Merge remote-tracking branch 'gitea/main' into feat/revenue-intelligence

packages/db/src/seed/demo.ts

@@ -41,6 +41,7 @@ import { createDatabase } from '../client';
 import {
   accounts,
   activities,
+  facts,
   allocations,
   capacityCommitments,
   capacityRequests,
@@ -660,7 +661,166 @@ async function seedDemo() {
     }
   }
 
+  // -------------------------------------------------- agent-derived facts
+  //
+  // Without these the fact-review queue and every provenance tooltip are
+  // empty, which hides the thing that makes an agent-written CRM trustworthy:
+  // that each claim carries a score, a band, evidence and a source, and that
+  // only verified claims apply themselves.
+  //
+  // The mix is deliberate. Two `applied` facts show what a confident agent
+  // writes unprompted; four `proposed` show what waits for a human; one is a
+  // near-miss that a reviewer should reject, so the queue is not a row of
+  // obvious approvals.
+  const factSeeds: {
+    accountDomain?: string;
+    contactName?: string;
+    field: string;
+    value: string;
+    score: string;
+    band: 'verified' | 'probable' | 'possible';
+    status: 'applied' | 'proposed';
+    method: string;
+    sourceUrl?: string;
+    evidence: Record<string, unknown>;
+  }[] = [
+    {
+      accountDomain: 'coreweave.com',
+      field: 'supplierType',
+      value: 'neocloud',
+      score: '0.960',
+      band: 'verified',
+      status: 'applied',
+      method: 'web_search',
+      sourceUrl: 'https://www.coreweave.com/',
+      evidence: {
+        quote: 'Describes itself as an AI hyperscaler providing GPU cloud infrastructure.',
+        corroboration: 2,
+      },
+    },
+    {
+      accountDomain: 'nebius.com',
+      field: 'jurisdiction',
+      value: 'European Union',
+      score: '0.910',
+      band: 'verified',
+      status: 'applied',
+      method: 'web_search',
+      sourceUrl: 'https://nebius.com/',
+      evidence: {
+        quote: 'Operates a datacentre in Finland, inside the EU data-residency perimeter.',
+        matters: 'Determines eligibility for customers with EU residency requirements.',
+      },
+    },
+    {
+      accountDomain: 'crusoe.ai',
+      field: 'certifications',
+      value: 'SOC 2 Type II',
+      score: '0.720',
+      band: 'probable',
+      status: 'proposed',
+      method: 'web_search',
+      sourceUrl: 'https://crusoe.ai/',
+      evidence: {
+        quote: 'A trust page references SOC 2, but the report scope and observation window are not stated.',
+        caution: 'Scope matters — a report can cover only some products.',
+      },
+    },
+    {
+      accountDomain: 'lambda.ai',
+      field: 'supplierType',
+      value: 'neocloud',
+      score: '0.680',
+      band: 'probable',
+      status: 'proposed',
+      method: 'web_search',
+      sourceUrl: 'https://lambda.ai/',
+      evidence: { quote: 'Markets GPU cloud and on-premises clusters.' },
+    },
+    {
+      contactName: 'Dana Whitfield',
+      field: 'title',
+      value: 'VP Infrastructure',
+      score: '0.540',
+      band: 'possible',
+      status: 'proposed',
+      method: 'inference',
+      evidence: {
+        reasoning: 'A conference bio lists a VP title; the CRM records Head of Infrastructure.',
+        conflict: 'Sources disagree, and neither is dated.',
+      },
+    },
+    {
+      accountDomain: 'runpod.io',
+      field: 'customerSegment',
+      value: 'frontier_lab',
+      score: '0.310',
+      band: 'possible',
+      status: 'proposed',
+      method: 'inference',
+      evidence: {
+        reasoning: 'Inferred from a blog post mentioning large training runs.',
+        warning:
+          'Weak. This is a supply-side provider, not a frontier lab — a reviewer should reject it.',
+      },
+    },
+  ];
+
+  let factsAdded = 0;
+  for (const seed of factSeeds) {
+    let accountId: string | undefined;
+    let contactId: string | undefined;
+
+    if (seed.accountDomain) {
+      const [row] = await db
+        .select({ id: accounts.id })
+        .from(accounts)
+        .where(eq(accounts.domain, seed.accountDomain))
+        .limit(1);
+      accountId = row?.id;
+    }
+    if (seed.contactName) {
+      const [row] = await db
+        .select({ id: contacts.id })
+        .from(contacts)
+        .where(eq(contacts.fullName, seed.contactName))
+        .limit(1);
+      contactId = row?.id;
+    }
+    if (!accountId && !contactId) continue;
+
+    // Idempotent on the natural key: one claim per subject per field per value.
+    const [existing] = await db
+      .select({ id: facts.id })
+      .from(facts)
+      .where(
+        and(
+          accountId ? eq(facts.accountId, accountId) : eq(facts.contactId, contactId!),
+          eq(facts.field, seed.field),
+          eq(facts.value, seed.value),
+        ),
+      )
+      .limit(1);
+    if (existing) continue;
+
+    await db.insert(facts).values({
+      accountId,
+      contactId,
+      field: seed.field,
+      value: seed.value,
+      score: seed.score,
+      band: seed.band,
+      status: seed.status,
+      method: seed.method,
+      sourceUrl: seed.sourceUrl,
+      evidence: seed.evidence,
+      observedAt: at(-Math.round(Math.random() * 6) - 1),
+    });
+    factsAdded += 1;
+  }
+
   console.log('  4 capacity commitments, with sites, MSAs and negotiated SLAs');
+  console.log(`  ${factSeeds.length} agent-derived facts (${factsAdded} new) — 2 applied, 4 awaiting review`);
   console.log('  6 demand deals across the pipeline, 5 supply deals');
   console.log('  Allocations including one unconverted hold and internal research burn');
   console.log('\nEverything is prefixed "DEMO — ". Remove it with: npm run db:demo -- --clear');