Back to docs

Tool Cognitive Interpolate

Category: Cognitive Transport · Since: v18.27.0 · Tier: All

Generate an interpolated persona at any point on the Wasserstein geodesic between two known personas. Geodesic interpolation preserves the internal coupling structure between traits. Simple averaging does not.

Why This Matters

Simple averaging: the midpoint between ADHD (high creativity + low patience) and power-user (moderate creativity + high patience) gives medium values for both. The creativity-patience correlation is destroyed.

Wasserstein geodesic: the midpoint keeps the creativity-patience coupling intact. Traits stay meaningfully related. The result is a coherent cognitive profile, not a bland average.

Parameters

Parameter Type Required Default Description
personaA string Yes — Starting persona
personaB string Yes — Ending persona
position number No 0.5 Position on geodesic (0 = persona A, 0.5 = midpoint, 1 = persona B)

Example

Generate a persona halfway between first-timer and power-user

Create a persona at 25% from elderly-user toward impatient-user

Response

{
  "interpolatedPersona": {
    "name": "first-timer-power-user-50pct",
    "position": 0.5,
    "traits": {
      "patience": 0.52,
      "riskTolerance": 0.58,
      "comprehension": 0.71,
      "frustrationResponse": 0.31,
      ...
    }
  },
  "description": "Persona at 50% along the Wasserstein geodesic from first-timer to power-user",
  "topChangedTraits": [
    { "trait": "comprehension", "value": 0.71, "fromA": -0.21 },
    { "trait": "riskTolerance", "value": 0.58, "fromA": -0.28 },
    { "trait": "socialProofSensitivity", "value": 0.45, "fromA": 0.25 }
  ]
}

Use Cases

  • Custom persona creation: "Give me someone between a first-timer and an expert"
  • Trait sensitivity analysis: walk along the geodesic and measure score changes. Identifies which trait transitions cause the biggest accessibility impact.
  • Adversarial testing: find the worst-case persona on the geodesic with findWorstCaseOnGeodesic

Mathematical Foundation

McCann displacement interpolation: μ_t = ((1-t)Id + tT)_# μ₀

For Gaussian cognitive profiles, the mean interpolates linearly. The covariance interpolates via matrix square root. This preserves trait correlations. Based on McCann (1997) and Zhu et al. (ICML 2023).

Related Tools

From the Blog