model: opus
permissions: read, bash, write, grep, glob
context: default (primary context, not isolated)
timeout: 600s

STEP 1: Intake & Scoping
  - Receive PDF path or dialogue source
  - Scan for dialogue content
  - Estimate scope (turns, richness, unit count)
  - Present preview to human for approval to proceed

STEP 2: PDF Extraction
  - Invoke pdf-extractor agent with PDF path
  - Receive dialogue.json with confidence scores
  - Validate metadata (turns ≥8, richness ≥5%, confidence ≥70%)
  - Present extracted dialogue to human for review/approval

STEP 3: Blank Insertion
  - Invoke blank-inserter agent with dialogue.json
  - Receive dialogue_blanked.json with pedagogical scoring
  - Validate compliance (BUCKET_A % meets threshold)
  - Present blanked dialogue to human for review/approval

STEP 4: Content Validation
  - Invoke content-validator agent with dialogue_blanked.json
  - Run all 7 validators (spelling, vocabulary, tonality, flow, alternatives, insights)
  - Apply HIGH confidence (≥95%) auto-fixes automatically
  - Present validation report to human for MEDIUM/LOW confidence findings

STEP 5: Fix Loop (if needed)
  - If validation issues found:
    - CRITICAL issues: Reject scenario, return to Step 3
    - MODERATE issues: Apply suggested fixes, re-validate
    - MINOR issues: Flag for downstream refinement

STEP 6: Scenario Transformation
  - Invoke scenario-transformer agent with validation_report.json
  - Generate RoleplayScript TypeScript code
  - Validate code syntax and structure
  - Present generated code to human for final approval

STEP 7: Human Final Approval
  - Show complete scenario to human (dialogue + blanks + answers + insights)
  - Request approval/rejection/edit
  - If approved: Proceed to merge
  - If rejected: Return to Step 3 with feedback
  - If edit: Apply edits, re-validate, return to Step 7

STEP 8: Data Integration
  - Merge approved RoleplayScript into staticData.ts
  - Update scenario count in data file
  - Run npm run validate (auto-check via hook)
  - Confirm zero TypeScript errors

STEP 9: Build Verification
  - Run npm run build
  - Verify bundle size <350 kB JS, <45 kB CSS (gzipped)
  - Generate build report (scenarios count, bundle size, modules)
  - Confirm successful production build

STEP 10: Final QA Report
  - Generate extraction report:
    * Scenario count (total, new, by category)
    * Quality metrics (avg compliance, avg insight score)
    * Build statistics (bundle size, modules, errors)
    * Human intervention % (target ≤20%)
  - Commit changes to git (if requested by human)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📋 EXTRACTION APPROVAL REQUEST
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

📄 Source: New Headway Advanced Unit 4.pdf
✓ Extraction Confidence: 92%
✓ Dialogue Turns: 18
✓ Richness Score: 8.5/10

SAMPLE DIALOGUE (first 3 turns):
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Person A: Good morning, John. How are you?
Person B: I'm fine, thanks. How about you?
Person A: Not too bad. Tell me about your family.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

OPTIONS:
  [A] Approve and proceed to blank insertion
  [B] Request edits (describe changes)
  [C] Reject and skip this unit
  [?] Preview full dialogue

Your choice: _

{
  "scenario_id": "advanced-7",
  "extraction": {
    "confidence": 0.92,
    "turns": 18,
    "richness": 8.5,
    "quality_score": 0.91
  },
  "blank_insertion": {
    "bucket_a_percent": 84,
    "pedagogical_avg": 62,
    "compliance_status": "PASS",
    "quality_score": 0.88
  },
  "validation": {
    "validators_pass": 7,
    "validators_fail": 0,
    "flags_human": 2,
    "overall_confidence": 0.91,
    "quality_score": 0.89
  },
  "transformation": {
    "code_valid": true,
    "fields_complete": true,
    "quality_score": 0.95
  },
  "final_score": 0.91,
  "human_interventions": 2,
  "status": "APPROVED"
}

# Extraction Pipeline Summary

**Date:** 2026-02-08
**Duration:** 2 hours 15 minutes
**Operator:** Orchestrator-QA Agent

## Input
- Source: New Headway Advanced Units 4-6
- Target: 5-8 scenarios

## Output
- Scenarios Extracted: 7
- Scenarios Approved: 6 (85.7%)
- Scenarios Rejected: 1 (14.3%)

## Quality Metrics
- Avg Extraction Confidence: 91%
- Avg BUCKET_A Compliance: 83% (target 80%)
- Avg Validation Score: 90%
- Avg Transformation Quality: 94%
- Overall Average Quality: 89.5%

## Build Results
- Scenarios before: 39
- Scenarios after: 45 (+6)
- Bundle size: 342 kB JS / 43 kB CSS (gzipped) ✓
- TypeScript errors: 0 ✓
- Build status: SUCCESS ✓

## Human Interventions
- Total interactions: 18
- Approval requests: 7
- Fix requests: 4
- Manual edits: 2
- Human intervention %: 25% (target ≤20%)

## Recommendations
- Consider C1-C2 content for future: Advanced scenarios validated well
- Scenario quality improving over time: Later units had higher quality scores
- Consider batch processing: Can process 5 scenarios in parallel next time

---

**Ready for production:** ✓ All scenarios tested, zero data integrity issues, build successful.

# Verify new scenarios
npm run validate

# Expected output:
# ✓ 45 scenarios total
# ✓ All pass data integrity checks
# ✓ Zero TypeScript errors

# Build production bundle
npm run build

# Expected output:
# ✓ Build successful
# ✓ Bundle size: 342 kB JS / 43 kB CSS (gzipped)
# ✓ All modules compiled

# Extract 1 unit from Headway Advanced Unit 4
orchestrator-qa extract "Source Materials/New Headway Advanced Unit 4.pdf" --unit=4

# Extract 4 units in parallel
orchestrator-qa extract \
  --files="Unit4.pdf,Unit5.pdf,Unit6.pdf,Unit7.pdf" \
  --parallel=4

# Interactive mode: Extract, validate, approve, merge
orchestrator-qa extract --interactive "Source Materials/New Headway Advanced.pdf"

# Generate extraction metrics report
orchestrator-qa report

# Output: EXTRACTION_REPORT_2026-02-08.md