@inerrata-corporation/errata 2.0.21-dev.3745 → 2.0.21-dev.3763

package/generalize-worker.mjs

@@ -14651,6 +14651,9 @@ var CastaliaIngestPayloadSchema = external_exports.object({
   symbolSummaries: SymbolSummarySidecarSchema.optional()
 });
 
+// ../../packages/shared/src/history-read.ts
+var CLOUD_ATTR_DUPLICATE_KEYS = ["attrsJson", "embedding"];
+
 // ../../packages/shared/src/history-revision.ts
 var Sha = external_exports.string().regex(/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/);
 var HistoryRevisionRequestSchema = external_exports.strictObject({
@@ -14980,7 +14983,59 @@ function installEditPlaceholderRetirement(db) {
   });
 }
 
+// ../../packages/local-graph/src/cloud-attr-compaction.ts
+var CLOUD_ATTR_COMPACTION_KEY = "cloud-attr-compaction-1";
+var DUPES = CLOUD_ATTR_DUPLICATE_KEYS.map((k) => `'${k}'`).join(",");
+function compacted(row) {
+  const packed = `json_extract(${row}.attrs_json, '$.attrsJson')`;
+  const nested = `CASE WHEN json_type(${row}.attrs_json, '$.attrsJson') = 'text' AND json_valid(${packed})
+    AND json_type(${packed}) = 'object' THEN ${packed} ELSE '{}' END`;
+  const value = `CASE WHEN type IN ('array','object') THEN json(value) WHEN type IN ('true','false') THEN json(type) ELSE value END`;
+  return `(SELECT json_group_object(key, ${value}) FROM (
+      SELECT key, value, type, min(pri) FROM (
+        SELECT key, value, type, 0 AS pri FROM json_each(${row}.attrs_json)
+        UNION ALL
+        SELECT key, value, type, 1 AS pri FROM json_each(${nested})
+      ) WHERE key NOT IN (${DUPES}) GROUP BY key))`;
+}
+function dirty(row) {
+  const has = CLOUD_ATTR_DUPLICATE_KEYS.map((k) => `json_type(${row}.attrs_json, '$.${k}') IS NOT NULL`).join(" OR ");
+  return `json_extract(${row}.attrs_json, '$.source') = 'cloud' AND (${has})`;
+}
+function installCloudAttrCompaction(db) {
+  if (db.prepare("SELECT value FROM store_meta WHERE key = ?").get(CLOUD_ATTR_COMPACTION_KEY)) return;
+  db.transaction(() => {
+    for (const operation of ["INSERT", "UPDATE"]) {
+      db.exec(`CREATE TRIGGER IF NOT EXISTS cloud_attr_compact_${operation.toLowerCase()}
+        AFTER ${operation} ON nodes WHEN ${dirty("NEW")}
+        BEGIN UPDATE nodes SET attrs_json = ${compacted("nodes")} WHERE id = NEW.id; END;`);
+    }
+    db.exec(`UPDATE nodes SET attrs_json = ${compacted("nodes")} WHERE ${dirty("nodes")}`);
+    db.prepare("INSERT INTO store_meta(key,value) VALUES (?,?)").run(CLOUD_ATTR_COMPACTION_KEY, "installed");
+  });
+}
+
+// ../../packages/math/src/relevance.ts
+function pageRankScale(ranks) {
+  const positive = [...ranks].filter((r) => Number.isFinite(r) && r > 0).sort((a, b) => a - b);
+  return positive.length ? Math.max(1e-9, positive[Math.floor((positive.length - 1) * 0.95)]) : 1;
+}
+
 // ../../packages/local-graph/src/store.ts
+var PAGE_RANK_SCALE_KEY = "pageRankScale.p95.v1";
+var TEXT_SEARCH_COLUMNS = `id, label, description, page_rank,
+  json_object('test', json(CASE WHEN json_type(attrs_json, '$.test') = 'true' THEN 'true' ELSE 'false' END),
+    'relPath', json_extract(attrs_json, '$.relPath'),
+    'resolutionSuspect', json(CASE WHEN json_type(attrs_json, '$.resolutionSuspect') = 'true' THEN 'true' ELSE 'false' END),
+    'resolvedAt', json_extract(attrs_json, '$.resolvedAt'),
+    'refutedAt', json_extract(attrs_json, '$.refutedAt'),
+    'refuteReasons', json_extract(attrs_json, '$.refuteReasons'),
+    'reopenedFrom', json_extract(attrs_json, '$.reopenedFrom'),
+    'resolvedAs', json_extract(attrs_json, '$.resolvedAs')) AS search_attrs`;
+function textSearchNode(row) {
+  const attrs = JSON.parse(row.search_attrs);
+  return { id: row.id, label: row.label, description: row.description, pageRank: row.page_rank, attrs };
+}
 var LOCAL_RULE_OVERRIDES = {
   ANCHORED_AT: {
     from: EDGE_RULES["ANCHORED_AT"]?.from,
@@ -15124,6 +15179,9 @@ CREATE INDEX IF NOT EXISTS edges_session_evidence ON edges(id) WHERE ${SESSION_E
 DROP INDEX IF EXISTS nodes_live_fingerprint;
 DROP INDEX IF EXISTS nodes_instance_fingerprint;
 CREATE INDEX IF NOT EXISTS nodes_instance_fingerprint_v2 ON nodes(label, last_updated_at, last_reinforced_at_seq, created_at_seq) WHERE valid_to IS NULL AND label IN ('Problem','Vulnerability','Solution','RootCause');
+-- Calibrate rank from a covering scalar index; the global page-rank index
+-- requires a table lookup per node to exclude frozen versions.
+CREATE INDEX IF NOT EXISTS nodes_live_page_rank ON nodes(valid_to, page_rank);
 CREATE INDEX IF NOT EXISTS nodes_logical ON nodes(logical_id, version);
 CREATE INDEX IF NOT EXISTS edges_live ON edges(valid_to);
 -- Expression index on the per-node relPath (stored in attrs_json) so the
@@ -15136,6 +15194,12 @@ CREATE INDEX IF NOT EXISTS nodes_live_description ON nodes(description, label) W
 -- Problems by the file their anchor hint names: the harvest upgrades hints for
 -- every edited file of every turn, and a scan parses every Problem's attrs.
 CREATE INDEX IF NOT EXISTS nodes_anchor_hint_path ON nodes(json_extract(attrs_json, '$.anchorHint.path')) WHERE valid_to IS NULL;
+-- Problems by the command / checker / file that opened them. The generalizer
+-- looked these up per shell or tool EVENT by hydrating every Problem: ~12s of
+-- every minute of daemon main-thread time on a 3.4k-Problem store (2026-10-05).
+CREATE INDEX IF NOT EXISTS nodes_command_signature ON nodes(json_extract(attrs_json, '$.commandSignature')) WHERE valid_to IS NULL;
+CREATE INDEX IF NOT EXISTS nodes_diag_command ON nodes(json_extract(attrs_json, '$.diagCommand')) WHERE valid_to IS NULL;
+CREATE INDEX IF NOT EXISTS nodes_lsp_rel_path ON nodes(json_extract(attrs_json, '$.lspRelPath')) WHERE valid_to IS NULL;
 -- Nodes still waiting on a root-cause ask. The MCP server checks this on EVERY
 -- tool response (the enrichment piggyback), so it must not parse every
 -- Solution's and Problem's attrs per call. Not partial: the Problem pull reads
@@ -15222,6 +15286,7 @@ var SqliteGraphStore = class {
    *  ALL upserts incl. code reindex) — this counter is the knowledge-arrival axis
    *  Δseq is measured over, so code reindex churn must NOT advance it. */
   ingestSeq = 0;
+  initialPageRankScale;
   constructor(opts) {
     this.db = openDatabase(opts.path);
     this.db.exec(SCHEMA_SQL);
@@ -15238,6 +15303,7 @@ var SqliteGraphStore = class {
     this.db.exec(TEMPORAL_INDEX_SQL);
     installHistoryCacheFloor(this.db);
     installEditPlaceholderRetirement(this.db);
+    installCloudAttrCompaction(this.db);
     if (!this.db.prepare("SELECT value FROM store_meta WHERE key = ?").get(SESSION_REPAIR_KEY)) {
       this.db.transaction(() => {
         this.repairSessionAliases();
@@ -15841,6 +15907,34 @@ var SqliteGraphStore = class {
     ).all(...terms.map((t) => `%${t}%`), Math.max(1, limit));
     return rows.map((r) => ({ node: rowToNode(r), matched: Number(r.matched) }));
   }
+  searchTextCandidates(terms) {
+    if (terms.length === 0) return [];
+    const rows = this.db.prepare(`WITH matches AS MATERIALIZED (
+      SELECT n.id, COUNT(*) AS matched FROM nodes n INDEXED BY nodes_live_description, json_each(:terms) t
+      WHERE n.valid_to IS NULL AND n.description LIKE '%' || t.value || '%'
+      GROUP BY n.id
+    ) SELECT ${TEXT_SEARCH_COLUMNS}, matches.matched FROM nodes
+      JOIN matches USING (id) ORDER BY page_rank DESC, last_updated_at DESC`).all({ terms: JSON.stringify(terms) });
+    return rows.map((row) => ({ node: textSearchNode(row), matched: row.matched }));
+  }
+  getTextSearchNode(id) {
+    const row = this.db.prepare(`SELECT ${TEXT_SEARCH_COLUMNS} FROM nodes WHERE id = ? AND valid_to IS NULL`).get(id);
+    return row ? textSearchNode(row) : null;
+  }
+  readPageRankScale() {
+    const rows = this.db.prepare("SELECT page_rank FROM nodes WHERE valid_to IS NULL AND page_rank > 0").all();
+    return pageRankScale(rows.map((row) => row.page_rank));
+  }
+  pageRankScale() {
+    const value = Number(this.getMeta(PAGE_RANK_SCALE_KEY));
+    if (Number.isFinite(value) && value > 0) return value;
+    return this.initialPageRankScale ??= this.readPageRankScale();
+  }
+  refreshPageRankScale() {
+    const scale = this.readPageRankScale();
+    this.setMeta(PAGE_RANK_SCALE_KEY, String(scale));
+    return scale;
+  }
   typedEdgeStmt(sql) {
     let stmt = this.typedEdgeStmts.get(sql);
     if (!stmt) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inerrata-corporation/errata",
-  "version": "2.0.21-dev.3745",
+  "version": "2.0.21-dev.3763",
   "description": "errata - local-first observation engine for AI coding agents",
   "type": "module",
   "bin": {