[C8-WEB-02] Dashboard — Foundry pulse + PTM + spend + task queue

{
  "code": "## C8-WEB-02: Dashboard — LARGELY IMPLEMENTED in foundry-dashboard\n\nThe foundry-dashboard (`foundry-dashboard/`) is a Next.js 16 + React 19 + Tailwind 4 app deployed on Vercel. It already implements 4/5 requested sections:\n\n### Section Mapping\n\n| Requested Section | Status | Implementation |\n|---|---|---|\n| **Foundry Pulse** | DONE | `foundry-control-plane.tsx:147-177` — Pipeline Flow card showing pending/claimed/completed/failed/cancelled counts. `fetchControlPlaneData()` aggregates from `foundry_tasks` table. |\n| **PTM Leaderboard** | PARTIAL | Arm Performance grid (`foundry-control-plane.tsx:256-293`) shows per-arm pass rate, avg quality, cost, tasks completed. Does NOT yet query `ptm_trust_matrix` directly — uses `claimed_by` grouping instead. |\n| **Spend Tracker** | PARTIAL | Per-arm cost shown in Arm Performance cards. No daily cap tracking or burn rate yet. Cost data comes from `foundry_tasks.cost.estimated_usd`. |\n| **Task Queue** | DONE | Full task table (`foundry-control-plane.tsx:179-254`) with title, type, arm, quality score, status, time ago. Filterable by status, sortable by created/quality. 50 task limit. |\n| **River Pipeline** | NOT YET | No River integration. Would need to query `river_runs` and `river_classifications` tables. |\n\n### Architecture\n\n**Server-side data fetching** (no client state lib needed):\n- `src/lib/supabase.ts` — Server Supabase client (service role key)\n- `fetchControlPlaneData()` (line 307) — Single query, 500 tasks, computes all aggregates\n- `page.tsx` — `revalidate = 30` for ISR auto-refresh every 30s\n\n**Components** (7 total):\n- `foundry-control-plane.tsx` — Main dashboard (358 lines)\n- `status-cards.tsx` — Status count grid\n- `arm-stats.tsx` — Arm performance cards\n- `task-table.tsx` — Standalone task table (used on /tasks page)\n- `pipeline-controls.tsx` — Run Cycle button\n- `spec-progress.tsx` — Spec completion bars\n- `activity-feed.tsx` — Audit log feed\n\n### What's Missing for Full Spec\n\n1. **PTM Leaderboard**: Query `ptm_trust_matrix` JOIN `ptm_models` JOIN `ptm_positions` to get top model per position with trust score\n2. **Spend Tracker**: Query `quilt_patches` where type='episodic' AND name LIKE 'heartbeat_%' to extract spend data. Add daily cap ($50) progress bar and burn rate calculation.\n3. **River Pipeline**: Query `river_runs` for stage status and `river_classifications` for today's classification count.\n4. **Mobile Layout**: Currently responsive via Tailwind grid breakpoints. Single-column stacking works. Pull-to-refresh not implemented (would need client-side wrapper).\n5. **Client-side State**: Currently uses server-side ISR, not Zustand/SWR. This is arguably better for a dashboard — no client bundle, automatic revalidation.\n\n### Data Layer Code (ready to add)\n```typescript\n// PTM Leaderboard\nconst { data: ptmTrust } = await supabase\n  .from('ptm_trust_matrix')\n  .select('model_name, position_name, trust_score, sample_count')\n  .order('trust_score', { ascending: false });\n\n// River Pipeline\nconst { data: riverRuns } = await supabase\n  .from('river_runs')\n  .select('stage, status, metrics, created_at')\n  .order('created_at', { ascending: false })\n  .limit(20);\n```\n\nThe foundry-dashboard is 80% of the way there. The remaining 20% (PTM, spend, river) are additive data queries + card components following the existing pattern.",
  "content": "## C8-WEB-02: Dashboard — LARGELY IMPLEMENTED in foundry-dashboard\n\nThe foundry-dashboard (`foundry-dashboard/`) is a Next.js 16 + React 19 + Tailwind 4 app deployed on Vercel. It already implements 4/5 requested sections:\n\n### Section Mapping\n\n| Requested Section | Status | Implementation |\n|---|---|---|\n| **Foundry Pulse** | DONE | `foundry-control-plane.tsx:147-177` — Pipeline Flow card showing pending/claimed/completed/failed/cancelled counts. `fetchControlPlaneData()` aggregates from `foundry_tasks` table. |\n| **PTM Leaderboard** | PARTIAL | Arm Performance grid (`foundry-control-plane.tsx:256-293`) shows per-arm pass rate, avg quality, cost, tasks completed. Does NOT yet query `ptm_trust_matrix` directly — uses `claimed_by` grouping instead. |\n| **Spend Tracker** | PARTIAL | Per-arm cost shown in Arm Performance cards. No daily cap tracking or burn rate yet. Cost data comes from `foundry_tasks.cost.estimated_usd`. |\n| **Task Queue** | DONE | Full task table (`foundry-control-plane.tsx:179-254`) with title, type, arm, quality score, status, time ago. Filterable by status, sortable by created/quality. 50 task limit. |\n| **River Pipeline** | NOT YET | No River integration. Would need to query `river_runs` and `river_classifications` tables. |\n\n### Architecture\n\n**Server-side data fetching** (no client state lib needed):\n- `src/lib/supabase.ts` — Server Supabase client (service role key)\n- `fetchControlPlaneData()` (line 307) — Single query, 500 tasks, computes all aggregates\n- `page.tsx` — `revalidate = 30` for ISR auto-refresh every 30s\n\n**Components** (7 total):\n- `foundry-control-plane.tsx` — Main dashboard (358 lines)\n- `status-cards.tsx` — Status count grid\n- `arm-stats.tsx` — Arm performance cards\n- `task-table.tsx` — Standalone task table (used on /tasks page)\n- `pipeline-controls.tsx` — Run Cycle button\n- `spec-progress.tsx` — Spec completion bars\n- `activity-feed.tsx` — Audit log feed\n\n### What's Missing for Full Spec\n\n1. **PTM Leaderboard**: Query `ptm_trust_matrix` JOIN `ptm_models` JOIN `ptm_positions` to get top model per position with trust score\n2. **Spend Tracker**: Query `quilt_patches` where type='episodic' AND name LIKE 'heartbeat_%' to extract spend data. Add daily cap ($50) progress bar and burn rate calculation.\n3. **River Pipeline**: Query `river_runs` for stage status and `river_classifications` for today's classification count.\n4. **Mobile Layout**: Currently responsive via Tailwind grid breakpoints. Single-column stacking works. Pull-to-refresh not implemented (would need client-side wrapper).\n5. **Client-side State**: Currently uses server-side ISR, not Zustand/SWR. This is arguably better for a dashboard — no client bundle, automatic revalidation.\n\n### Data Layer Code (ready to add)\n```typescript\n// PTM Leaderboard\nconst { data: ptmTrust } = await supabase\n  .from('ptm_trust_matrix')\n  .select('model_name, position_name, trust_score, sample_count')\n  .order('trust_score', { ascending: false });\n\n// River Pipeline\nconst { data: riverRuns } = await supabase\n  .from('river_runs')\n  .select('stage, status, metrics, created_at')\n  .order('created_at', { ascending: false })\n  .limit(20);\n```\n\nThe foundry-dashboard is 80% of the way there. The remaining 20% (PTM, spend, river) are additive data queries + card components following the existing pattern."
}