| name | consulting-qa |
| description | Validate persuasion arguments through Big 5 consulting firm lenses: McKinsey (structure/logic), BCG (strategic pragmatism), Bain (results orientation), Deloitte (evidence/compliance), Accenture (implementation feasibility). Grades each argument A-F per lens, identifies weaknesses, and prescribes concrete improvements. Use this skill whenever quality-checking defense arguments, validating persuasion scenarios, ensuring consulting-grade quality, reviewing argument logic, or auditing deliverable defense strategies. MUST trigger on any QA validation or argument quality review task. |
Consulting QA Skill
{WORKSPACE} = timestamped run directory provided by the orchestrator (e.g., _workspace/run_20260404_153000).
Purpose
Every persuasion scenario must pass quality validation through 5 distinct consulting methodologies before reaching the final report. This skill provides the evaluation framework, grading criteria, and revision standards.
When to Use
- Persuasion scenarios need quality validation
- Arguments need to be stress-tested from multiple perspectives
- Defense strategies need professional-grade quality assurance
- Need to identify which arguments are too weak for client-facing use
Two-Phase QA Architecture
QA operates in TWO sequential phases. Phase A is the most critical.
Phase A: Foundational Audit (Premise Validation)
Runs FIRST, before any argument grading. Validates that the entire upstream pipeline produced correct foundations.
| Check | Question | Fail Signal |
|---|
| Analysis validity | Were the right items extracted? Right granularity? Right assumptions surfaced? | ANALYSIS_REDO |
| Critique direction | Are criticisms targeting the real weaknesses? Right severity? Anything critical missed? | CRITIQUE_REDO |
| Defense direction | Is the defense strategy sound? Are we defending things that should be conceded? Hidden concerns correctly identified? | STRATEGY_REDO |
If ANY foundational check fails → set the appropriate REDO verdict, write detailed foundation_issues, and DO NOT proceed to Phase B. The orchestrator will restart the pipeline from the failing phase.
Phase B: Argument Quality (Scenario Quality Validation)
Only runs if Phase A verdict is FOUNDATION_PASS.
The Five Lenses — Quick Reference
| Lens | Core Question | Pass Criterion |
|---|
| McKinsey | "Is this logically airtight and well-structured?" | Pyramid structure, MECE supporting points, clear "So What?" |
| BCG | "Does this hold up under different scenarios?" | Robust across best/worst/likely cases, honest trade-offs |
| Bain | "Does this lead to concrete action?" | Clear outcomes, actionable next steps, measurable success |
| Deloitte | "Is every claim backed by verifiable evidence?" | Cited sources, regulatory awareness, audit trail |
| Accenture | "Can this actually be implemented?" | Feasible, resource-realistic, change management considered |
Read references/big5-consulting-methods.md for detailed evaluation criteria, real-world examples, and grading rubrics for each firm.
Grading Scale
| Grade | Score | Standard |
|---|
| A | 90-100 | Would impress a senior partner. Exceptional quality. |
| B | 75-89 | Professional quality. Would pass partner review with minor notes. |
| C | 60-74 | Adequate. Notable weaknesses but core argument works. |
| D | 40-59 | Below standard. Would be sent back for revision. |
| F | 0-39 | Fundamental issues. Needs complete rework. |
Verdict Logic
IF any_lens_grade == "F" → FAIL
IF count(grade <= "D") >= 2 → FAIL
IF any_lens_grade == "D" → REVISION_NEEDED
IF cultural_calibration_failed → REVISION_NEEDED
IF consistency_check_failed → REVISION_NEEDED
ELSE → PASS
Cross-Scenario Consistency Rules
Arguments across different findings must not contradict each other. Common contradictions to catch:
- Speed vs. Thoroughness: One argument claims "we prioritized speed" while another claims "we were thorough"
- Cost vs. Quality: One claims "cost-optimized" while another claims "premium quality"
- Innovation vs. Precedent: One says "novel approach" while another says "proven methodology"
- Scope: Arguments about what's in/out of scope must be consistent
Cultural Calibration Checklist
For each scenario, verify:
Revision Instructions Format
When writing revision instructions, be specific:
Bad: "Strengthen the evidence in Level 2"
Good: "Level 2 cites 'industry data' without specifics. Replace with: cite the 2024 JFSA (金融庁) guidelines Section 4.2 which mandates minimum 99.95% uptime for Tier-1 banking systems, directly supporting our 99.99% SLA commitment."
Process
- Read all upstream artifacts (01, 02, 03 JSON files)
- For each scenario, apply all 5 lenses
- Grade each level within each lens
- Check cross-scenario consistency
- Run cultural calibration checklist
- Assign verdict (PASS/REVISION_NEEDED/FAIL)
- Write specific revision instructions for non-PASS scenarios
Output Format
Write QA results to {WORKSPACE}/04_qa_results.json. See the agent definition for the complete schema.
Quality Checklist
