@jhizzard/termdeck 0.9.0 → 0.10.0

package/packages/server/src/graph-routes.js

@@ -0,0 +1,555 @@
+'use strict';
+
+// Graph routes — Sprint 38 T4. Powers the D3 force-directed knowledge-graph
+// view in the dashboard.
+//
+// Endpoints:
+//   GET /api/graph/project/:name          per-project node + edge set
+//   GET /api/graph/memory/:id?depth=2     per-memory N-hop neighborhood
+//   GET /api/graph/stats                  global topology counts
+//
+// All three accept the same { app, getPool } injection so tests can stub the
+// pg pool. getPool() returns null when DATABASE_URL is absent or the pool
+// failed to initialize — endpoints respond `{ enabled: false }` in that case
+// (mirrors the rumen endpoints' graceful-degrade pattern).
+//
+// Schema notes:
+//   - memory_items.project is a single text column (not array).
+//   - memory_relationships.relationship_type is the edge type (CHECK enforces
+//     a vocabulary; T1's migration 009 expands to 8 values).
+//   - T2 may add weight + inferred_at + inferred_by mid-sprint. We SELECT them
+//     conditionally via to_jsonb so missing columns don't break the query.
+
+const UUID_RE = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
+const PROJECT_RE = /^[a-zA-Z0-9][a-zA-Z0-9._-]{0,63}$/;
+const NODE_LABEL_LEN = 200;
+const MAX_DEPTH = 4;
+const DEFAULT_DEPTH = 2;
+const MAX_NODES_PER_PROJECT = 2000;
+
+function snippet(content, len = NODE_LABEL_LEN) {
+  if (!content) return '';
+  const s = String(content).replace(/\s+/g, ' ').trim();
+  return s.length > len ? s.slice(0, len - 1) + '…' : s;
+}
+
+function asIso(v) {
+  if (!v) return null;
+  if (v instanceof Date) return v.toISOString();
+  return v;
+}
+
+// SELECT shape — keep column list explicit so additions to memory_items
+// (Sprint 39+) don't accidentally leak into the API.
+const NODE_COLUMNS = [
+  'id', 'content', 'source_type', 'category', 'project',
+  'created_at', 'updated_at', 'is_active', 'archived', 'superseded_by',
+];
+const NODE_COLUMNS_SQL = NODE_COLUMNS.join(', ');
+const NODE_COLUMNS_PREFIXED = (alias) => NODE_COLUMNS.map((c) => `${alias}.${c}`).join(', ');
+
+const EDGE_COLUMNS_BASE = ['id', 'source_id', 'target_id', 'relationship_type', 'created_at'];
+const EDGE_COLUMNS_BASE_SQL = EDGE_COLUMNS_BASE.join(', ');
+
+// T2 will add weight/inferred_at/inferred_by mid-sprint. Use to_jsonb so a
+// SELECT on a fresh pre-T2 box doesn't error on missing columns. The to_jsonb
+// row carries whichever columns exist; we read weight/inferred_at/inferred_by
+// off the resulting JSON.
+const EDGE_COLUMNS_T2_SQL = `to_jsonb(memory_relationships) AS _row`;
+
+function rowToNode(row) {
+  const createdAt = asIso(row.created_at);
+  const updatedAt = asIso(row.updated_at);
+  const ageMs = createdAt ? Date.now() - new Date(createdAt).getTime() : null;
+  const ageDays = ageMs == null ? null : ageMs / 86_400_000;
+  return {
+    id: row.id,
+    label: snippet(row.content, 80),
+    snippet: snippet(row.content, NODE_LABEL_LEN),
+    source_type: row.source_type || null,
+    category: row.category || null,
+    project: row.project || 'global',
+    createdAt,
+    updatedAt,
+    ageDays: ageDays == null ? null : Number(ageDays.toFixed(2)),
+    degree: typeof row.degree === 'number' ? row.degree : Number(row.degree || 0),
+    superseded: !!row.superseded_by,
+  };
+}
+
+function rowToEdge(row) {
+  const meta = row._row && typeof row._row === 'object' ? row._row : {};
+  return {
+    id: row.id,
+    source: row.source_id,
+    target: row.target_id,
+    kind: row.relationship_type,
+    createdAt: asIso(row.created_at),
+    weight: typeof meta.weight === 'number' ? meta.weight : null,
+    inferredAt: asIso(meta.inferred_at),
+    inferredBy: meta.inferred_by || null,
+  };
+}
+
+function rowToFullMemory(row) {
+  return {
+    id: row.id,
+    content: row.content || '',
+    source_type: row.source_type || null,
+    category: row.category || null,
+    project: row.project || 'global',
+    createdAt: asIso(row.created_at),
+    updatedAt: asIso(row.updated_at),
+    isActive: row.is_active !== false,
+    archived: !!row.archived,
+    supersededBy: row.superseded_by || null,
+  };
+}
+
+function disabledPayload(extra = {}) {
+  return Object.assign({ enabled: false, reason: 'DATABASE_URL not configured' }, extra);
+}
+
+async function fetchProjectGraph(pool, projectName) {
+  // One round-trip:
+  //   1. nodes for the project (with degree computed via subquery so we don't
+  //      pull edges into JS just to count).
+  //   2. edges where BOTH endpoints belong to the project's node set.
+  //
+  // We deliberately scope to is_active=true and archived=false (matches the
+  // mnestra recall convention) so superseded/archived noise stays out of the
+  // visualization.
+  const nodesSql = `
+    WITH proj_nodes AS (
+      SELECT ${NODE_COLUMNS_SQL}
+        FROM memory_items
+       WHERE project = $1
+         AND is_active = TRUE
+         AND archived = FALSE
+       ORDER BY created_at DESC
+       LIMIT ${MAX_NODES_PER_PROJECT}
+    )
+    SELECT
+      n.*,
+      COALESCE((
+        SELECT COUNT(*)::int
+          FROM memory_relationships r
+         WHERE r.source_id = n.id OR r.target_id = n.id
+      ), 0) AS degree
+      FROM proj_nodes n
+  `;
+  const nodesRes = await pool.query(nodesSql, [projectName]);
+  const nodes = nodesRes.rows.map(rowToNode);
+  const idSet = new Set(nodes.map((n) => n.id));
+  if (idSet.size === 0) {
+    return { nodes: [], edges: [] };
+  }
+  const ids = Array.from(idSet);
+
+  // ANY($1::uuid[]) is the safe way to pass a uuid array param.
+  const edgesSql = `
+    SELECT ${EDGE_COLUMNS_BASE_SQL}, ${EDGE_COLUMNS_T2_SQL}
+      FROM memory_relationships
+     WHERE source_id = ANY($1::uuid[])
+       AND target_id = ANY($1::uuid[])
+  `;
+  const edgesRes = await pool.query(edgesSql, [ids]);
+  const edges = edgesRes.rows.map(rowToEdge);
+  return { nodes, edges };
+}
+
+async function fetchNeighborhood(pool, rootId, depth) {
+  // Inline recursive CTE so T4 ships independently of T1's
+  // expand_memory_neighborhood RPC. When that RPC lands, the CTE here can be
+  // replaced with `SELECT * FROM expand_memory_neighborhood($1, $2)` without
+  // changing the return shape (memory_id, depth).
+  //
+  // Cycle-safety: the path[] column accumulates visited ids; we only descend
+  // into ids not already in the path. Caps at depth so a runaway loop can't
+  // OOM the server.
+  const safeDepth = Math.max(1, Math.min(MAX_DEPTH, depth));
+  const sql = `
+    WITH RECURSIVE walk AS (
+      SELECT $1::uuid AS memory_id, 0 AS depth, ARRAY[$1::uuid] AS path
+      UNION ALL
+      SELECT
+        CASE WHEN r.source_id = w.memory_id THEN r.target_id ELSE r.source_id END AS memory_id,
+        w.depth + 1 AS depth,
+        w.path || (CASE WHEN r.source_id = w.memory_id THEN r.target_id ELSE r.source_id END)
+        FROM walk w
+        JOIN memory_relationships r
+          ON (r.source_id = w.memory_id OR r.target_id = w.memory_id)
+       WHERE w.depth < ${safeDepth}
+         AND NOT (
+           (CASE WHEN r.source_id = w.memory_id THEN r.target_id ELSE r.source_id END)
+             = ANY (w.path)
+         )
+    )
+    SELECT DISTINCT memory_id, MIN(depth) AS depth
+      FROM walk
+     GROUP BY memory_id
+  `;
+  const walkRes = await pool.query(sql, [rootId]);
+  const memoryIds = walkRes.rows.map((r) => r.memory_id);
+  const depthByMemoryId = new Map(walkRes.rows.map((r) => [r.memory_id, Number(r.depth)]));
+  if (memoryIds.length === 0) {
+    return { nodes: [], edges: [], depthByMemoryId };
+  }
+
+  // Fetch full memory rows for each id in the walk + their degree.
+  const nodesSql = `
+    SELECT
+      ${NODE_COLUMNS_PREFIXED('m')},
+      COALESCE((
+        SELECT COUNT(*)::int
+          FROM memory_relationships r
+         WHERE r.source_id = m.id OR r.target_id = m.id
+      ), 0) AS degree
+      FROM memory_items m
+     WHERE m.id = ANY($1::uuid[])
+       AND m.is_active = TRUE
+       AND m.archived = FALSE
+  `;
+  const nodesRes = await pool.query(nodesSql, [memoryIds]);
+  const nodes = nodesRes.rows.map((row) => {
+    const node = rowToNode(row);
+    node.depth = depthByMemoryId.get(row.id) ?? 0;
+    return node;
+  });
+
+  // Edges where BOTH endpoints are in the walk.
+  const edgesSql = `
+    SELECT ${EDGE_COLUMNS_BASE_SQL}, ${EDGE_COLUMNS_T2_SQL}
+      FROM memory_relationships
+     WHERE source_id = ANY($1::uuid[])
+       AND target_id = ANY($1::uuid[])
+  `;
+  const edgesRes = await pool.query(edgesSql, [memoryIds]);
+  const edges = edgesRes.rows.map(rowToEdge);
+  return { nodes, edges, depthByMemoryId };
+}
+
+async function fetchStats(pool) {
+  const memoriesSql = `
+    SELECT
+      COUNT(*)::int AS total,
+      COUNT(*) FILTER (WHERE is_active = TRUE AND archived = FALSE)::int AS active,
+      COUNT(DISTINCT project)::int AS projects
+      FROM memory_items
+  `;
+  const edgesSql = `
+    SELECT
+      COUNT(*)::int AS total,
+      relationship_type AS kind,
+      COUNT(*)::int AS by_type
+      FROM memory_relationships
+     GROUP BY ROLLUP(relationship_type)
+  `;
+  const projectSql = `
+    SELECT project, COUNT(*)::int AS n
+      FROM memory_items
+     WHERE is_active = TRUE AND archived = FALSE
+     GROUP BY project
+     ORDER BY n DESC
+     LIMIT 30
+  `;
+  // T2's inferred_at column may not exist yet — guard with a try/catch and
+  // fall back to a null lastInferredAt so the endpoint stays alive on a
+  // pre-T2 box.
+  const lastInferredSqlRaw = `
+    SELECT to_jsonb(memory_relationships) - 'embedding' AS _row
+      FROM memory_relationships
+     ORDER BY id DESC
+     LIMIT 1
+  `;
+
+  const [memoriesRes, edgesRes, projectsRes] = await Promise.all([
+    pool.query(memoriesSql),
+    pool.query(edgesSql),
+    pool.query(projectSql),
+  ]);
+
+  const memoriesRow = memoriesRes.rows[0] || { total: 0, active: 0, projects: 0 };
+
+  let totalEdges = 0;
+  const byType = {};
+  for (const row of edgesRes.rows) {
+    if (row.kind === null) {
+      // ROLLUP NULL row carries the grand total.
+      totalEdges = Number(row.by_type || 0);
+    } else {
+      byType[row.kind] = Number(row.by_type || 0);
+    }
+  }
+
+  const byProject = {};
+  for (const row of projectsRes.rows) {
+    byProject[row.project] = Number(row.n || 0);
+  }
+
+  let lastInferredAt = null;
+  let lastInferredBy = null;
+  try {
+    const r = await pool.query(lastInferredSqlRaw);
+    const meta = r.rows[0] && r.rows[0]._row ? r.rows[0]._row : null;
+    if (meta) {
+      lastInferredAt = asIso(meta.inferred_at) || null;
+      lastInferredBy = meta.inferred_by || null;
+    }
+  } catch (_e) {
+    // pre-T2: column or extension missing — leave nulls.
+  }
+
+  return {
+    enabled: true,
+    memories: {
+      total: Number(memoriesRow.total || 0),
+      active: Number(memoriesRow.active || 0),
+      projects: Number(memoriesRow.projects || 0),
+    },
+    edges: {
+      total: totalEdges,
+      byType,
+    },
+    byProject,
+    lastInferredAt,
+    lastInferredBy,
+    generatedAt: new Date().toISOString(),
+  };
+}
+
+// Sprint 38 T2 — inference-pipeline stats. Distinct from the topology-focused
+// /api/graph/stats above: this answers "is the graph-inference cron healthy?"
+// using memory_relationships' inferred_* columns plus pg_cron's job_run_details.
+//
+// All optional columns are read via to_jsonb so a pre-T1 box (no inferred_at /
+// inferred_by / weight columns) still gets a usable response with nulls.
+async function fetchInferenceStats(pool) {
+  const totalsSql = `
+    SELECT
+      COUNT(*)::int AS total,
+      (
+        SELECT COUNT(*)::int FROM memory_items
+         WHERE is_active = TRUE AND archived = FALSE
+           AND id NOT IN (
+             SELECT source_id FROM memory_relationships
+             UNION
+             SELECT target_id FROM memory_relationships
+           )
+      ) AS orphan_memories
+      FROM memory_relationships
+  `;
+
+  // cron-tagged rows + last cron-inferred timestamp. Anonymous to_jsonb dance
+  // so a missing column on a pre-T1 box returns nulls rather than errors.
+  const cronSummarySql = `
+    SELECT to_jsonb(memory_relationships) AS _row
+      FROM memory_relationships
+     ORDER BY id DESC
+     LIMIT 200
+  `;
+
+  const [totalsRes, sampleRes] = await Promise.all([
+    pool.query(totalsSql),
+    pool.query(cronSummarySql).catch(() => ({ rows: [] })),
+  ]);
+
+  const totalsRow = totalsRes.rows[0] || { total: 0, orphan_memories: 0 };
+
+  let cronInferredEdges = 0;
+  let lastInferenceAt = null;
+  for (const row of sampleRes.rows) {
+    const meta = row._row && typeof row._row === 'object' ? row._row : null;
+    if (!meta) continue;
+    const inferredBy = typeof meta.inferred_by === 'string' ? meta.inferred_by : null;
+    if (inferredBy && inferredBy.startsWith('cron-')) {
+      cronInferredEdges++;
+      const at = asIso(meta.inferred_at);
+      if (at && (!lastInferenceAt || at > lastInferenceAt)) {
+        lastInferenceAt = at;
+      }
+    }
+  }
+
+  // Authoritative count for cron-inferred edges (separate query so the sample
+  // above stays a fast probe; this one only runs after the totals land).
+  let cronInferredTotal = cronInferredEdges;
+  let lastCronInferredAt = lastInferenceAt;
+  try {
+    const exactRes = await pool.query(
+      `SELECT COUNT(*)::int AS n, MAX(inferred_at) AS last_at
+         FROM memory_relationships
+        WHERE inferred_by ILIKE 'cron-%'`,
+    );
+    const exactRow = exactRes.rows[0];
+    if (exactRow) {
+      cronInferredTotal = Number(exactRow.n || 0);
+      lastCronInferredAt = asIso(exactRow.last_at) || lastInferenceAt;
+    }
+  } catch (_e) {
+    // Pre-T1: inferred_by column missing — fall back to the sample probe.
+  }
+
+  // Last 5 graph-inference-tick runs from pg_cron's job_run_details.
+  // Wrapped in try/catch because some Supabase plans gate access to cron.*
+  // tables and we'd rather return nulls than 500.
+  let recentRuns = [];
+  let lastRunDurationMs = null;
+  try {
+    const cronRes = await pool.query(
+      `SELECT
+         start_time, end_time, status, return_message,
+         EXTRACT(EPOCH FROM (end_time - start_time)) * 1000 AS duration_ms
+         FROM cron.job_run_details
+        WHERE jobname = 'graph-inference-tick'
+        ORDER BY start_time DESC
+        LIMIT 5`,
+    );
+    recentRuns = cronRes.rows.map((row) => ({
+      startedAt: asIso(row.start_time),
+      endedAt: asIso(row.end_time),
+      status: row.status || null,
+      durationMs: row.duration_ms == null ? null : Number(row.duration_ms),
+      message: row.return_message || null,
+    }));
+    if (recentRuns.length > 0 && recentRuns[0].durationMs != null) {
+      lastRunDurationMs = recentRuns[0].durationMs;
+    }
+  } catch (_e) {
+    // pg_cron not enabled or cron schema not readable — recentRuns stays [].
+  }
+
+  return {
+    enabled: true,
+    totalEdges: Number(totalsRow.total || 0),
+    cronInferredEdges: cronInferredTotal,
+    orphanMemories: Number(totalsRow.orphan_memories || 0),
+    lastInferenceAt: lastCronInferredAt,
+    lastRunDurationMs,
+    recentRuns,
+    generatedAt: new Date().toISOString(),
+  };
+}
+
+function createGraphRoutes({ app, getPool }) {
+  if (!app) throw new Error('app required');
+  if (typeof getPool !== 'function') throw new Error('getPool callback required');
+
+  app.get('/api/graph/project/:name', async (req, res) => {
+    const project = req.params.name;
+    if (!project || !PROJECT_RE.test(project)) {
+      return res.status(400).json({ error: 'invalid project name' });
+    }
+    const pool = getPool();
+    if (!pool) return res.json(disabledPayload({ project, nodes: [], edges: [] }));
+
+    try {
+      const { nodes, edges } = await fetchProjectGraph(pool, project);
+      const byType = {};
+      for (const e of edges) {
+        byType[e.kind] = (byType[e.kind] || 0) + 1;
+      }
+      res.json({
+        enabled: true,
+        project,
+        stats: {
+          nodes: nodes.length,
+          edges: edges.length,
+          byType,
+          truncated: nodes.length >= MAX_NODES_PER_PROJECT,
+        },
+        nodes,
+        edges,
+      });
+    } catch (err) {
+      console.warn('[graph] /api/graph/project failed:', err.message);
+      res.status(500).json({ error: 'graph query failed', detail: err.message });
+    }
+  });
+
+  app.get('/api/graph/memory/:id', async (req, res) => {
+    const id = req.params.id;
+    if (!UUID_RE.test(id)) {
+      return res.status(400).json({ error: 'invalid memory id' });
+    }
+    const depthRaw = parseInt(req.query.depth, 10);
+    const depth = Number.isFinite(depthRaw) ? depthRaw : DEFAULT_DEPTH;
+
+    const pool = getPool();
+    if (!pool) return res.json(disabledPayload({ root: null, nodes: [], edges: [] }));
+
+    try {
+      // Fetch the root memory in full first so we can return content even when
+      // it has no edges yet.
+      const rootRes = await pool.query(
+        `SELECT ${NODE_COLUMNS} FROM memory_items WHERE id = $1`,
+        [id],
+      );
+      if (rootRes.rows.length === 0) {
+        return res.status(404).json({ error: 'memory not found' });
+      }
+      const root = rowToFullMemory(rootRes.rows[0]);
+
+      const { nodes, edges } = await fetchNeighborhood(pool, id, depth);
+      res.json({
+        enabled: true,
+        root,
+        depth: Math.max(1, Math.min(MAX_DEPTH, depth)),
+        nodes,
+        edges,
+        stats: {
+          nodes: nodes.length,
+          edges: edges.length,
+        },
+      });
+    } catch (err) {
+      console.warn('[graph] /api/graph/memory failed:', err.message);
+      res.status(500).json({ error: 'graph query failed', detail: err.message });
+    }
+  });
+
+  app.get('/api/graph/stats', async (req, res) => {
+    const pool = getPool();
+    if (!pool) return res.json(disabledPayload());
+    try {
+      const stats = await fetchStats(pool);
+      res.json(stats);
+    } catch (err) {
+      console.warn('[graph] /api/graph/stats failed:', err.message);
+      res.status(500).json({ error: 'graph stats query failed', detail: err.message });
+    }
+  });
+
+  // Sprint 38 T2 — operational view of the graph-inference cron pipeline.
+  // Distinct from /api/graph/stats (T4 topology). Express routes by exact
+  // path so the more-specific /stats/inference does not shadow /stats.
+  app.get('/api/graph/stats/inference', async (req, res) => {
+    const pool = getPool();
+    if (!pool) return res.json(disabledPayload());
+    try {
+      const stats = await fetchInferenceStats(pool);
+      res.json(stats);
+    } catch (err) {
+      console.warn('[graph] /api/graph/stats/inference failed:', err.message);
+      res.status(500).json({ error: 'graph inference stats query failed', detail: err.message });
+    }
+  });
+}
+
+module.exports = {
+  createGraphRoutes,
+  // Exported for tests + reuse:
+  fetchProjectGraph,
+  fetchNeighborhood,
+  fetchStats,
+  fetchInferenceStats,
+  rowToNode,
+  rowToEdge,
+  rowToFullMemory,
+  snippet,
+  UUID_RE,
+  PROJECT_RE,
+  MAX_NODES_PER_PROJECT,
+  MAX_DEPTH,
+  DEFAULT_DEPTH,
+};

package/packages/server/src/index.js

@@ -64,6 +64,7 @@ const { themes, statusColors } = require('./themes');
 const { loadConfig, addProject, updateConfig } = require('./config');
 const { createAuthMiddleware, verifyWebSocketUpgrade, hasAuth } = require('./auth');
 const { createSprintRoutes } = require('./sprint-routes');
+const { createGraphRoutes } = require('./graph-routes');
 const orchestrationPreview = require('./orchestration-preview');
 
 // Sprint 37 T3 — lazy resolution of T2's CLI modules. The orchestration-preview
@@ -170,6 +171,17 @@ function createServer(config) {
   const mnestraBridge = createBridge(config);
   console.log(`[mnestra-bridge] mode=${mnestraBridge.mode}`);
 
+  // Sprint 38 / T3 — let RAGIntegration delegate vector recall to the
+  // bridge so we don't duplicate the embed pipeline. Graph recall stays
+  // in rag.js because it's a different RPC and doesn't share the
+  // direct/webhook/mcp mode shape.
+  rag.setBridge(mnestraBridge);
+  if (rag.graphRecall) {
+    console.log(
+      `[rag] graph-aware recall ENABLED (depth=${rag.graphRecallDepth}, k=${rag.graphRecallK}, half-life=${rag.graphRecallRecencyHalflifeDays}d)`
+    );
+  }
+
   // Initialize transcript writer (Session Transcripts — Sprint 6)
   const transcriptConfig = config.transcripts || {};
   const transcriptEnabled = transcriptConfig.enabled !== undefined
@@ -902,6 +914,15 @@ function createServer(config) {
     getSession: (id) => sessions.get(id),
   });
 
+  // Graph endpoints (Sprint 38 T4) — knowledge-graph view backing graph.html.
+  // Reuses the petvetbid pg pool (same DATABASE_URL serves memory_items +
+  // memory_relationships alongside rumen_*). Graceful-degrades when the pool
+  // is absent.
+  createGraphRoutes({
+    app,
+    getPool: getRumenPool,
+  });
+
   // GET /api/sessions/:id - get session details
   app.get('/api/sessions/:id', (req, res) => {
     const session = sessions.get(req.params.id);

package/packages/server/src/preflight.js

@@ -136,6 +136,82 @@ async function checkDatabase() {
   }
 }
 
+// Sprint 38 / T3 — graph-health check. Returns:
+//   pass : memory_relationships has rows AND last inferred_at < 48h ago
+//   warn : has rows but last inference > 48h ago (T2 cron may have drifted)
+//   fail : pg unreachable, table missing, or zero edges
+//
+// Reads `inferred_at` (T1's migration 009 column). Falls back to `created_at`
+// for the 749 pre-T2 edges that have no inferred_at value yet, so the check
+// doesn't perma-warn on the substrate that already exists.
+async function checkGraphHealth(config) {
+  // Only meaningful when graph features are enabled. Treat as pass with a
+  // descriptive detail so the banner doesn't FAIL on installs that haven't
+  // opted into graph recall yet.
+  const graphEnabled = config.rag?.graphRecall === true;
+  if (!graphEnabled) {
+    return { name: 'graph_health', passed: true, detail: 'graph recall disabled' };
+  }
+
+  const dbUrl = process.env.DATABASE_URL;
+  if (!dbUrl) {
+    return { name: 'graph_health', passed: false, detail: 'DATABASE_URL not set — cannot check graph' };
+  }
+
+  let pg;
+  try { pg = require('pg'); } catch (err) { pg = null; }
+  if (!pg) {
+    return { name: 'graph_health', passed: false, detail: 'pg module not installed' };
+  }
+
+  const pool = new pg.Pool({
+    connectionString: dbUrl,
+    max: 1,
+    connectionTimeoutMillis: 5000,
+  });
+
+  try {
+    // Single round-trip: edge count + last inference timestamp. coalesce on
+    // inferred_at so the substrate's pre-T2 edges register their created_at
+    // (otherwise max() returns NULL and the staleness check trips).
+    const res = await pool.query(
+      `SELECT
+         count(*)::int AS edges,
+         max(coalesce(inferred_at, created_at)) AS last_inferred_at
+       FROM memory_relationships`
+    );
+    const row = res.rows[0] || {};
+    const edges = Number(row.edges || 0);
+    if (edges === 0) {
+      return {
+        name: 'graph_health', passed: false,
+        detail: 'memory_relationships is empty — run T2 inference cron or seed edges manually',
+      };
+    }
+
+    const last = row.last_inferred_at ? new Date(row.last_inferred_at) : null;
+    if (!last) {
+      return {
+        name: 'graph_health', passed: true,
+        detail: `${edges.toLocaleString()} edges, last inference timestamp unknown`,
+      };
+    }
+
+    const agoMs = Date.now() - last.getTime();
+    const agoH = (agoMs / 3_600_000).toFixed(1);
+    const stale = agoMs > 48 * 3_600_000; // 48h cron drift threshold
+    return {
+      name: 'graph_health',
+      passed: !stale,
+      detail: stale
+        ? `${edges.toLocaleString()} edges, last inference ${agoH}h ago (stale — expected within 48h)`
+        : `${edges.toLocaleString()} edges, last inference ${agoH}h ago`,
+    };
+  } finally {
+    await pool.end().catch(() => {});
+  }
+}
+
 async function checkProjectPaths(config) {
   const projects = config.projects || {};
   const names = Object.keys(projects);
@@ -282,6 +358,10 @@ async function runPreflight(config) {
       name: 'shell_sanity', passed: false,
       detail: `check failed — ${err.message}`,
     })),
+    checkGraphHealth(config).catch((err) => ({
+      name: 'graph_health', passed: false,
+      detail: `check failed — ${err.message}`,
+    })),
   ]);
 
   const result = {
@@ -330,6 +410,7 @@ const REMEDIATION = {
   database_url: 'Set DATABASE_URL in ~/.termdeck/secrets.env',
   project_paths: 'Fix paths in ~/.termdeck/config.yaml → projects',
   shell_sanity: 'Check $SHELL and your login profile (~/.zshrc or ~/.bashrc)',
+  graph_health: 'Run T2 inference cron or apply migrations 009/010 to populate edges',
 };
 
 const CHECK_LABELS = {
@@ -339,6 +420,7 @@ const CHECK_LABELS = {
   database_url: 'Database',
   project_paths: 'Project paths',
   shell_sanity: 'Shell',
+  graph_health: 'Graph',
 };
 
 function printHealthBanner(result) {