REP-VE-COL-001 — Color, Contrast, Low Vision, and Color-Vision Deficiency

Executive synthesis

Bottom line

There is no universal minimum color contrast that makes content usable for every person. Contrast need is a function of at least:

observer × task × target geometry × typography × polarity × adaptation
× ambient light/glare × display × surrounding context × time

The strongest defensible engineering position is:

1. Treat WCAG 2.2 as the current normative web conformance baseline:
   • ordinary text: 4.5:1 minimum (Level AA);
   • large text: 3:1 minimum;
   • ordinary text at enhanced Level AAA: 7:1;
   • large text at enhanced Level AAA: 4.5:1;
   • essential UI component boundaries, states, focus indicators, and graphical objects: 3:1 against adjacent colors;
   • never use color as the only means of conveying information.
2. Do not describe those ratios as thresholds that guarantee readability, accessibility, comfort, or comprehension for “people with low vision.”
3. For important reading, use a stronger default than the AA floor where the visual design permits it, while preserving a user-selectable lower-luminance or reversed-polarity presentation for people affected by glare or photophobia.
4. Treat color-vision deficiency primarily as a discriminability and encoding problem, not merely a foreground/background contrast problem. Add text, shape, pattern, position, line style, or direct labels.
5. Validate rendered components and representative tasks under zoom, text scaling, forced colors, light/dark schemes, CVD simulation, grayscale, glare/low-brightness stress, and real affected users. A passing hex pair is not a passing experience.

What is established, what is provisional

Objective and scope

This package asks:

• What does “minimum needed contrast” mean physiologically, empirically, and normatively?
• Which contrast requirements protect people with low vision or color-vision deficiency, and where do they fail?
• What additional variables must a production design system control?
• Do the repository hypotheses about perceptual color spaces and relational color survive falsification?

The focus is self-luminous digital interfaces. Print, safety signage, aviation, medical imaging, and mixed reality are included only where they expose transfer limits. This is not medical advice and does not define clinical impairment.

Definitions that prevent category errors

Physical luminance is emitted or reflected light weighted by a standard observer function. Relative luminance in WCAG is a normalized calculation from encoded sRGB components. Luminance contrast describes a relationship between light and dark. It is not the same as hue difference.

Contrast sensitivity is an observer's ability to detect luminance variation, often measured across spatial frequencies. Visual acuity measures resolution of detail. One does not substitute for the other.

Visibility means that a target can be detected. Legibility means that characters or symbols can be identified. Readability adds fluent reading. Comprehension, task success, preference, and comfort are further outcomes. A metric validated for one should not silently stand in for another.

Color-vision deficiency (CVD) is reduced or altered discrimination in parts of color space. “Color blind” is common language but often misleading: most affected people perceive many colors. Congenital protan and deutan conditions are common; tritan conditions are rarer. Disease, medication, aging, and ocular or neural changes can also produce acquired deficiencies.

Contrast reserve is the ratio between available contrast and the observer's threshold for the task. Reading near threshold is fragile. A nominally visible stimulus may require substantially more contrast for fluent, sustained use.

Repository context, separated from external evidence

Existing repository claims

The repository correctly treats contrast as multidimensional, recommends redundant cues, and warns against low-contrast secondary text. Its DF-ATLAS-COLOR-005 measurement layer also correctly separates source color, viewing conditions, derived perceptual spaces, task metrics, and context validation.

The risky repository assumptions are:

• that modern perceptual spaces might adequately predict interface color differences;
• that relational principles from art pedagogy might transfer to task performance;
• that “perceptually meaningful color spaces and measurable contrast” could be read as a sufficiently complete accessibility method;
• that stronger contrast can safely be recommended without naming brightness, polarity, photophobia, glare, target size, or user override.

External evidence judgment

External evidence supports the layered measurement architecture but requires a sharper separation:

color appearance/difference metric
        ≠
foreground/background luminance contrast
        ≠
observer contrast threshold
        ≠
fluent reading or task performance
        ≠
semantic decoding

No single color space or ratio spans all five.

What the standards actually require

WCAG 2.2 requires 4.5:1 for ordinary text, 3:1 for large text, 7:1 for ordinary text at AAA, and 4.5:1 for large text at AAA. Large text is at least 18 point (24 CSS px) or 14 point (about 18.66 CSS px) and bold. Required visual information that identifies UI components and states, and essential graphical objects, needs 3:1 against adjacent color. Color cannot be the only visual means of conveying information.

These are exact threshold tests for conformance: a computed 2.999:1 does not round up to 3:1. Evaluation uses the specified/rendered colors rather than trying to infer anti-aliased edge pixels, because rasterization is outside the author's control.

WCAG's own rationale says 4.5:1 starts from a 3:1 recommendation for standard text and normal vision, then multiplies by an estimated 1.5 contrast-sensitivity loss associated with 20/40 acuity. The 7:1 value analogizes to 20/80. This is a pragmatic accessibility model. It is not a modern dose-response study showing that every observer, font, size, display, and task crosses from unreadable to readable at those exact ratios.

Important conformance gaps

• Inactive controls and logos have exemptions. An exemption is not evidence that low contrast is usable.
• Text contrast evaluates the nominal foreground and background pair, not font stroke width, angular size, glare, ambient illumination, or surrounding adaptation.
• Non-text contrast only applies when the visual information is required to identify or understand the component or graphic.
• Passing “use of color” through an added label does not necessarily make adjacent regions or data series easy to discriminate.
• A page can conform and still prevent a particular low-vision user from using it comfortably or efficiently.

APCA and WCAG 3

APCA is a polarity-sensitive candidate method that aims to model perceived text contrast and connect contrast requirements to font size and weight. Those are directionally important corrections to WCAG 2's symmetric, pair-only ratio.

It must not, however, be presented as an approved WCAG requirement. The W3C Visual Contrast subgroup page explicitly says the subgroup is inactive, that its legacy material was not approved to move forward, and that it should not be used as W3C guidance. The March 2026 WCAG 3 publication is a Working Draft and states that publication does not imply W3C endorsement. Therefore:

• use WCAG 2.2 for current conformance;
• APCA may be run as an additional design diagnostic or research variable;
• do not fail a product solely on an experimental APCA target unless the organization has explicitly adopted that internal policy;
• record version and parameters, because experimental methods can change.

Low vision: why one minimum cannot work

“Low vision” combines materially different limitations:

• reduced acuity;
• reduced contrast sensitivity;
• central or peripheral field loss;
• cloudy ocular media and intraocular scatter;
• photophobia and glare sensitivity;
• reduced light adaptation;
• impaired color discrimination;
• combinations of the above.

Rubin and Legge measured 19 low-vision readers and found critical reading contrasts higher than normal in 16, averaging 3.9 times higher. Critical contrast correlated strongly with letter contrast sensitivity, while peak reading rate also depended on field loss. Five read reversed-polarity text faster; four had cloudy ocular media. This directly falsifies both “one ratio fits low vision” and “one polarity fits low vision.”

Later work found that reducing text contrast from 90% to 45% produced at least a 20% reading-speed loss in roughly one third of a group at 0.6 logMAR acuity and two thirds at 1.0 logMAR. The result also shows why testing only maximum-contrast text hides functional differences.

Contrast is necessary but coupled to size

Small or thin strokes contain high spatial frequencies, where sensitivity is lower and more vulnerable to blur, age, glare, and display rasterization. Increasing font size and weight can increase visibility even when the nominal color pair is unchanged. Conversely, a 4.5:1 pair can be fragile when used with very thin type, small text, low pixel density, transparency, background imagery, or a font whose actual strokes are much thinner than its nominal weight implies.

This supports a design rule: never spend the entire visibility budget on color contrast. Provide adequate size, stroke width, spacing, zoom, text scaling, and reflow.

Contrast is coupled to luminance and adaptation

The same nominal pair can perform differently as display brightness, black level, ambient light, reflections, and the viewer's adaptation change. Operational cockpit studies found that under low/moderate ambient light some symbols reached a performance plateau at low ratios, while extreme daylight adaptation benefited from ratios up to 30:1. Medical-display research similarly shows that reflected ambient light raises dark luminance and reduces effective contrast.

A contrast ratio therefore describes a pair under a model; it does not fully describe retinal contrast in the field.

Polarity and the “dark mode” contradiction

Controlled studies in normally sighted adults frequently find a positive polarity advantage—dark text on a light background—for proofreading and fine detail. Higher overall luminance constricts the pupil and can sharpen the retinal image. The advantage grows for smaller characters.

That population result cannot become a universal accessibility rule. People with cloudy ocular media may experience more scatter from a large bright background and read faster with light text on dark. In a survey of 133 adults with low vision, 46% preferred reversed contrast, 39% preferred dark-on-light, and 14% reported no preference. Preference is not the same as performance, but the heterogeneity is decisive.

The correct conclusion is:

• positive polarity is a reasonable general default for dense fine-detail reading when glare and light sensitivity are controlled;
• a well-designed dark/reversed theme is a necessary alternative for many;
• neither should be forced;
• avoid absolute white on black as the only dark option when blooming, halation, or astigmatic blur is a concern, but do not lower contrast below conformance in pursuit of “softness”;
• allow brightness and theme choice at the user or system level.

More contrast is not identical to more light

People can have both reduced contrast sensitivity and photophobia. Making a background brighter can increase available contrast yet cause pain, glare, or scatter. A high ratio made from two relatively low-luminance colors may be more tolerable than the same ratio made with a glaring white field, depending on the display and observer. WCAG relative luminance does not encode screen luminance in cd/m² or total luminous area.

This defeats the theory that “maximize black–white contrast” is a universal optimization. The robust solution is sufficient contrast plus theme, brightness, and color customization.

Color-vision deficiency: what actually fails

Prevalence is variable, not one folklore number

A 2025 systematic review and meta-analysis covering 1,703,619 participants from 56 studies in 21 countries estimated congenital CVD at 2.59% globally, 4.38% in males and 0.64% in females, with very high between-study heterogeneity. Deutan conditions were most common. The often-repeated “8% of men” estimate describes some populations, particularly people of Northern European ancestry, not humanity.

Acquired CVD broadens the affected population. Aging lenses and cataract can produce blue-yellow loss; retinal disease often affects short-wavelength discrimination; optic-nerve disease often affects red-green discrimination. Medication and systemic disease can also matter.

CVD is not solved by hue naming

The failure is not simply that a person calls red “green.” Colors that are distinct for a standard trichromat can collapse toward the same percept under a particular deficiency. Protan conditions also alter perceived lightness of long-wavelength reds, making dark red especially easy to lose against black. Tritan and acquired losses create different confusion axes.

No fixed list such as “never use red and green” is sufficient. Blue/purple, green/brown, pink/gray, yellow/light green, and other pairs can fail depending on lightness, saturation, size, adjacency, and deficiency. Conversely, some red and green colors remain distinguishable when their lightness and other cues are strongly separated. The safer rule is to avoid depending on any hue pair alone.

Redundant encoding is the strongest general intervention

For critical status, instructions, selection, errors, warnings, routes, or data series, combine color with one or more of:

• direct text labels;
• icons whose shapes are distinguishable without color;
• line style or marker shape;
• pattern or texture;
• position or grouping;
• borders or separators;
• explicit state text such as “Error,” “Selected,” or “Offline”;
• programmatically available names, roles, values, and status messages.

Redundancy should be information-preserving, not decorative. Two colored icons with nearly identical silhouettes are still color-dependent. A legend that requires repeated color matching may technically add text while leaving the task slow and error-prone; direct labels are usually stronger.

Simulation is a stress test, not an affected user's testimony

The Machado–Oliveira–Fernandes model provides a physiologically grounded, experimentally evaluated simulation across anomalous trichromacy and dichromacy. It is useful for finding likely collapses. But:

• CVD type and severity vary continuously;
• acquired deficiencies may not match congenital models;
• display spectra, profiles, gamut mapping, and ambient conditions matter;
• a simulation shown to a standard observer does not recreate a lifetime of adaptation or learned strategies;
• image appearance is not equivalent to task performance.

Use multiple simulated severities and types, then validate critical workflows with affected participants. Grayscale conversion is a quick test for whether lightness carries structure, but unique grayscale values do not prove universal CVD safety or adequate contrast.

Perceptual color spaces: hypothesis VE-COL-H1

Initial hypothesis

Modern perceptual spaces adequately predict interface color differences.

Falsification result

Rejected as stated; retained only as a bounded engineering tool.

Oklab/OKLCH improves palette manipulation, interpolation, and approximate perceptual spacing over raw RGB or HSL. CAM16 and CAM16-UCS explicitly model viewing-condition adaptation and have strong color-science uses. Experimental wide-gamut comparisons show CAM16-UCS can outperform several alternatives while still performing unevenly across hue regions.

But a color-difference metric predicts neither:

• text/background readability;
• critical edge visibility;
• color-category naming;
• semantic interpretation;
• discrimination for every CVD observer;
• resistance to glare, blur, crowding, or small target size;
• attention priority or task performance.

Even a perceptually uniform space is uniform only for the observer model, dataset, magnitude range, and viewing assumptions used to build or test it. Simultaneous contrast and adaptation can change appearance without changing the stored coordinates.

Revised hypothesis

Task-appropriate color spaces improve palette construction and color-difference prediction under stated viewing assumptions, but accessibility requires separate luminance-contrast, CVD, geometry, context, and task validation.

Confidence: high.

Art-pedagogy transfer: hypothesis VE-COL-H2

Initial hypothesis

Relational color principles transfer from art pedagogy to digital task performance.

Falsification result

Narrowed substantially.

The relational insight survives: surrounding color, area, adaptation, contour, and spatial frequency can change appearance. This is compatible with modern simultaneous-contrast and appearance research. Historical frameworks are useful for generating variations and noticing context.

What does not survive is the inference that named contrasts—complementary, warm/cool, saturation, or hue category—predict detection, readability, comprehension, urgency, or accessibility. A complementary pair can have poor luminance separation; high chroma can compete with hierarchy; context can produce contrast, assimilation, or little effect.

Revised hypothesis

Historical relational color systems are generative vocabularies. Their variables must be translated into measurable luminance, color-difference, spatial, semantic, and task predictions before they can justify interface decisions.

Confidence: medium-high.

A layered engineering model

Layer 1 — Normative floor

Meet the applicable WCAG 2.2 criterion on every allowed component state and theme. Test hover, focus, selected, disabled-but-readable content, validation, loading, and high-contrast variants—not only default screenshots.

Layer 2 — Visibility reserve

For important text and critical controls, avoid designing exactly on the threshold. Account for thin strokes, alpha, overlays, gradients, imagery, anti-aliasing, low-quality displays, glare, and aging. Prefer a higher contrast reserve for:

• long-form or dense text;
• small or thin text;
• time-critical information;
• safety, health, finance, or destructive actions;
• outdoor or mobile use;
• older or low-vision audiences;
• information that cannot be recovered after an error.

This is a risk rule, not a new universal ratio. A candidate internal default is 7:1 for ordinary critical text where practical, with representative-user tests and user customization. Do not claim AAA alone makes a workflow accessible.

Layer 3 — Geometry and typography

Validate:

• rendered font family, variable-font axes, size, weight, and actual stroke;
• visual angle at realistic viewing distance;
• zoom to 200% and text enlargement beyond it where supported;
• reflow and no clipped text;
• spacing, crowding, target size, focus-indicator area, and boundary thickness;
• icons at their smallest deployed size.

Layer 4 — Color discriminability

• Do not encode meaning solely by hue.
• Check color pairs after CVD simulation across protan, deutan, and tritan types and multiple severities.
• Check grayscale to expose hidden lightness dependencies.
• For categorical palettes, prefer fewer simultaneous categories, direct labels, and separable lightness plus hue.
• For ordered data, use a monotonic perceptual-lightness sequence unless the task requires a justified alternative.
• Avoid rainbow maps for ordered magnitude because they introduce non-monotonic lightness and false boundaries.

Layer 5 — Adaptability

• Respect prefers-color-scheme; support both light and dark themes when feasible.
• Allow forced colors to replace author colors. Use semantic CSS system colors and test forced-colors: active.
• Avoid forced-color-adjust: none except for a narrowly justified element whose meaning remains perceivable and whose custom rendering is fully tested.
• Do not put required information only in CSS background images, box shadows, or color fills that forced colors may remove.
• Do not block browser or OS contrast, color, font, zoom, or brightness preferences.
• Preserve programmatic semantics so nonvisual presentation remains possible.

Layer 6 — Context and task validation

Test the rendered experience, not only token pairs:

For consequential workflows, measure accuracy, completion time, errors, confidence, fatigue, and abandonment with affected participants. Preference and performance should both be recorded and not conflated.

Practical decision table

Anti-patterns

• “It passes 4.5:1, therefore it is accessible.”
• Lowering secondary text to the AA boundary to manufacture hierarchy.
• Relying on 3:1 large-text relief while the implementation can render smaller due to responsive type, zoom, or font substitution.
• Green/red status dots with identical shape and no text.
• Treating CVD as one simulated switch at maximum severity.
• Recoloring content to “correct” CVD without preserving stable semantic cues.
• Using ΔE, OKLCH distance, saturation, or complementary hue as a substitute for contrast testing.
• Checking a token against one background while components add opacity, gradients, overlays, images, elevation, or state colors.
• Preventing forced colors to preserve brand identity.
• Assuming dark mode reduces eye strain for everyone.
• Assuming preference scores prove faster or more accurate performance.
• Treating a standard-observer calculation as a measurement of a specific display or human.

Evidence registry

Hypothesis registry changes proposed

Contradiction ledger

Failed assumptions and negative results

• A search for an empirically demonstrated universal ratio for low vision did not find one. Evidence instead showed observer and condition heterogeneity.
• The WCAG 4.5:1 rationale did not resolve to a direct modern interface trial; it is a multiplication of a normal-observer recommendation by a generalized loss factor.
• APCA could not be promoted as the WCAG 3 method. Current W3C pages explicitly block that claim.
• A single “best dark mode” recommendation did not survive polarity evidence.
• A universal CVD-safe named palette did not survive variation in deficiency, severity, display, and task.
• A single ΔE threshold did not survive the distinction between color matching, categorical discrimination, small-symbol visibility, and reading.
• More searching for brand or UX blog recommendations had low value because those sources mostly repeat WCAG or palette folklore without primary evidence.

Source acceptance and rejection

Accepted source classes

• W3C normative specifications and official understanding documents for conformance claims;
• CIE material for color appearance scope;
• primary psychophysics, ergonomics, and visualization studies;
• systematic reviews for prevalence;
• affected-user studies for variability and preferences;
• official CSS specifications for customization behavior.

Rejected or down-weighted

• vendor contrast-checker pages: useful tools, not independent evidence;
• design blogs and palette lists: mostly derivative and rarely task-validated;
• APCA promotional documentation for normative status: primary project documentation but interested and not W3C-approved;
• Reddit discussions: useful as leads or objections, not load-bearing evidence;
• studies using only standard-vision observers to claim CVD accessibility;
• studies reporting preference alone as readability or performance;
• simulation-only interface studies presented as affected-user validation;
• papers without retrievable methods or clear population/task descriptions.

Engineering implementation requirements

Token metadata

Each semantic color token should declare:

token_id:
semantic_role:
allowed_foregrounds:
allowed_backgrounds:
allowed_states:
themes:
wcag_target:
criticality:
redundant_cues:
forced_colors_mapping:
cvd_checks:
rendered_component_tests:
known_exceptions:
owner:

Do not certify isolated palette swatches. Certify allowed role × background × state × theme × component combinations.

Automated checks

• compute WCAG 2.2 contrast without rounding;
• resolve alpha compositing and all actual backgrounds;
• cover theme and state matrices;
• flag color-only differences in status, charts, selection, and instructions;
• run screenshot checks under forced colors and CVD simulation;
• detect gradients or images that create variable local contrast;
• retain manual review gates for focus visibility, semantic redundancy, typography, and charts.

Automation should report uncertainty rather than fabricate a pass when the background, compositing, display profile, or image region is unknown.

Recommended validation program

Experiment 1 — contrast reserve × typography

Factor ordinary text across WCAG ratios (3, 4.5, 7, and higher), size, weight, polarity, and background luminance. Recruit typical vision, older adults, and stratified low-vision participants. Measure reading accuracy, rate, critical print size, errors, fatigue, and preference.

Disconfirmation target: if 4.5:1 produces no worse task outcomes than higher ratios across risk-relevant groups and conditions, the proposed reserve rule should narrow.

Experiment 2 — semantic status encoding

Compare color-only, color + icon, color + text, and color + icon + text across CVD types and low-vision groups. Measure identification accuracy and time under normal, simulated glare, and forced-colors conditions.

Disconfirmation target: if redundant cues slow or confuse users without improving accuracy, determine whether cue design—not redundancy itself—is the cause.

Experiment 3 — metric tournament

Test whether WCAG ratio, APCA versioned score, ΔEOK, CAM16-UCS distance, and simple lightness difference predict rendered text, icon, boundary, and categorical-color performance. Pre-register separate outcome models and test on held-out devices and observers.

Disconfirmation target: no metric should be promoted if it wins only on the development dataset or one target class.

Experiment 4 — customization versus universal theme

Compare a fixed high-contrast theme with user choice among tested light, dark, and low-luminance high-contrast themes. Measure task outcomes and switching behavior over sustained use.

Disconfirmation target: if customization produces errors from inconsistent semantics or does not improve outcomes, constrain options while preserving system preferences.

Research debt and open questions

• Direct evidence connecting WCAG ratios to modern variable fonts, high-density displays, and representative low-vision task performance is limited.
• APCA needs independent, preregistered comparison against alternatives across affected populations and real tasks.
• Better models are needed for local contrast over images, gradients, transparency, and HDR tone mapping.
• CVD simulation should be compared with affected-user discrimination across modern wide-gamut displays.
• Forced-colors behavior needs cross-browser and component-library field data.
• The interaction of photophobia, contrast sensitivity, total luminous area, and sustained fatigue needs stronger trials.
• More evidence is needed for low-vision users with multiple simultaneous impairments, not diagnosis-isolated samples.
• Semantic color meaning must be tested cross-culturally and separated from simple visual discriminability.
• Non-Latin scripts may have different stroke-density and size requirements.
• Visual accessibility under outdoor mobile conditions, AR/VR, and automotive use cannot be inferred from desktop WCAG calculations alone.

Theory impact

Retain

• contrast and appearance are relational and viewing-condition dependent;
• color should be modeled in layers;
• redundant encoding is a governing safety principle;
• historical color theory is a vocabulary, not a causal law.

Revise

• replace “use perceptually meaningful spaces and measurable contrast” with “use task-appropriate color spaces, normative contrast floors, independent semantic cues, user adaptation, and representative task validation”;
• replace “target stronger contrast for critical information” with a bounded reserve rule that also protects against excess luminance and forced polarity;
• make observer, task, geometry, luminance environment, device, and state mandatory fields in contrast evidence.

Do not promote yet

• a repository-wide APCA requirement;
• a universal internal contrast ratio beyond WCAG;
• an approved categorical palette;
• a claim that OKLCH or CAM16-UCS distance establishes accessibility;
• light or dark mode as the universally preferred accessible theme.

Repository updates proposed

1. Add REP-VE-COL-001 as a research-draft execution package.
2. Add VE-COL-H3 through VE-COL-H6 to the future hypothesis registry checkpoint.
3. Add EV-VE-COL-001 through EV-VE-COL-020 to the future evidence registry, after ID-collision and bibliography audit.
4. Update agent-context/UI-FOUNDATIONS.md to distinguish conformance floor, contrast reserve, absolute luminance, and user customization.
5. Add forced-colors and multi-severity CVD checks to agent-context/UI-DECISION-CHECKLIST.md.
6. Preserve DF-ATLAS-COLOR-005 but add a warning that color-difference metrics do not validate text or UI contrast.
7. Gate the section's engineering-validation prompt on affected-user testing plans and a versioned metric tournament.

No canonical file was changed by this package.

Research journal

Cycle 1 — standards

Inspected WCAG 2.2 text, enhanced text, non-text contrast, use-of-color, and official rationale. Expected a strong empirical threshold basis. Found a useful but simplified derivation, changing the interpretation from human threshold to conformance proxy.

Cycle 2 — low vision

Searched contrast sensitivity, reading contrast, polarity, glare, and digital reading studies. Expected “higher is better” with a possible plateau. Found large observer heterogeneity, polarity reversals linked to ocular scatter, and conflicting luminance needs from photophobia.

Cycle 3 — color-vision deficiency

Searched prevalence, congenital/acquired mechanisms, simulation, and accessible visualization. Rejected the “8% of men” universal statement. Retained multi-channel redundancy and simulation as a screen, not proof.

Cycle 4 — metric challenge

Compared WCAG, APCA status, Oklab/OKLCH, CAM16, and uniform color-difference evidence. Found that these address different target variables. APCA's W3C status was weaker than common design discourse suggests.

Cycle 5 — environmental and implementation boundaries

Reviewed polarity, pupil-size, cockpit adaptation, display glare, CSS forced colors, and user preferences. Concluded that absolute luminance, total area, adaptation, rendering, and customization are first-class variables.

Search strategies recorded

• official-domain searches for WCAG 2.2, WCAG 3, low-vision requirements, forced colors, CSS Color 4, and CIECAM16;
• PubMed/PMC searches combining low vision, contrast sensitivity, reading, polarity, ambient illumination, aging, and CVD;
• title/author search for the Machado CVD simulation model;
• repository-wide rg for contrast, CVD, APCA, OKLCH, ΔE, low vision, and existing Project Atlas claims;
• contradiction-first searches for maximum contrast, dark-mode advantage, WCAG empirical validation, and APCA approval status.

Access limitations

• Several standards referenced by WCAG (older ISO and ANSI documents) were not freely available in full.
• Some primary papers were accessible only through abstracts.
• No original participant testing was performed.
• The search was broad but not a formal systematic review with dual screening, protocol registration, or database export.

Exact resume point

Next, execute the four experiments above or perform a preregistered systematic review of interface contrast studies after 2008. Before registry integration, audit IDs, normalize full bibliographic records, and have a low-vision/CVD specialist and affected-user reviewers challenge the engineering defaults.

Sources

Standards and official guidance

• W3C. Web Content Accessibility Guidelines (WCAG) 2.2.
• W3C. Understanding SC 1.4.6: Contrast (Enhanced).
• W3C. Understanding SC 1.4.11: Non-text Contrast.
• W3C. Technique G111: Using color and pattern.
• W3C. Accessibility Requirements for People with Low Vision.
• W3C. CSS Color Adjustment Module Level 1.
• W3C. CSS Color Module Level 4.
• W3C. WCAG 3.0 Working Draft.
• W3C. Visual Contrast of Text Subgroup archive.
• CIE. CIE 248:2022, CIECAM16.

Research and reviews

• Rubin GS, Legge GE. Psychophysics of reading VI: the role of contrast in low vision. Vision Research. 1989.
• Giacomelli G, et al. Contrast reduction and reading: assessment and reliability with the Reading Explorer test. 2010.
• Legge GE. Reading Digital with Low Vision. Visible Language. 2017.
• Crossland MD, et al. Online Survey of Digital Reading by Adults with Low Vision. 2020.
• Piepenbrock C, Mayr S, Buchner A. Smaller pupil size and better proofreading performance with positive polarity displays. Ergonomics. 2014.
• Buchner A, Mayr S, Brandt M. The advantage of positive text-background polarity is due to high display luminance. Ergonomics. 2009.
• Jeong YD, et al. Global prevalence of congenital color vision deficiency among children and adolescents, 1932–2022. Ophthalmology. 2025.
• National Research Council. Tests of Visual Functions. In Visual Impairments. 2002.
• Machado GM, Oliveira MM, Fernandes LAF. A physiologically-based model for simulation of color vision deficiency. IEEE TVCG. 2009.
• Nuñez JR, Anderton CR, Renslow RS. Optimizing colormaps with consideration for color vision deficiency. PLOS ONE. 2018.
• Huang M, et al. Testing uniform colour spaces using colour differences of a wide colour gamut. Optics Express. 2021.
• Silverstein LD, et al. Luminance and luminance contrast requirements for legibility of self-luminous displays in aircraft cockpits. Aviation, Space, and Environmental Medicine. 2005.

Completion self-audit

What did I expect? A tiered recommendation centered on 4.5:1, 7:1, and a color-blind palette.

What most challenged it? Low-vision polarity reversals, photophobia plus reduced contrast sensitivity, the indirect WCAG ratio derivation, and the explicitly unapproved status of archived APCA work at W3C.

Strongest conclusion: contrast compliance is necessary but cannot be a complete model of visual accessibility. Independent evidence converges on geometry, observer, adaptation, redundancy, and customization.

Most fragile conclusion: using 7:1 as a candidate internal target for critical ordinary text. It is prudent but not established as universally optimal; it requires task and population validation.

Likely overgeneralization risk: transferring reading and cockpit findings to all interactive components.

Missing stakeholders: people with combined low vision and CVD, photophobia, non-Latin-script readers, mobile outdoor users, and users of high-contrast assistive themes.

What would a skeptical expert dispute? The evidence grade, the treatment of APCA, whether preference justifies theme support, and whether “contrast reserve” can guide design without a fixed number.

What evidence would most change the roadmap? Independent, preregistered trials comparing contrast metrics across representative affected users, typographies, devices, lighting conditions, and consequential interface tasks.

Reconstructability: source classes, accepted/rejected evidence, hypotheses, contradictions, queries, limitations, proposed updates, and exact next work are recorded above.