Insight

Each step's output is input to the next. Skipping a step produces garbage downstream.

If a step fails or produces low-quality output:

• Retry the specific step with refined input (up to 2 retries per step)
• Do NOT skip to the next step
• If a step fails 3 times, document the failure and proceed with best-effort output </HARD-GATE>

Procedure

Step 1: Root Cause Drilling

1. Call insight-root-cause-drilling SOP with { gap, evidence, knowledge }
2. Receive RootCauseResult — surfaceGap, whyChain[], rootCause, hiddenAssumptions[], unexploredAngles[]

Step 2: Stakeholder Mapping

3. Call insight-stakeholder-mapping SOP with { gap, rootCauseOutput: JSON.stringify(step1) }
4. Receive StakeholderResult — stakeholders[], underservedStakeholder, conflictingNeeds

Step 3: Tension Mining

5. Call insight-tension-mining SOP with { gap, stakeholderOutput: JSON.stringify(step2) }
6. Receive TensionResult — tensions[], primaryTension, tensionInsight

Step 4: Question Reformulation

7. Call insight-question-reformulation SOP with { tensions: JSON.stringify(step3) }
8. Receive HmwResult — hmwQuestions[], rankedQuestions[], rankingRationale

Step 5: Abstraction Laddering

9. Call insight-abstraction-laddering SOP with { hmwQuestions: JSON.stringify(step4) }
10. Receive AbstractionResult — ladder[], insightFromLadder, reframedProblem

Step 6: Assumption Audit

11. Call insight-assumption-audit SOP with { insightSoFar: JSON.stringify({ step1, step2, step3, step4, step5 }) }
12. Receive AssumptionResult — assumptions[], criticalAssumptions[], safeAssumptions[], recommendation

Step 7: Validation Gate

13. Call insight-validation SOP with { fullInsight: JSON.stringify({ step1, step2, step3, step4, step5, step6 }) }
14. Receive ValidationResult — gates[], overallVerdict (PASS|FAIL|REVISE), failedGates[], revisionGuidance?

Validation Gate Logic

IF overallVerdict == PASS:
  → Return complete ValidatedInsight to caller

IF overallVerdict == REVISE AND attempt < 2:
  → Read revisionGuidance.fromStep (1-6)
  → Re-run pipeline from that step forward, carrying prior outputs for unchanged steps
  → Increment attempt counter
  → Go to Step 7 again

IF overallVerdict == REVISE AND attempt >= 2:
  → Report partial insight to caller with status: REVISE_EXHAUSTED
  → Include best available output and failedGates[] for caller to decide

IF overallVerdict == FAIL:
  → Report failure to caller with failedGates[] and reasoning
  → Do NOT return an insight — structural failure cannot be revised

Revision Re-entry Points

• fromStep: 1 — re-run all 7 steps
• fromStep: 2 — re-run Steps 2-7, carry Step 1 output
• fromStep: 3 — re-run Steps 3-7, carry Steps 1-2 outputs
• fromStep: 4 — re-run Steps 4-7, carry Steps 1-3 outputs
• fromStep: 5 — re-run Steps 5-7, carry Steps 1-4 outputs
• fromStep: 6 — re-run Steps 6-7, carry Steps 1-5 outputs

Output JSON Structure

{
  "status": "PASS | REVISE_EXHAUSTED | FAIL",
  "gap": "<original gap input>",
  "insightChain": {
    "rootCause": {
      "surfaceGap": "...",
      "whyChain": ["why1", "why2", "why3", "why4", "why5"],
      "rootCause": "...",
      "hiddenAssumptions": ["..."],
      "unexploredAngles": ["..."]
    },
    "stakeholders": {
      "stakeholders": [{ "name": "...", "goals": "...", "painPoints": "..." }],
      "underservedStakeholder": "...",
      "conflictingNeeds": "..."
    },
    "tensions": {
      "tensions": ["..."],
      "primaryTension": "...",
      "tensionInsight": "..."
    },
    "hmwQuestions": {
      "hmwQuestions": ["..."],
      "rankedQuestions": ["..."],
      "rankingRationale": "..."
    },
    "abstraction": {
      "ladder": ["..."],
      "insightFromLadder": "...",
      "reframedProblem": "..."
    },
    "assumptions": {
      "assumptions": ["..."],
      "criticalAssumptions": ["..."],
      "safeAssumptions": ["..."],
      "recommendation": "..."
    }
  },
  "validation": {
    "gates": [{ "name": "...", "status": "PASS|FAIL", "rationale": "..." }],
    "overallVerdict": "PASS",
    "failedGates": []
  },
  "attempts": 1
}

Yield Report