Game-Theoretic Research Strategy Analysis

Dimension Extraction Template

From the literature knowledge base, extract independent axes of variation:

dimensions:
  D1_representation:
    name: "Feature Representation"
    options:
      - id: d1a
        name: "Raw time-domain"
        known_in: ["Paper1", "Paper2"]     # who already used this
      - id: d1b
        name: "Frequency-domain (FFT/STFT)"
        known_in: ["Paper3"]
    # ... more options

  D2_architecture:
    name: "Core Architecture"
    options: [...]

  # Typically 4-7 dimensions, 3-5 options each

Independence rule: Changing one dimension must NOT force changes in another. Mutual exclusivity: Options within a dimension are mutually exclusive.

Strategy Card Template

For each viable combination:

strategy_id: S-{number}
combination: { D1: d1c, D2: d2d, D3: d3b, D4: d4b, D5: d5b }

scores:
  feasibility: X/10          # Can we actually build this?
  novelty: X/10              # Is this new enough for the target venue?
  sota_potential: X/10        # Probability of beating SOTA?
  theoretical_soundness: X/10 # Is the math clean?

backward_induction:
  target_outcome: "Metric > threshold on Dataset"
  required_properties:
    - property: "Must capture multi-scale features"
      provided_by: "D1=d1c (CWT)"
    - property: "Must be robust to noise"
      provided_by: "D3=d3b (Lipschitz)"
  critical_success_factors:
    - "Parameter X must be tuned correctly"

expected_value:
  improvement_over_sota: "+X.X%"
  confidence: "LOW | MEDIUM | HIGH"

Backward Induction Procedure

Define terminal payoffs: For each strategy, estimate P(beat SOTA) and E[improvement]
Identify chance nodes: Training stability, data quality, hyperparameter sensitivity
Assign probabilities: Based on literature evidence and domain knowledge
Propagate backwards: E[value] = Σ p(chance) × payoff(decision)
Select: Strategy with highest risk-adjusted E[value]

Risk adjustment: risk_adjusted = E[value] × (1 - variance_penalty)

Sensitivity Analysis Template

For top 3 strategies, evaluate each critical parameter:

Parameter         Base    Range        Impact    Notes
────────────────────────────────────────────────────
[param_name]      [val]   [low, high]  HIGH/MED  [why it matters]

Robustness verdict:

All critical params right → X% chance of beating SOTA
2/3 right → Y% chance
1/3 right → Z% chance

Probability Calibration Rules