feat(growth): add inspectable company research pilot

.github/workflows/ci.yml

@@ -42,6 +42,7 @@ jobs:
       posthog: ${{ steps.scope.outputs.posthog }}
       scripts_tests: ${{ steps.scope.outputs.scripts_tests }}
       growth_lifecycle: ${{ steps.scope.outputs.growth_lifecycle }}
+      growth_research: ${{ steps.scope.outputs.growth_research }}
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         with:
@@ -271,6 +272,24 @@ jobs:
       - run: npx nx run lifecycle:check
       - run: npx nx build lifecycle
 
+  growth-research:
+    name: Growth Research — Node 24
+    needs: ci-scope
+    if: github.event_name == 'push' || needs.ci-scope.outputs.growth_research == 'true'
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+      - uses: actions/setup-node@53b83947a5a98c8d113130e565377fae1a50d02f # v6.3.0
+        with:
+          node-version: 24
+          cache: npm
+      - run: npm ci --ignore-scripts
+      - run: npx nx lint growth-research
+      - run: npx nx test growth-research
+      - run: npx nx check growth-research
+      - run: npx nx build growth-research
+
   cockpit:
     name: Cockpit — build / test
     needs: ci-scope
@@ -868,6 +887,7 @@ jobs:
       - scripts-tests
       - growth-lifecycle
       - lifecycle
+      - growth-research
     # `CI — required` is the only required status context. A merge queue
     # waits on it for each candidate, so it must report on merge_group too —
     # otherwise every queued merge blocks forever on a check that never runs.
@@ -895,6 +915,7 @@ jobs:
           RESULT_SCRIPTS_TESTS: ${{ needs.scripts-tests.result }}
           RESULT_GROWTH_LIFECYCLE: ${{ needs.growth-lifecycle.result }}
           RESULT_LIFECYCLE: ${{ needs.lifecycle.result }}
+          RESULT_GROWTH_RESEARCH: ${{ needs.growth-research.result }}
           SCOPE_LIBRARY: ${{ needs.ci-scope.outputs.library }}
           SCOPE_ANGULAR_COMPATIBILITY: ${{ needs.ci-scope.outputs.angular_compatibility }}
           SCOPE_WEBSITE: ${{ needs.ci-scope.outputs.website }}
@@ -909,6 +930,7 @@ jobs:
           SCOPE_POSTHOG: ${{ needs.ci-scope.outputs.posthog }}
           SCOPE_SCRIPTS_TESTS: ${{ needs.ci-scope.outputs.scripts_tests }}
           SCOPE_GROWTH_LIFECYCLE: ${{ needs.ci-scope.outputs.growth_lifecycle }}
+          SCOPE_GROWTH_RESEARCH: ${{ needs.ci-scope.outputs.growth_research }}
           # The preview lanes need repository secrets, so they skip on fork
           # PRs. Their scope keys are computed from changed files alone, so on
           # a fork they can be in scope yet legitimately skipped. This mirrors
@@ -998,6 +1020,7 @@ jobs:
           require_scoped "scripts_tests" "Scripts — generator / proxy vitest suites" "$RESULT_SCRIPTS_TESTS" "$SCOPE_SCRIPTS_TESTS"
           require_scoped "growth_lifecycle" "Growth lifecycle — Node 22" "$RESULT_GROWTH_LIFECYCLE" "$SCOPE_GROWTH_LIFECYCLE"
           require_scoped "growth_lifecycle" "Lifecycle — Node 24" "$RESULT_LIFECYCLE" "$SCOPE_GROWTH_LIFECYCLE"
+          require_scoped "growth_research" "Growth Research — Node 24" "$RESULT_GROWTH_RESEARCH" "$SCOPE_GROWTH_RESEARCH"
 
           if [[ "$failed" -ne 0 ]]; then
             exit 1

apps/growth-research/README.md

@@ -1,4 +1,101 @@
-# Growth research compatibility application
+# Growth research application
+
+## Local company research pilot
+
+The local pilot compares one bounded Dawn agent with the existing lifecycle enrichment
+generator on identical captured company evidence. It has no Growth database connection,
+does not resolve people or employment, and cannot send email. The managed deployment
+still exposes only the synthetic compatibility graph documented below. Pilot routes,
+operator adapters, and their generated graph are excluded from its staged artifact.
+
+Use Node 24 and the existing workspace dependencies. Build before running the agent:
+
+```sh
+npx nx build growth-research
+npx tsx apps/growth-research/scripts/research-pilot.mts synthetic --output /absolute/private/pilot
+npx tsx apps/growth-research/scripts/research-pilot.mts acquire --output /absolute/private/pilot --domains threadplane.ai,dawnai.org,neon.tech,vercel.com,resend.com,langchain.com
+```
+
+These commands return UUIDs for immutable JSON files in the selected output directory.
+Acquisition records include complete, partial, empty and failed outcomes. The existing
+fetcher can skip unusable pages, so missing paths have an unknown reason; redirects may
+make the original path indeterminate. Review the captured corpus before model calls:
+remove personal biography/contact snippets, retain empty cases and failures, and fill
+expected claims/unknowns from the actual captured evidence. Save the reviewed corpus
+under a new name/version. Acquisition is preparation, not a human quality label.
+
+Set `GROWTH_RESEARCH_PILOT_MODE=local-company-only` and configure `OPENAI_API_KEY`
+for the agent or `ANTHROPIC_API_KEY` for the baseline through the operator environment.
+Never include keys in arguments, fixtures, reports or commits. The local in-process
+case context is also required: an environment flag alone cannot authorize pilot tools.
+
+```sh
+npx tsx apps/growth-research/scripts/research-pilot.mts run --output /absolute/private/pilot --corpus /absolute/private/pilot/CORPUS_UUID.json --approach agent
+npx tsx apps/growth-research/scripts/research-pilot.mts run --output /absolute/private/pilot --corpus /absolute/private/pilot/CORPUS_UUID.json --approach baseline
+npx tsx apps/growth-research/scripts/research-pilot.mts inspect --output /absolute/private/pilot --run RUN_UUID
+```
+
+Each case/approach/repetition has a separate run ID, deadline, budget and terminal
+record. Runs execute sequentially. The agent permits six provider requests, six evidence
+reads, 1,024 output tokens per request, no provider retries, a 20-second request timeout,
+and a 90-second run deadline. It can use the evidence skill and plan, read only captured
+case evidence, and submit a candidate. It cannot delegate, use memory, fetch URLs or
+read arbitrary files. Candidate acceptance is structural validation, not a truth label.
+Explicit inspection shows company sources and candidate findings; ordinary progress
+prints only opaque IDs and outcome codes. Reports use restrictive atomic writes and
+refuse overwrites. Preserve the final index and all failed attempts when comparing runs.
+
+The baseline uses its existing provider/model and 1,200-token/30-second request bounds.
+It receives company mode, synthetic adapter form context and zero progress score.
+Its raw citations are captured before production normalization. It does not return
+quotes: `not_provided` is distinct from failing or passing exact-quote validation.
+Provider failure records retain known request/usage/citation diagnostics. Missing usage
+and cost are unavailable, never zero. This comparison measures whole approaches with
+different providers/models; it does not isolate Dawn's causal contribution.
+
+Raw automatic tracing is disabled for local pilot runs. The record reports
+`tracing: unavailable`; this slice does not claim sanitized LangSmith tracing is live.
+No research findings are automatically published to Growth or typed memory.
+
+### Human comparison
+
+Each invocation emits a blinded review packet. To combine baseline and agent results
+for the same corpus, pass their index UUIDs; mixed corpus hashes/classes are rejected:
+
+```sh
+npx tsx apps/growth-research/scripts/research-pilot.mts review --output /absolute/private/pilot --indices BASELINE_INDEX_UUID,AGENT_INDEX_UUID
+npx tsx apps/growth-research/scripts/research-pilot.mts score --output /absolute/private/pilot --packet PACKET_UUID --labels /absolute/private/pilot/human-labels.json
+```
+
+The packet omits model and approach labels. Reviewers inspect each claim and profile
+against the captured sources, including failed cases. `human-labels.json` is an array:
+
+```json
+[{"reviewId":"RUN_UUID","supportedClaims":0,"reviewedClaims":0,"supportedFields":0,"applicableFields":0,"correctAbstentions":3,"applicableAbstentions":3,"contradictionsMissed":0}]
+```
+
+Use actual UUIDs and counts for each case. Reviewed claim count must match the packet;
+applicable fields and abstentions come from its expected unknowns. Imported labels and
+per-approach scores are persisted as a new review artifact. Aggregate quality scores
+remain unavailable while reviews are incomplete, preventing success-only denominators.
+Human semantic review is not replaced by model grading or string matching.
+
+### Dogfooding findings ledger
+
+| Finding | Evidence / owning layer | Status and next verification |
+| --- | --- | --- |
+| Nullable tool fields become required strings | Dawn 0.8.24 compiler JSON schema conversion; observed generated submit schema and failed unknown-field submissions | Upstream core and LangChain conversion regression/fix in progress. Pilot uses the supported authored Zod schema export; a package upgrade must rerun the original extraction probe before declaring the upstream defect released. |
+| Bound model calls bypass subclass generation hooks | Real bound-model regression in this application | Guards, request counts and JSON usage capture live at the actual provider fetch boundary; generated graph tests verify it. |
+| Page capture yields empty, partial, or mostly navigation evidence | Company-only acquisition against the six documented domains | Outcomes retained. Evaluate extraction improvements separately; do not hide failures by swapping cases. |
+| Baseline provider rejects billing state | Live baseline synthetic calls returned a classified billing rejection | External provider funding/configuration required; no quality comparison can be claimed from failed calls. |
+| Managed interruption precedes later child checkpoint | Recorded local/cloud Agent Server 0.13.4-node24 probe | Still a live-person integration gate; local cancellation tests are not proof of managed cancellation. |
+| Disabled memory and shared harness persistence behavior | Earlier synthetic compatibility probe on Dawn 0.8.24 | Reproduction-needed against current Dawn before assigning a fix. Pilot has no memory and graph tests use isolated state. |
+
+Keep source snapshots, generated reports and review labels outside git. The full growth
+funnel/contact journey and real install/runtime-triggered enrichment are subsequent
+slices, after supported company context and the managed data lifecycle are verified.
+
+## Synthetic compatibility deployment
 
 Private synthetic Dawn application, separate from lifecycle and the Python cockpit.
 Published Dawn packages are pinned to `0.8.24`; the app and deployment require Node 24.

apps/growth-research/dawn.config.ts

@@ -5,7 +5,7 @@ import { candidateMemoryStore, syntheticEmbedder, trustedFixtureScope } from './
 export default {
   appDir: 'src/app',
   build: { targets: ['langsmith'] },
-  toolOutput: { noOffloadTools: ['readFixture', 'coordinatorSummary', 'readSkill', 'writeTodos', 'recall', 'remember'] },
+  toolOutput: { noOffloadTools: ['readFixture', 'coordinatorSummary', 'readSkill', 'writeTodos', 'recall', 'remember', 'readEvidence', 'submitCandidate'] },
   summarization: { enabled: false },
   memory: {
     store: candidateMemoryStore,

apps/growth-research/eslint.config.mjs

@@ -3,4 +3,13 @@ import baseConfig from '../../eslint.config.mjs';
 export default [
   { ignores: ['**/.dawn/**', '**/.deployment/**'] },
   ...baseConfig,
+  {
+    // Local-only benchmark adapters exercise the exact lifecycle baseline.
+    // They are excluded from the standalone deployment; copying it would bias comparisons.
+    files: [
+      'apps/growth-research/src/pilot/baseline.ts',
+      'apps/growth-research/src/pilot/acquisition.ts',
+    ],
+    rules: { '@nx/enforce-module-boundaries': 'off' },
+  },
 ];

apps/growth-research/project.json

@@ -3,8 +3,21 @@
   "$schema": "../../node_modules/nx/schemas/project-schema.json",
   "sourceRoot": "apps/growth-research/src",
   "projectType": "application",
-  "tags": ["scope:internal", "scope:growth-research", "type:app", "runtime:node24"],
+  "tags": [
+    "scope:internal",
+    "scope:growth-research",
+    "type:app",
+    "runtime:node24"
+  ],
   "targets": {
+    "pilot": {
+      "executor": "nx:run-commands",
+      "cache": false,
+      "options": {
+        "command": "tsx apps/growth-research/scripts/research-pilot.mts",
+        "forwardAllArgs": true
+      }
+    },
     "smoke-langsmith": {
       "executor": "nx:run-commands",
       "cache": false,
@@ -17,7 +30,9 @@
     "test-memory-integration": {
       "executor": "nx:run-commands",
       "cache": false,
-      "options": { "command": "npx vitest run --config apps/growth-research/vitest.memory-integration.config.ts --reporter=verbose" }
+      "options": {
+        "command": "npx vitest run --config apps/growth-research/vitest.memory-integration.config.ts --reporter=verbose"
+      }
     },
     "test": {
       "executor": "@nx/vitest:test",
@@ -28,7 +43,10 @@
       "cache": false,
       "options": {
         "cwd": "apps/growth-research",
-        "commands": ["node scripts/dawn-cli.mts check", "node ../../node_modules/typescript/bin/tsc --noEmit -p tsconfig.json"],
+        "commands": [
+          "node scripts/dawn-cli.mts check",
+          "node ../../node_modules/typescript/bin/tsc --noEmit -p tsconfig.json"
+        ],
         "parallel": false
       }
     },
@@ -38,7 +56,11 @@
       "outputs": ["{projectRoot}/.deployment"],
       "options": {
         "cwd": "apps/growth-research",
-        "commands": ["node scripts/dawn-cli.mts build --clean", "node scripts/package-langsmith.mts", "node scripts/verify-langsmith-artifact.mts"],
+        "commands": [
+          "node scripts/dawn-cli.mts build --clean",
+          "node scripts/package-langsmith.mts",
+          "node scripts/verify-langsmith-artifact.mts"
+        ],
         "parallel": false
       }
     },

apps/growth-research/scripts/package-langsmith.mts

@@ -43,6 +43,9 @@ async function contained(root: string, path: string): Promise<void> {
 async function copySource(root: string, path: string, output: string): Promise<void> {
   await contained(root, path);
   const name = basename(path);
+  const local = relative(root, path);
+  if (local === 'src/app/enrichment/company-pilot') return;
+  if (local.startsWith('src/pilot/') && !['context.ts', 'contracts.ts', 'validation.ts'].includes(name)) return;
   if (name.startsWith('.') || name === 'node_modules' || /\.(spec|test)\.[cm]?ts$/.test(name)) return;
   if ((await lstat(path)).isDirectory()) {
     await mkdir(output, { recursive: true });
@@ -116,7 +119,9 @@ export async function stageLangSmith(appRoot: string): Promise<string> {
   const config = await readObject(join(root, '.dawn/build/langgraph.json'));
   const generatedGraphs = object(config['graphs'], 'graphs');
   const specialistId = '/enrichment/research/subagents/researcher#agent';
-  if (Object.keys(generatedGraphs).some(key => key !== graphId && key !== specialistId)) throw new Error('Unexpected generated graph');
+  const pilotId = '/enrichment/company-pilot#agent';
+  if (Object.keys(generatedGraphs).some(key => key !== graphId && key !== specialistId && key !== pilotId)) throw new Error('Unexpected generated graph');
+  if (pilotId in generatedGraphs && generatedGraphs[pilotId] !== './.dawn/build/enrichment-company-pilot.ts:graph') throw new Error('Unexpected pilot graph');
   if (specialistId in generatedGraphs) {
     if (generatedGraphs[specialistId] !== './.dawn/build/enrichment-research-subagents-researcher.ts:graph') throw new Error('Unexpected specialist graph entry');
     await validateReference(root, generatedGraphs[specialistId], 'specialist graph');
@@ -141,6 +146,7 @@ export async function stageLangSmith(appRoot: string): Promise<string> {
   await copyFile(join(root, 'dawn.config.ts'), join(output, 'dawn.config.ts'));
   const copySchemas = async (path: string, target: string): Promise<void> => {
     await contained(root, path);
+    if (['.dawn/routes/enrichment/company-pilot', '.dawn/routes/enrichment-company-pilot'].includes(relative(root, path))) return;
     if ((await lstat(path)).isDirectory()) {
       await mkdir(target, { recursive: true });
       for (const name of await readdir(path)) await copySchemas(join(path, name), join(target, name));
@@ -151,6 +157,7 @@ export async function stageLangSmith(appRoot: string): Promise<string> {
   };
   await copySchemas(join(root, '.dawn/routes'), join(output, '.dawn/routes'));
   for (const name of await readdir(join(root, '.dawn/build'))) {
+    if (name === 'enrichment-company-pilot.ts') continue;
     if (!name.endsWith('.ts')) continue;
     await contained(root, join(root, '.dawn/build', name));
     await copyFile(join(root, '.dawn/build', name), join(output, '.dawn/build', name));