Gap Analysis

Read the topicSize from state to select the correct column. These are HARD FLOORS.

State Ledger

When a tactic returns a Yield Report, ADD its counts to the Current column. </HARD-GATE>

Procedure — Iterative Loop

WHILE (unvalidated_gaps AND iteration < MAX_ITERATIONS):

  STEP 1 — IDENTIFY GAPS:
  From surveyResult.knowledge + surveyResult.gaps, enumerate candidate gaps:
  - What problems remain unsolved?
  - What methods are missing or underexplored?
  - What assumptions are unquestioned?
  - What benchmarks or datasets are absent?

  STEP 2 — INVESTIGATE (parallel):
  Call academic-research tactic: 3 queries targeting each candidate gap's prior work
  Call web-research tactic: 3 queries targeting each candidate gap's practical context
  Merge new papersRead and urlsVisited into state

  STEP 3 — DEEPEN with INSIGHT pipeline:
  For each candidate gap with unclear root cause or non-obvious stakeholder need:
    Call insight tactic with:
      - gap: the candidate gap JSON
      - evidence: accumulated evidence string from Steps 1-2
      - knowledge: current knowledge string
    Collect InsightResult (PASS / REVISE / FAIL verdict)
    - PASS: promote gap to validated_gaps[]
    - REVISE: refine gap framing, re-queue for next iteration
    - FAIL: move to rejected_gaps[] with reason

  STEP 4 — VALIDATE via review tactic:
  Call review tactic on the set of PASS gaps from this iteration
  Review may escalate to multiagent-debate for controversial or high-stakes gaps
  Gaps that survive review → confirmed in validatedGaps[]
  Gaps that fail review → rejected with reason

Budget Gate

1. Papers read >= 80% of target?
2. Web pages fetched >= 80% of target?
3. Gaps investigated >= minimum from budget table?

If ANY condition fails → CONTINUE iterating. Do NOT evaluate stop conditions. </HARD-GATE>

  STEP 5 — STOP CHECK:
  - All candidate gaps processed? → STOP (success)
  - No new gaps validated for 2 consecutive iterations? → STOP (plateau)
  - MAX_ITERATIONS hit? → STOP (budget)

END LOOP

Completeness Probe — Gap-Analysis

Evidence balance:

• For each gap, do I have BOTH supporting AND contradicting evidence?
• Have I searched for papers that explicitly address (or fail to address) each gap?
• Is there a gap I identified early but never investigated deeply?

Dimension completeness:

• Have I considered gaps across multiple dimensions? (methodology gaps, data gaps, evaluation gaps, theoretical gaps)
• Are there gaps that are "obvious" to practitioners but missing from the literature?
• Did I check if any gap has been partially addressed in very recent work (last 6 months)?

Cross-domain hint (optional but encouraged):

• Are there analogous problems in other fields that might reframe this gap?
• Could a technique from another domain directly address one of these gaps?

If ANY checkbox in the first two sections fails → add 1-3 targeted queries and run ONE MORE iteration. Maximum 2 extra completeness iterations. Cross-domain leads are a bonus, not a requirement here. </HARD-GATE>

State Management

• knowledge: inherited from lit-survey surveyResult; extended with gap-analysis findings each iteration
• papersRead: inherited from lit-survey; deduplicated, extended with new papers read in Steps 2-3
• urlsVisited: inherited from lit-survey; deduplicated, extended with new URLs visited in Steps 2-3
• validatedGaps: string[] of gaps that passed INSIGHT + review; grows each iteration
• rejectedGaps: string[] with rejection reasons; for transparency

Output

{
  "validatedGaps": [
    {
      "id": "gap_001",
      "description": "Gap description...",
      "rootCause": "Why this gap exists...",
      "stakeholders": ["who cares and why"],
      "evidence": "Supporting evidence from papers/web...",
      "insightVerdict": "PASS",
      "reviewVerdict": "approved"
    }
  ],
  "rejectedGaps": [
    {
      "id": "gap_002",
      "description": "Gap description...",
      "rejectionReason": "Already addressed by X et al. (2024)..."
    }
  ],
  "papersRead": ["paper1", "paper2", "..."],
  "urlsVisited": ["url1", "url2", "..."],
  "knowledge": "Extended knowledge string...",
  "iterationsUsed": 4,
  "stopReason": "all_gaps_processed | plateau | max_iterations"
}