| name | qmdr-enrich |
| description | 3-pass systematic enrichment framework. Use when output needs quality elevation from baseline 60-70% to publication-grade 95-99%. Triggers on research enrichment, gap analysis, depth injection, quality improvement, or when KTG-DIRECTIVE output needs refinement. Deploys 8 expert roles across 6 priorities. Modes: QUICK 80-85%, STANDARD 92-95%, DEEP 97-99%. |
QMDR-ENRICH
Extract • Normalize • Reason • Integrate • Compress • Harmonize
3-pass quality elevation: 60% → 85% → 99%
Quick Reference
| Mode | Time | Quality | Priorities Used |
|---|
| QUICK | 5-10 min | 80-85% | 1, 4 (critical only) |
| STANDARD | 20-30 min | 92-95% | 1, 2, 3 (light), 4 |
| DEEP | 40-60 min | 97-99% | All 1-6 + compression |
6 Priorities
| # | Priority | Expert | Detects | Embeds |
|---|
| 1 | Gaps | Gap Expert | A→C jumps, undefined terms, missing context | Inline definitions, bridge content |
| 2 | Depth | Evidence + Mechanism | Unsupported claims, no "why" | Statistics, sources, mechanism |
| 3 | Perspective | Perspective Expert | One-sided, absolutes, no tradeoffs | Counter-views, conditions |
| 4 | Structure | Flow Expert | Abrupt shifts, disconnected sections | Bridge sentences, transitions |
| 5 | History | History Expert | Arbitrary recommendations, no precedent | Origin stories, validation |
| 6 | Mission | Mission Expert | Generic output, user constraints ignored | Context-specific application |
Compression Expert (MLDoE): Removes redundancy, consolidates examples, extracts meta-patterns.
3-Pass Execution
Pass 1: Baseline (60-70%)
Execute standard research:
├─ MR.RUG: 3-5 experts per domain
├─ Reasoning: FCoT (nuanced) / CoC (deterministic)
├─ Structure: SoT light
├─ Verification: CoVE-1 (basic)
└─ Output: Functional but gap-heavy
Time: 40-60% of budget
Pass 2: Enrich (85-90%)
Priority 1 (Gaps):
├─ Scan line-by-line for jumps, undefined terms
├─ Score severity 1-10, prioritize highest
├─ Embed inline, validate claim stands alone
├─ Cross-check: fix doesn't create new gaps
Priority 2 (Depth):
├─ Identify unsupported claims, missing mechanism
├─ Research evidence (stats, studies, cases)
├─ Explain WHY not just WHAT
├─ Compress: stat + mechanism + example in 1-2 sentences
Priority 4 (Structure):
├─ Find disconnected sections, abrupt shifts
├─ Create bridge: why does B follow A?
├─ Test: read A→Bridge→B, no backtracking needed
Time: 35-45% of budget
Pass 3: Polish (95-99%)
Priority 3 (Perspective):
├─ Find one-sided views, absolutes ("always"/"never")
├─ Add tradeoffs, conditions, failure scenarios
├─ Transform: "X is best" → "X when Y, else Z"
Priority 5 (History):
├─ Ground recommendations in precedent
├─ Add origin stories for "best practices"
├─ Validate patterns with real examples
Priority 6 (Mission):
├─ Align to user's actual goal
├─ Embed user constraints (regulatory, technical)
├─ Ensure output is executable by user
Compression (MLDoE):
├─ Remove redundancy from enrichments
├─ Consolidate weak examples
├─ Extract meta-patterns
└─ Target: same pages, 5x density
Time: 15-25% of budget
Quality Gates
| Gate | PASS | FAIL |
|---|
| Gap | Standalone comprehension | Needs background, undefined acronyms, skipped steps |
| Evidence | Claims traceable to source, mechanism explained | "It works" without why, stats without context |
| Perspective | Benefits AND drawbacks, explicit tradeoffs | Only benefits, absolutes, no conditions |
| Flow | Sections naturally follow, no backtracking | Abrupt shifts, isolated sections |
| History | Principles grounded in precedent | Arbitrary recommendations, patterns seem timeless |
| Mission | Output helps user accomplish goal | Generic answer, user constraints ignored |
Action on FAIL: Research → embed → re-test.
Mode Selection
QUICK: Urgent deadline, breadth over depth
Skip: P2, P3, P5, P6
STANDARD: Balanced research, most use cases
Skip: Full P5, P6
DEEP: High stakes, regulated, expert audience
All priorities + cascading improvements
Integration
KTG-DIRECTIVE → generates baseline (60-70%)
QMDR-ENRICH → elevates to target quality
Output Gate → routes to delivery format
Cascading Patterns
Pattern 1: Definition gap → depth cascade
P1: Acronym undefined
P2: Acronym explained but mechanism missing
P3: Mechanism shown but no alternative comparison
→ Insert once: definition + mechanism + comparison (solves all 3)
Pattern 2: Structure break → missing history
P4: Flow broken between A→B
Why? → B seems arbitrary without history
→ Add history of why B invented (fixes both P4 + P5)
Pattern 3: Mission goal → multi-priority trigger
User: "Present to HIPAA compliance officer"
→ P1: Too high-level for compliance
→ P2: Needs regulatory citations
→ P3: Compliance implications + benefits
→ P5: Historical precedent for approach
→ P6: Execution-ready for officer
Troubleshooting
| Issue | Cause | Solution |
|---|
| Enrichment creates new gaps | New concept introduced without definition | Embed context immediately or mark out-of-scope |
| Document growing too long | Compression layer not applied | MLDoE pass: consolidate, extract patterns |
| Output doesn't match mission | Missed P6 alignment | Revisit user goal, add context-specific recs |
QMDR-ENRICH v11 | Kevin Tan (ktg.one)
