Feedback Loop Handoff

-- Feedback Loop Infrastructure (Migration 005) -- Run in Supabase SQL Editor -- 1. Add tester_role to clone_feedback ALTER TABLE clone_feedback ADD COLUMN IF NOT EXISTS tester_role TEXT DEFAULT 'team'; -- 2. Add target_block + consensus columns to improvement_items ALTER TABLE improvement_items ADD COLUMN IF NOT EXISTS target_block TEXT; ALTER TABLE improvement_items ADD COLUMN IF NOT EXISTS consensus_score NUMERIC DEFAULT 0; ALTER TABLE improvement_items ADD COLUMN IF NOT EXISTS flagged_by JSONB DEFAULT '[]'; ALTER TABLE improvement_items ADD COLUMN IF NOT EXISTS prompt_version TEXT; ALTER TABLE improvement_items ADD COLUMN IF NOT EXISTS extraction_version TEXT; -- 3. Backfill target_block from existing skill_target UPDATE improvement_items SET target_block = skill_target WHERE target_block IS NULL AND skill_target IS NOT NULL; -- 4. Performance indexes CREATE INDEX IF NOT EXISTS idx_improvement_items_run_status ON improvement_items(factory_run_id, status) WHERE factory_run_id IS NOT NULL; CREATE INDEX IF NOT EXISTS idx_improvement_items_expert_scenario ON improvement_items(expert_slug, source_scenario) WHERE source_scenario IS NOT NULL; CREATE INDEX IF NOT EXISTS idx_improvement_items_target_block ON improvement_items(target_block, status) WHERE target_block IS NOT NULL;

## Feedback Loop Integration — From Audit Session (Mar 30) ### What's Built (folio-saas, deployed on Vercel) The audit flow now creates structured `improvement_items` that need to flow into factory block re-runs. Here's what exists: **Pages:** - `align360.asapai.net/samuel-audit-v3` — Interactive expert audit (19 cards, live chat, voice+text feedback) - `align360.asapai.net/samuel-extraction-review` — Expert reviews 6 extraction JSONs (295 items across 9 cards) - `align360.asapai.net/improvement-items` — Triage dashboard with "Push to Factory" button - `align360.asapai.net/feedback-loop-architecture` — Full PRD for the corrections loop **API Endpoints (all live at folio-saas on Vercel):** - `POST /api/a360/audit-submit` — Saves audit, publishes report, emails team, auto-triggers analyze - `POST /api/a360/audit-analyze` — Classifies findings into improvement_items (voice_gap, framework_gap, tone_drift, etc.) - `POST /api/a360/extraction-review-submit` — Saves extraction review, creates improvement_items per flagged item - `GET /api/a360/improvement-items?expert_slug=samuel-ngu&status=pending` — Query items with filters - `PATCH /api/a360/improvement-items/{id}` — Update status, priority, expert_correction, target_block, factory_run_id **Database (Supabase):** - `improvement_items` table — columns: id, feedback_id, expert_slug, finding_type, skill_target, target_block, priority, source_scenario, source_text, expert_correction, status, consensus_score, flagged_by, factory_run_id, factory_block_id, prompt_version, extraction_version - `clone_feedback` table — audit submissions with chat_transcript, chat_feedback, tester_role - Every improvement_item has `target_block` set (maps to factory block IDs) ### What YOU Need to Build **`POST /api/corrections-ingest`** — receives grouped corrections from the improvement-items dashboard. The "Push to Factory" button on the dashboard calls: ``` POST factory.asapai.net/api/corrections-ingest Content-Type: application/json { "run_id": "uuid-of-factory-run", "corrections": [ { "block_id": "voice-extractor", "source": "improvement-items-dashboard", "instruction": "Apply 3 corrections from expert feedback", "items": [ { "finding": "Clone uses 'leverage' which Samuel never says", "correction": "Replace with 'use' or 'lean into'", "priority": "critical", "item_id": "uuid-of-improvement-item" } ] }, { "block_id": "soul-extractor", "source": "improvement-items-dashboard", "items": [ { "finding": "Soul extraction missed the 'second chance' narrative", "correction": "Core narrative should center on redemption through service", "priority": "critical", "item_id": "uuid" } ] } ] } ``` **Expected Response:** ```json { "success": true, "run_id": "uuid", "results": [ { "block_id": "voice-extractor", "items_count": 3, "status": "stored" }, { "block_id": "soul-extractor", "items_count": 1, "status": "stored" } ], "total_items": 4 } ``` **How it works:** 1. Store corrections in the block's state (`block.state.corrections[]`) 2. When block is re-run via `/api/execute`, inject corrections into LLM context as high-priority instructions 3. After block completes, clear corrections from state so they don't re-apply **Block ID mapping (these are the `target_block` values the dashboard sends):** | target_block | Factory Block | |---|---| | voice-extractor | Voice Extractor | | soul-extractor | Soul Extractor | | framework-extractor | Framework Extractor | | resource-extractor | Resource Extractor | | offer-extractor | Offer Extractor | | clone-compiler | Clone Compiler | | gap-analyzer | Gap Analyzer | | clone-tester | Clone Tester | **After building corrections-ingest:** - The dashboard button will work end-to-end (it already has the JS to call your endpoint) - Operator resets targeted blocks in factory GUI, blocks re-run with corrections injected - After re-run completes, dashboard can update items to "resolved" **Reference:** - Full PRD: align360.asapai.net/feedback-loop-architecture (11 sections) - Your own v5 handoff doc: memory/factory-v5-corrections-api.md (you wrote this — it aligns)

## Feedback Loop — FORGE Integration Brief (Mar 30) ### What Exists A complete feedback loop is being wired between three systems: 1. **Audit Pages** (folio-saas/Vercel) — Expert tests the clone, rates scenarios, provides corrections 2. **Improvement Items** (folio-saas API + Supabase) — Structured findings with target_block mapping 3. **Process Factory** (factory.asapai.net) — DAG engine that re-runs extraction/build blocks with corrections ### FORGE's Role: Monitor + Trigger FORGE should know about these endpoints for monitoring and automation: **Read improvement items:** ``` GET https://align360.asapai.net/api/a360/improvement-items?expert_slug=samuel-ngu&status=pending → Returns: { items: [...], total: N, summary: { by_status, by_block, by_priority } } ``` **Read improvement items by status:** ``` GET https://align360.asapai.net/api/a360/improvement-items?status=in_progress GET https://align360.asapai.net/api/a360/improvement-items?status=resolved GET https://align360.asapai.net/api/a360/improvement-items?target_block=voice-extractor ``` **Update an improvement item:** ``` PATCH https://align360.asapai.net/api/a360/improvement-items/{id} Body: { "status": "resolved", "resolved_in_version": "v7.2", "resolution_notes": "Fixed in re-run" } ``` **Push corrections to factory (once factory builds the endpoint):** ``` POST https://factory.asapai.net/api/corrections-ingest Body: { "run_id": "uuid", "corrections": [{ "block_id": "voice-extractor", "items": [...] }] } ``` ### What FORGE Can Automate (Future) 1. **Nightly check**: Poll improvement_items for new pending items, alert Jason via dashboard 2. **Auto-push**: When items accumulate past threshold (e.g., 5+ pending for same block), auto-push to factory 3. **Convergence tracking**: After factory re-run completes, trigger clone-tester on flagged scenarios, compare scores to previous run 4. **Status sync**: When factory block completes with corrections, mark corresponding improvement_items as resolved ### Key URLs - Dashboard: align360.asapai.net/improvement-items - Audit v3: align360.asapai.net/samuel-audit-v3 - Extraction Review: align360.asapai.net/samuel-extraction-review - Factory: factory.asapai.net - PRD: align360.asapai.net/feedback-loop-architecture ### Database Tables (Supabase) - `improvement_items` — findings from audits/reviews, with target_block, status, priority, expert_correction - `clone_feedback` — raw audit submissions (ratings, chat transcripts, feedback notes) - `system_prompts` — versioned system prompts (for tracking which version fixed what)