Color-blindness simulators only validate part of the picture

What it is

Color-blindness simulation tools — browser extensions, design-tool plugins, image editors' "protanopia/deuteranopia preview" filters — approximate what a full dichromat, someone entirely missing one cone type, would see. But full dichromacy is only the smaller, more severe end of the real population of color-deficient people; most are anomalous trichromats, whose third cone type is still present but spectrally shifted, so their actual perception sits somewhere between normal trichromacy and full dichromacy — a continuum that a standard simulator's binary "protanopia/deuteranopia" output does not represent. In other words, passing a simulator check shows a design holds up against the theoretically most severe case, but it does not validate the design against the far more common, generally milder, and individually variable anomalous trichromat population.

Why it happens

The mainstream simulation algorithms (Brettel, Viénot, and Machado are the well-known ones) work by transforming an image's colors into LMS cone-response space, zeroing out or otherwise modeling the "missing" cone type's response to simulate a specific full dichromacy, then transforming back to RGB for display. This is a well-defined transform, validated against actual dichromats' discrimination behavior for that specific, extreme case. But anomalous trichromacy is not "a missing cone" — it is a spectrally shifted cone still contributing a real, partial, graded signal, and that graded case does not map cleanly onto an algorithm built for the extreme case. Some tools offer an "anomalous trichromacy severity" slider (Machado and colleagues extended the original algorithm to model this), but even then the slider represents an averaged theoretical individual, not the actual variation across real users.

Studying it

Validation works by comparing simulator predictions against the measured discrimination performance of clinically diagnosed individuals — matching anomaloscope-graded severity to what the simulation predicts. This is exactly how the extended, severity-graded simulation algorithms were built and calibrated, and ongoing work continues to refine these models, but they remain population-level approximations rather than precise predictions for a specific individual.

Where it stops holding

• Predictions are relatively reliable for full dichromacy, only approximate for the more common anomalous trichromacy. The extreme theoretical case is well validated; the majority-case population only gets a rough severity slice that does not represent the actual spread within that group.
• Simulators do not cover rarer variants. Combined or acquired color vision anomalies beyond the standard red-green and blue-yellow types are typically not covered by standard simulation algorithms.
• Simulation results depend on the accuracy of the source image's color values. Screenshot compression and color-management errors introduce distortion before simulation even begins, and the simulator's own approximation error stacks on top of that.
• "Passed the simulator check" and "validated as usable for real color-deficient users" are two different claims. The former only means the design survives one of the theoretically most severe test cases.

Applying it

• Treat color-blindness simulators as a first-pass screen, not final sign-off: colors that remain indistinguishable after simulation are definitely a problem worth fixing immediately, but colors that "look fine" after simulation do not prove they will be fine for real color-deficient users.
• For color carrying critical information (alerts, financial gains/losses, medical status), after a simulator passes, do a second validation using confusion-line-based calculation or testing with actual color-deficient users, rather than stopping at a simulator screenshot comparison.
• Run simulations for at least both common types — protanopia and deuteranopia — separately, since their confusion lines point in different directions; a palette that survives one type is not guaranteed to survive the other, and checking only one type misses half the risk.
• How to check: run key screens through multiple simulation types and compare results side by side; anything that becomes indistinguishable under any one of them needs redesign, rather than concluding "color-vision-safe" from a single, most-common simulation type.

Related

• Same group: A1.06.1 Red-green deficiency is the most common type, far more prevalent in men · A1.06.2 Red-green semantic opposition fails for this population · A1.06.3 People with color vision deficiency still discriminate lightness differences
• Nearby: U4.04 Color-vision-deficiency-safe palettes
• Search terms: color blindness simulator · Brettel algorithm · Viénot algorithm · anomalous trichromacy severity