feat: activate Drops Agent V3 evidence

[svg8bit]

Summary

• activate the immutable V2 benchmark baseline from the canonical 120-case corpus
• attach the committed failure-clustering artifact and the verified 10-case Design Agent slice
• measure two current Vercel AI Gateway models with bounded privacy-safe probes
• persist only hashes, token/cost/latency metadata, and aggregate receipts in private immutable storage
• drive the internal dashboard and promotion gate from stored evidence instead of placeholder state

Validation

• npm audit --omit=dev --audit-level=high
• npm run guardrails:ui
• npm run lint
• npm run typecheck
• npm run build:vercel
• npm run test:unit
• npm run build-storybook
• npm run test:storybook
• npm run test:storybook:visual
• focused agent evidence tests: 11 passed
• focused fresh-server Playwright smoke: 1 passed

The first broad local E2E pass reused a stale 14:14 Next server after a 15:51 rebuild, causing missing-chunk 500s. After terminating that test-only process, a fresh-server browser check passed. No production runtime was touched by that cleanup.

CodeRabbit reviewed this change; three findings were addressed. One suggestion to expose raw storage SDK errors was intentionally rejected to preserve the existing secret/error-redaction boundary.

Summary by CodeRabbit

• New Features

  • Added release evidence activation with benchmark validation, failure analysis, design-agent results, and approved live-model checks.
  • Added evidence snapshots with privacy-conscious metadata and persistent storage.
  • Enhanced the evaluation dashboard to show evidence status, receipts, and expanded evidence coverage.
  • Added configuration options for enabling the compact agent core and legacy fallback.

• Bug Fixes

  • Improved handling and reporting when evidence storage is unavailable.

• Tests

  • Added coverage for activation success and failure, privacy protections, persistence, duplicate prevention, and snapshot validation.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
drops-studio	Building	Preview	Jul 31, 2026 4:02pm

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 9d35c7eb-a159-4bbb-98c1-90ba9325e675

📥 Commits

Reviewing files that changed from the base of the PR and between 8dd16e9 and c64de22.

📒 Files selected for processing (12)

• .env.example
• app/api/internal/agent-evals/run/route.ts
• app/api/internal/agent-evals/summary/route.ts
• app/internal/agent-evals/page.tsx
• components/agent-eval-dashboard.tsx
• lib/agent/evals/dashboard-evidence.ts
• lib/agent/evals/dashboard-types.ts
• lib/agent/evals/evidence-activation.ts
• lib/agent/evals/index.ts
• lib/agent/evals/store.ts
• lib/agent/evals/types.ts
• tests/agent-evidence-activation.test.mjs

---

📝 Walkthrough

Walkthrough

The PR adds V3 evidence activation with bounded live model probes, versioned snapshot storage, platform evidence receipts, and dashboard controls. Evaluation routes can activate and persist evidence, while pages and dashboards load snapshots and display their validation state.

Changes

Agent V3 evidence flow

Layer / File(s)	Summary
Evidence contracts and activation engine lib/agent/evals/types.ts, lib/agent/evals/evidence-activation.ts, lib/agent/evals/index.ts, tests/agent-evidence-activation.test.mjs	Adds versioned evidence snapshots, live model measurements, strict probe validation, baseline checks, model measurements, privacy guarantees, and activation tests.
Evidence snapshot persistence lib/agent/evals/store.ts, tests/agent-evidence-activation.test.mjs	Adds private snapshot storage with validation, size limits, write-once behavior, bounded listing, and persistence tests.
Route and platform evidence integration app/api/internal/agent-evals/run/route.ts, app/api/internal/agent-evals/summary/route.ts, app/internal/agent-evals/page.tsx, lib/agent/evals/dashboard-evidence.ts, lib/agent/evals/dashboard-types.ts	Adds evidence activation to evaluation runs and passes stored snapshots into platform evidence generation with receipts and storage-gate state.
Dashboard activation and operator configuration components/agent-eval-dashboard.tsx, .env.example	Updates the dashboard to run full evidence activation, display activation receipts, and document compact-core and release-evidence settings.

Estimated code review effort: 4 (Complex) | ~45 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Operator
  participant AgentEvalDashboard
  participant AgentEvalRunRoute
  participant activateAgentV3Evidence
  participant AIGateway
  participant AgentEvalStore
  Operator->>AgentEvalDashboard: start full evidence activation
  AgentEvalDashboard->>AgentEvalRunRoute: request release suite with activateEvidence
  AgentEvalRunRoute->>activateAgentV3Evidence: validate report and activate evidence
  activateAgentV3Evidence->>AIGateway: probe approved live models
  AIGateway-->>activateAgentV3Evidence: return structured results and usage
  activateAgentV3Evidence-->>AgentEvalRunRoute: return evidence snapshot
  AgentEvalRunRoute->>AgentEvalStore: persist snapshot
  AgentEvalRunRoute-->>AgentEvalDashboard: return activation status and snapshot ID

Loading

Possibly related PRs

• svg8bit/drops-studio#5: Both changes extend the V3 platform-evidence flow and its summary routes.
• svg8bit/drops-studio#18: Both changes update the evaluation page and dashboard platform-evidence flow.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/drops-agent-evidence-live

---

_{Comment @coderabbitai help to get the list of available commands.}