inverse design

Discrimination Test Designer

Optimizes vanishing-design reflectance pairs for color-vision testing directly from spectra.

figure generated by the instrument itself. hover to let the light in.

The problem

Pseudoisochromatic plates, the Ishihara-style tests used to screen color vision, are still constructed by trial and error, a point the review literature on color-vision testing makes plainly. A vanishing design needs a color pair that a normal trichromat separates easily but the target deficiency cannot tell apart. This tool searches for that pair directly in the space of physical reflectance spectra instead of hand-tuning pigments.

The method

Both spectra are parametrized as low-order shifted Chebyshev coefficients, synthesized onto the canonical 380 to 750 nm grid and clamped to [0,1], so the results are smooth, physically valid, manufacturable reflectances. The objective maximizes Oklab separation (Delta E OK) under normal trichromacy while penalizing any separation above a confusion ceiling for the deficient observer, default 0.02, roughly one just-noticeable difference. The deficiency is modeled with the Machado 2009 CVD simulation, the same model used everywhere else in the engine. A seeded (1+1) evolution strategy with step-size adaptation runs the search, chosen over gradient methods because the objective routes through clamping and CVD simulation and because determinism matters here.

What it promises

Deterministic: the same seed, order, and iteration count reproduce the same stimulus pair exactly.
Output spectra are physically valid reflectances on the canonical 38-sample grid; low Chebyshev order keeps them smooth enough to manufacture.
Reports the achieved Delta E OK under normal vision, under the target deficiency, and their ratio, so you judge the stimulus on numbers rather than trust.
Stated limit, kept in the response itself: this is a research stimulus designer, not a validated clinical screening test. Diagnostic use requires its own trial.

Calling it

{
  "tool": "design_discrimination_test",
  "arguments": {
    "target": {
      "type": "deutan",
      "severity": 1
    },
    "illuminant": "D65",
    "max_target_delta_e_ok": 0.02,
    "order": 6,
    "iterations": 3000,
    "seed": 42
  }
}

Clone the repo, build the server (cd mcp && npm install && npm run build), then register: claude mcp add maryslab -- node <repo>/mcp/dist/server.js

Known limits

Research stimulus design only. It has never been validated against human observers, and diagnostic use requires its own clinical trial.
Confusion for the deficient observer is judged by the Machado 2009 simulation applied to displayed sRGB, not by measured human responses.
The optimizer is a stochastic search: it returns a good stimulus with honestly reported Delta E values, not a certified global optimum.