 @@ var SqliteGraphStore = class {
       cloud: r.cloud === 1
     }));
   }
+  findNodeStampsByLabelAttrs(label, attrKeys) {
+    for (const k of attrKeys) if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(k)) throw new Error(`findNodeStampsByLabelAttrs: bad attr key ${k}`);
+    const keys = [...new Set(attrKeys)];
+    const rows = this.db.prepare(
+      `SELECT id, created_at, created_at_seq,
+                ${keys.map((k, i) => `json_type(attrs_json, '$.${k}') IS NOT NULL AS h${i}`).join(", ")}${keys.length ? "," : ""}
+                COALESCE(json_extract(attrs_json, '$.source') = 'cloud', 0) AS cloud
+           FROM nodes WHERE label = ? AND valid_to IS NULL`
+    ).all(label);
+    const out = new Map(keys.map((k) => [k, []]));
+    for (const r of rows) {
+      keys.forEach((k, i) => out.get(k).push({
+        id: r.id,
+        createdAt: r.created_at,
+        ...r.created_at_seq != null ? { createdAtSeq: r.created_at_seq } : {},
+        has: r[`h${i}`] === 1,
+        cloud: r.cloud === 1
+      }));
+    }
+    return out;
+  }
   findNodeAttrsByLabel(label) {
     const rows = this.stmts.attrsByLabel.all(label);
     return rows.map((r) => ({
@@ -17195,6 +17255,9 @@ var SqliteGraphStore = class {
   lineageChangedFiles(sinceSeq) {
     return readLineageChangedFiles(this.db, sinceSeq);
   }
+  changedNodeIds(sinceSeq, labels) {
+    return readChangedNodeIds(this.db, sinceSeq, labels);
+  }
   pruneChangeLog() {
     return pruneChangeLog(this.db);
   }
@@ -17385,11 +17448,26 @@ var SqliteGraphStore = class {
     ).get(...edgeTypes, label, cloudOnly ? 1 : 0);
     return { count: Number(r.count), total: Number(r.total) };
   }
+  liveEdgeTargets(edgeTypes) {
+    if (edgeTypes.length === 0) return [];
+    const rows = this.db.prepare(`SELECT DISTINCT to_id FROM edges INDEXED BY edges_type WHERE type IN (${edgeTypes.map(() => "?").join(",")}) AND valid_to IS NULL`).all(...edgeTypes);
+    return rows.map((r) => r.to_id);
+  }
   recentEdgeAttrCoverage(marker, attr, limit) {
+    if (marker === PRIOR_TAG_MARKER) {
+      const r2 = this.db.prepare(
+        `SELECT COUNT(*) AS total,
+                  COALESCE(SUM(CASE WHEN json_extract(attrs_json, '$.' || ?) = 1 THEN 1 ELSE 0 END), 0) AS count
+             FROM (SELECT attrs_json FROM edges INDEXED BY edges_live_prior_tag
+                    WHERE valid_to IS NULL AND json_extract(attrs_json, '$.priorTag') = 1
+                    ORDER BY created_at DESC LIMIT ?)`
+      ).get(attr, limit);
+      return { count: Number(r2.count), total: Number(r2.total) };
+    }
     const r = this.db.prepare(
       `SELECT COUNT(*) AS total,
                 COALESCE(SUM(CASE WHEN json_extract(attrs_json, '$.' || ?) = 1 THEN 1 ELSE 0 END), 0) AS count
-           FROM (SELECT attrs_json FROM edges
+           FROM (SELECT attrs_json FROM edges INDEXED BY edges_live_created
                   WHERE valid_to IS NULL AND attrs_json LIKE ?
                   ORDER BY created_at DESC LIMIT ?)`
     ).get(attr, `%${marker}%`, limit);
@@ -17537,8 +17615,9 @@ var SqliteGraphStore = class {
    *  so it must be a plain identifier. */
   nodesAsOfWhereTrue(t, attr) {
     if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(attr)) throw new Error(`nodesAsOfWhereTrue: bad attr ${JSON.stringify(attr)}`);
+    const indexed = attr === "revisit" ? " INDEXED BY nodes_revisit_true" : "";
     const rows = this.db.prepare(
-      `SELECT * FROM nodes WHERE valid_from <= :t AND (valid_to IS NULL OR valid_to > :t)
+      `SELECT * FROM nodes${indexed} WHERE valid_from <= :t AND (valid_to IS NULL OR valid_to > :t)
            AND json_type(attrs_json, '$.${attr}') = 'true'
          ORDER BY rowid`
     ).all({ t });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inerrata-corporation/errata",
-  "version": "2.0.22-dev.4617",
+  "version": "2.0.22-dev.4629",
   "description": "errata - local-first observation engine for AI coding agents",
   "type": "module",
   "bin": {

package/pass-worker.mjs

@@ -26185,6 +26185,12 @@ function changeLogCovers(db, sinceSeq) {
   const prunedTo = Number(db.prepare("SELECT value FROM store_meta WHERE key = ?").get(PRUNED_TO_KEY)?.["value"] ?? NaN);
   return Number.isFinite(prunedTo) && sinceSeq >= prunedTo;
 }
+function readChangedNodeIds(db, sinceSeq, labels) {
+  if (!changeLogCovers(db, sinceSeq)) return null;
+  if (!labels.length) return [];
+  const rows = db.prepare(`SELECT DISTINCT id FROM change_log WHERE change_seq > ? AND kind = 'node' AND label IN (${labels.map(() => "?").join(",")})`).all(sinceSeq, ...labels);
+  return rows.map((r) => r.id);
+}
 function readLineageChangedFiles(db, sinceSeq) {
   if (!changeLogCovers(db, sinceSeq)) return null;
   const rows = db.prepare(LINEAGE_CHANGED_FILES_SQL).all({ s: sinceSeq });
@@ -26828,6 +26834,7 @@ function pagerankLocal(input) {
 }
 
 // ../../packages/local-graph/src/node-select.ts
+var SELECT_INDEXES = ["nodes_live_local_community", "nodes_live_fold_candidate"];
 var EMBEDDING_DIM_SQL = "(CASE WHEN length(embedding_blob) >= 4 THEN length(embedding_blob) / 4 ELSE COALESCE(json_array_length(embedding_json), 0) END)";
 var KEY = /^[A-Za-z_][A-Za-z0-9_]*(\.[A-Za-z_][A-Za-z0-9_]*)*$/;
 var COLUMNS = /* @__PURE__ */ new Set([
@@ -26843,6 +26850,10 @@ function path(key) {
   if (!KEY.test(key)) throw new Error(`selectNodes: bad attr key ${key}`);
   return `'$.${key}'`;
 }
+function selectIndex(i2) {
+  if (!SELECT_INDEXES.includes(i2)) throw new Error(`selectNodes: bad index ${i2}`);
+  return i2;
+}
 function column(c) {
   if (!COLUMNS.has(c)) throw new Error(`selectNodes: bad column ${c}`);
   return c;
@@ -26858,6 +26869,12 @@ function compileNodeWhere(q) {
     } else if ("idPrefix" in w) {
       conds.push("substr(id, 1, ?) = ?");
       params.push(w.idPrefix.length, w.idPrefix);
+    } else if ("idIn" in w) {
+      if (w.idIn.length === 0) conds.push("0");
+      else {
+        conds.push(`id IN (${w.idIn.map(() => "?").join(", ")})`);
+        params.push(...w.idIn);
+      }
     } else if ("descriptionIn" in w) {
       if (w.descriptionIn.length === 0) conds.push("0");
       else {
@@ -26914,7 +26931,7 @@ 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 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" : ""}${attrCols} FROM nodes${q.indexedBy ? ` INDEXED BY ${selectIndex(q.indexedBy)}` : ""} WHERE ${where} ORDER BY ${order}${limit}`;
   return { sql, params: [...params, ...orderParams] };
 }
 function decodeAttr(value, type) {
@@ -27112,6 +27129,19 @@ var TEMPORAL_INDEX_SQL = `
 CREATE INDEX IF NOT EXISTS nodes_live ON nodes(label, valid_to);
 CREATE INDEX IF NOT EXISTS nodes_live_count ON nodes(valid_to);
 CREATE INDEX IF NOT EXISTS nodes_live_recent ON nodes(label, last_updated_at DESC) WHERE valid_to IS NULL;
+-- "Anything created since the watermark?" guards (the fold, regression and
+-- canonical-bind sweeps run one per session boundary). Without it the planner
+-- walks nodes_live and reads every live row of the label to test created_at:
+-- 56 MB per guard on a 1.1 GB store, to learn there was nothing to do. valid_to
+-- is a key column so the planner prefers this over nodes_live (one more term).
+CREATE INDEX IF NOT EXISTS nodes_live_created ON nodes(label, valid_to, created_at) WHERE valid_to IS NULL;
+-- Community members (the skills pull reads them twice per session boundary):
+-- ~15% of semantic nodes carry a localCommunity, and without this every live
+-- semantic row was read to find them (168 MB \u2192 18 MB per read).
+CREATE INDEX IF NOT EXISTS nodes_live_local_community ON nodes(label, valid_to, json_extract(attrs_json, '$.localCommunity')) WHERE valid_to IS NULL AND json_extract(attrs_json, '$.localCommunity') IS NOT NULL;
+-- Problems parked for the fold judge (read every settle boundary): ~1 in 5 open
+-- Problems; without it every live Problem was read to find them (56 MB).
+CREATE INDEX IF NOT EXISTS nodes_live_fold_candidate ON nodes(label, valid_to) WHERE valid_to IS NULL AND json_type(attrs_json, '$.foldCandidate') IS NOT NULL;
 CREATE INDEX IF NOT EXISTS nodes_session_evidence ON nodes(id) WHERE ${SESSION_EVIDENCE_PREDICATE};
 CREATE INDEX IF NOT EXISTS edges_session_evidence ON edges(id) WHERE ${SESSION_EVIDENCE_PREDICATE};
 -- Covering + partial index for labelFingerprint over the instance labels: the
@@ -27126,6 +27156,14 @@ CREATE INDEX IF NOT EXISTS nodes_instance_fingerprint_v2 ON nodes(label, last_up
 CREATE INDEX IF NOT EXISTS nodes_live_page_rank ON nodes(valid_to, page_rank);
 CREATE INDEX IF NOT EXISTS nodes_logical ON nodes(logical_id, version);
 CREATE INDEX IF NOT EXISTS edges_live ON edges(valid_to);
+-- Newest live edges first, so a "recent N matching edges" read stops early instead
+-- of matching and sorting every live edge (the hourly liveness check: 163 MB \u2192
+-- 69 MB). Named by recentEdgeAttrCoverage; the planner keeps edges_live otherwise.
+CREATE INDEX IF NOT EXISTS edges_live_created ON edges(created_at) WHERE valid_to IS NULL;
+-- Live prior-tag edges by recency: the liveness corroboration check reads the
+-- newest 100 of ~300, and walking all live edges newest-first to find them read
+-- 69 MB. Its predicate is the one recentEdgeAttrCoverage uses for that marker.
+CREATE INDEX IF NOT EXISTS edges_live_prior_tag ON edges(created_at) WHERE valid_to IS NULL AND json_extract(attrs_json, '$.priorTag') = 1;
 -- Expression index on the per-node relPath (stored in attrs_json) so the
 -- incremental reindex can fetch just the changed files' symbols instead of
 -- scanning every versioned node. Partial: only live rows are ever queried.
@@ -27216,6 +27254,7 @@ function rowToEdge(r) {
 }
 var FINGERPRINT_INDEXED_LABELS = ["Problem", "Vulnerability", "Solution", "RootCause"];
 var NODE_LABELS_CHUNK = 256;
+var PRIOR_TAG_MARKER = '"priorTag":true';
 var SqliteGraphStore = class {
   db;
   // Typed edge lookups build their SQL from the number of types asked for, and
@@ -27822,6 +27861,27 @@ var SqliteGraphStore = class {
       cloud: r.cloud === 1
     }));
   }
+  findNodeStampsByLabelAttrs(label, attrKeys) {
+    for (const k of attrKeys) if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(k)) throw new Error(`findNodeStampsByLabelAttrs: bad attr key ${k}`);
+    const keys = [...new Set(attrKeys)];
+    const rows = this.db.prepare(
+      `SELECT id, created_at, created_at_seq,
+                ${keys.map((k, i2) => `json_type(attrs_json, '$.${k}') IS NOT NULL AS h${i2}`).join(", ")}${keys.length ? "," : ""}
+                COALESCE(json_extract(attrs_json, '$.source') = 'cloud', 0) AS cloud
+           FROM nodes WHERE label = ? AND valid_to IS NULL`
+    ).all(label);
+    const out2 = new Map(keys.map((k) => [k, []]));
+    for (const r of rows) {
+      keys.forEach((k, i2) => out2.get(k).push({
+        id: r.id,
+        createdAt: r.created_at,
+        ...r.created_at_seq != null ? { createdAtSeq: r.created_at_seq } : {},
+        has: r[`h${i2}`] === 1,
+        cloud: r.cloud === 1
+      }));
+    }
+    return out2;
+  }
   findNodeAttrsByLabel(label) {
     const rows = this.stmts.attrsByLabel.all(label);
     return rows.map((r) => ({
@@ -28063,6 +28123,9 @@ var SqliteGraphStore = class {
   lineageChangedFiles(sinceSeq) {
     return readLineageChangedFiles(this.db, sinceSeq);
   }
+  changedNodeIds(sinceSeq, labels) {
+    return readChangedNodeIds(this.db, sinceSeq, labels);
+  }
   pruneChangeLog() {
     return pruneChangeLog(this.db);
   }
@@ -28253,11 +28316,26 @@ var SqliteGraphStore = class {
     ).get(...edgeTypes, label, cloudOnly ? 1 : 0);
     return { count: Number(r.count), total: Number(r.total) };
   }
+  liveEdgeTargets(edgeTypes) {
+    if (edgeTypes.length === 0) return [];
+    const rows = this.db.prepare(`SELECT DISTINCT to_id FROM edges INDEXED BY edges_type WHERE type IN (${edgeTypes.map(() => "?").join(",")}) AND valid_to IS NULL`).all(...edgeTypes);
+    return rows.map((r) => r.to_id);
+  }
   recentEdgeAttrCoverage(marker, attr, limit) {
+    if (marker === PRIOR_TAG_MARKER) {
+      const r2 = this.db.prepare(
+        `SELECT COUNT(*) AS total,
+                  COALESCE(SUM(CASE WHEN json_extract(attrs_json, '$.' || ?) = 1 THEN 1 ELSE 0 END), 0) AS count
+             FROM (SELECT attrs_json FROM edges INDEXED BY edges_live_prior_tag
+                    WHERE valid_to IS NULL AND json_extract(attrs_json, '$.priorTag') = 1
+                    ORDER BY created_at DESC LIMIT ?)`
+      ).get(attr, limit);
+      return { count: Number(r2.count), total: Number(r2.total) };
+    }
     const r = this.db.prepare(
       `SELECT COUNT(*) AS total,
                 COALESCE(SUM(CASE WHEN json_extract(attrs_json, '$.' || ?) = 1 THEN 1 ELSE 0 END), 0) AS count
-           FROM (SELECT attrs_json FROM edges
+           FROM (SELECT attrs_json FROM edges INDEXED BY edges_live_created
                   WHERE valid_to IS NULL AND attrs_json LIKE ?
                   ORDER BY created_at DESC LIMIT ?)`
     ).get(attr, `%${marker}%`, limit);
@@ -28405,8 +28483,9 @@ var SqliteGraphStore = class {
    *  so it must be a plain identifier. */
   nodesAsOfWhereTrue(t, attr) {
     if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(attr)) throw new Error(`nodesAsOfWhereTrue: bad attr ${JSON.stringify(attr)}`);
+    const indexed = attr === "revisit" ? " INDEXED BY nodes_revisit_true" : "";
     const rows = this.db.prepare(
-      `SELECT * FROM nodes WHERE valid_from <= :t AND (valid_to IS NULL OR valid_to > :t)
+      `SELECT * FROM nodes${indexed} WHERE valid_from <= :t AND (valid_to IS NULL OR valid_to > :t)
            AND json_type(attrs_json, '$.${attr}') = 'true'
          ORDER BY rowid`
     ).all({ t });
@@ -28492,19 +28571,16 @@ function conditionOf(node) {
 }
 function clearAnsweredRevisits(store2, ts) {
   const report = { cleared: 0, standing: 0 };
-  for (const label of SEMANTIC_NODE_LABELS) {
-    for (const row of store2.findNodeFieldsByLabel(label, ["revisit"])) {
-      if (row.attrs["revisit"] !== 1 && row.attrs["revisit"] !== true) continue;
-      const node = store2.getNode(row.id);
-      if (!node || node.attrs["revisit"] !== true) continue;
-      const condition = conditionOf(node);
-      if (condition === void 0 || !isRevisitAnswered(store2, node, condition)) {
-        report.standing++;
-        continue;
-      }
-      clearRevisit(store2, node.id, ts);
-      report.cleared++;
+  const semantic = new Set(SEMANTIC_NODE_LABELS);
+  for (const node of store2.nodesAsOfWhereTrue(ts, "revisit")) {
+    if (!semantic.has(node.label) || node.validTo != null) continue;
+    const condition = conditionOf(node);
+    if (condition === void 0 || !isRevisitAnswered(store2, node, condition)) {
+      report.standing++;
+      continue;
     }
+    clearRevisit(store2, node.id, ts);
+    report.cleared++;
   }
   return report;
 }
@@ -30107,7 +30183,14 @@ var MECHANISMS = [
 var COLUMN_INPUTS = {
   __createdAtSeq: (n) => n.createdAtSeq !== void 0 && n.createdAtSeq !== null
 };
-function evaluateFed(store2, d) {
+function evaluateFed(store2, d, stamps) {
+  const incomingRows = d.fedIncoming && store2.liveEdgeTargets ? stamps?.(d.fedIncoming.label, "source") : void 0;
+  if (d.fedIncoming && incomingRows) {
+    const targets = new Set(store2.liveEdgeTargets(d.fedIncoming.edgeTypes));
+    const rows = d.fedIncoming.cloudOnly ? incomingRows.filter((r) => r.cloud) : incomingRows;
+    const count2 = rows.filter((r) => targets.has(r.id)).length;
+    return { count: count2, total: rows.length, fraction: rows.length === 0 ? 0 : count2 / rows.length };
+  }
   if (d.fedIncoming) {
     const { count: count2, total } = store2.incomingEdgeCoverage?.(
       d.fedIncoming.label,
@@ -30127,7 +30210,7 @@ function evaluateFed(store2, d) {
   if (!d.fed) return { count: 0, total: 0, fraction: 0 };
   const column2 = COLUMN_INPUTS[d.fed.attr];
   const fed = d.fed;
-  const all = fed.labels.flatMap((label) => store2.findNodeStampsByLabel ? store2.findNodeStampsByLabel(label, fed.attr) : store2.findNodeAttrsByLabel(label));
+  const all = fed.labels.flatMap((label) => stamps?.(label, fed.attr) ?? (store2.findNodeStampsByLabel ? store2.findNodeStampsByLabel(label, fed.attr) : store2.findNodeAttrsByLabel(label)));
   const isCloudMirror = (n) => n.cloud ?? n.attrs?.["source"] === "cloud";
   const nodes = fed.localOnly ? all.filter((n) => !isCloudMirror(n)) : all;
   const has = (n) => column2 ? column2(n) : n.has ?? n.attrs?.[fed.attr] !== void 0;
@@ -30150,8 +30233,16 @@ function evaluateMechanisms(store2, invocations = /* @__PURE__ */ new Map(), mec
     if (!counts.has(label)) counts.set(label, store2.countNodesByLabel(label));
     return counts.get(label);
   };
+  const wanted = /* @__PURE__ */ new Map();
+  for (const d of mechanisms) for (const l of d.fed?.labels ?? []) wanted.set(l, (wanted.get(l) ?? /* @__PURE__ */ new Set()).add(d.fed.attr));
+  for (const d of mechanisms) if (d.fedIncoming) wanted.set(d.fedIncoming.label, (wanted.get(d.fedIncoming.label) ?? /* @__PURE__ */ new Set()).add("source"));
+  const byLabel = /* @__PURE__ */ new Map();
+  const stamps = store2.findNodeStampsByLabelAttrs ? (label, attr) => {
+    if (!byLabel.has(label)) byLabel.set(label, store2.findNodeStampsByLabelAttrs(label, [...wanted.get(label) ?? [attr]]));
+    return byLabel.get(label).get(attr);
+  } : void 0;
   return mechanisms.map((d) => {
-    const fed = evaluateFed(store2, d);
+    const fed = evaluateFed(store2, d, stamps);
     const inv = invocations.get(d.pass);
     const effectCount = d.effectIsLevel ? Math.max(
       0,
@@ -31204,6 +31295,8 @@ async function reconcileStaleFiles(store2, rootPath, workspaceId) {
 }
 
 // src/workspace-maintenance.ts
+var WM_CALIBRATION_AT_KEY = "wm-calibration-at";
+var WM_CALIBRATION_INTERVAL_MS = 30 * 6e4;
 function runWorkspaceMaintenance(store2, workspaceId) {
   const seqBefore = store2.currentIngestSeq();
   const report = runSemanticMaintenance(store2);
@@ -31214,9 +31307,13 @@ function runWorkspaceMaintenance(store2, workspaceId) {
   }
   let calibration = null;
   const calibrationStarted = Date.now();
-  try {
-    calibration = wmCalibrationCells(store2);
-  } catch {
+  const lastCalibration = Number(store2.getMeta?.(WM_CALIBRATION_AT_KEY) ?? 0);
+  if (!(calibrationStarted - lastCalibration < WM_CALIBRATION_INTERVAL_MS)) {
+    try {
+      calibration = wmCalibrationCells(store2);
+      store2.setMeta?.(WM_CALIBRATION_AT_KEY, String(calibrationStarted));
+    } catch {
+    }
   }
   return {
     report,

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

@@ -1,6 +1,6 @@
 {
   "name": "errata-context",
-  "version": "2.0.22-dev.4617",
+  "version": "2.0.22-dev.4629",
   "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 @@
-81a5a7875eb137fa180465a687ecfd2fd76f67967cec97962a20f6e695c81d2f  .codex-plugin/plugin.json
+6116713b2c7fa083cb68622899bbfe5c62167d3c1b8d6d62ec2c81e4cc391baa  .codex-plugin/plugin.json
 a7bebae9fb01c838d6c75fc784129e28be571233fa66ff5ee8039b574f35ccf7  assets/icon.png
 fc38b64e4df1e0b72c35764bef6c4c05430d97c9009ad0ecc8a9e75b4deab232  configure-local.mjs
 73dc6028b3b070257c7c31f252d11e55a9e096f4aed0f187f10cb21716e1e6d7  dist/client.js