Back

AccessibilityEnterprise UXResearch

Dark mode isn’t accessible by default

Dark mode gets treated as the accessible choice for dashboards. Across my time at Deloitte I tested that assumption with 73 people, then designed a system that doesn’t depend on it.

Role
Senior UX Designer, research lead
Scope
Research, unmoderated testing, design system recommendations, concept system
Delivery
Deloitte
Time frame
2023 – 2026

The problem

For people with astigmatism, bright text on black can blur and glow. Dashboards are read for hours in exactly that setup.

What I did

Led an unmoderated study of a standard dark mode against an astigmatism-friendly one with 73 participants, then reviewed the literature behind it.

What came of it

Recommendations for the design system, and a concept system of three modes and eight design decisions, each graded by the strength of its evidence.

00Context

One assumption, tested and then designed around

An HR analytics dashboard in an enterprise design system, read in dark mode for hours at a time. One study with 73 people, one literature review, and a concept system that came out of both.

73participants
2modes tested
3modes designed
8graded design decisions
The problem
For people with astigmatism, bright text on black can blur and glow. Dashboards are read for hours in exactly that setup.
What I did
Led an unmoderated study of a standard dark mode against an astigmatism-friendly one with 73 participants, then reviewed the literature behind it.
What came of it
Recommendations for the design system, and a concept system of three modes and eight design decisions, each graded by the strength of its evidence.

01The problem

The accessible default, met by the eyes it suits least

Dark mode is the safe choice for enterprise dashboards. Astigmatism is common, and it changes what bright text on a black field looks like.

A · The default

The safe choice

Dark mode is easier on the eyes in a dim room, it looks considered, and it is widely recommended as an accessibility feature.

B · The reader

Four in ten adults

About four in ten adults have some astigmatism, from a 2023 pooled estimate of about 40%. In UK Biobank, 29 to 30% of eyes had corneal astigmatism of 1.0 D or more.

C · The eye

Two points of focus

In an astigmatic eye, light focuses at two points instead of one. A bright letter on a black field can scatter into a halo.

D · The screen

Read for hours

A dashboard is exactly that setup: dense, small, bright numbers, read for hours at a time.

The question was never light versus dark.

It was whether one dark theme can serve everyone who reads in it.

02The study

Two dark modes, the same tasks, seventy-three people

I ran this research independently across my time at Deloitte, on an HR analytics dashboard in the enterprise design system. I built two versions of the dashboard: a standard dark mode, and an astigmatism-friendly dark mode designed to reduce glare for astigmatic readers.

  1. 73 participants

    An unmoderated study, at a scale that gave me enough people to see variation.

  2. Same tasks, both modes

    Every participant completed the same tasks in standard dark and in astigmatism-friendly dark.

  3. Two signals

    What people did and what they reported: strong signal for the design system.

Mode A · Standard dark #FFFFFF on #000000, 21:1 contrast
Mode B · Astigmatism-friendly dark #E4E7EC on #161A22, soft contrast, looser spacing, distinct letterforms

The study informed recommendations for the design system.

03The evidence

What the research says, in both directions

After the study I went to the literature. The case for a softer dark mode is an inference from adjacent research, so I mapped what each body of evidence shows.

Points towards a softer dark

01

Polarity studies

Dark text on a light ground performs better for small text, and the advantage grows as characters shrink. Pupils are larger for white-on-black text, the leading explanation.

Piepenbrock et al., 2014 · Bernal-Molina et al., 2019
02

Induced astigmatism

In 12 young adults, 2 D of induced cylinder raised median symptom scores while reading rate held. In 21 presbyopes, reading speed, near acuity and rated screen clarity all declined with cylinder power.

Rosenfield et al., 2012 · Wolffsohn et al., 2011
03

The optical chain

A dark surround enlarges the pupil, which adds aberration and a halo around bright glyphs, stretched along one axis by astigmatism.

Optical inference

Points the other way

04

Chart reading

Each polarity helped comparable proportions of people, and the best-performing polarity often differed from stated preference.

While and Sarvghad, 2024
05

The people who need dark

Some people with astigmatism use dark mode happily, and people with photophobia, migraine or dry eye often need it.

Comfort and speed are separate outcomes, and the evidence pulls in two directions. So I didn’t design against dark mode. I designed for choice and tunability, and I graded every decision by how much evidence stands behind it.

04The system

One dashboard, three modes, a toggle in the header

I rebuilt the study dashboard in Figma as a concept system. The same HR analytics content (tab strip, trend and heatmap, at-risk roles and funnel) appears in each mode, so the only variable is the theme. Switch between them here.

The dashboard in astigmatism-friendly dark: soft off-white text on a lifted navy canvas with pale blue charts.
Intent
Cut halation and stay dark.
What changes
A soft text tier on a lifted canvas, with a contrast floor.

The heatmap in each mode

Engagement heatmap in standard dark.
Standard dark
Engagement heatmap in astigmatism-friendly dark.
Astigmatism-friendly dark
Engagement heatmap in light mode.
Light

05The decisions

Eight decisions, each graded by its evidence

Because the evidence is indirect, each decision carries a grade and a starting point to validate.

  1. Research-backed
  2. System default
  3. Practitioner opinion
  4. Extrapolation to test
DecisionGradeStarting point to validate
Never default to #000 with #FFF text in dark themesSystem default · practitioner opinionPrimer’s default dark text is #F0F6FC on #0D1117, about 17.4:1. Pure white appears only in its high-contrast theme.
Offer three dark text tiers: soft, default, high contrastSystem default · extrapolation for astigmatismSoft about #D1D7E0 (13.1:1), default #F0F6FC, high #FFFFFF. Test the soft tier.
Keep a contrast floor under the soft tierResearch-backed, indirectlyAcuity in astigmatism worsens at lower contrast. A 7:1 floor is a judgement call until testing refines it.
Cap body-text contrast in dark themesPractitioner opinionThe APCA author proposes Lc 90 as a cap. Run it as a lint.
Lighter weight in dark, never above regular to mediumPractitioner opinion · extrapolationKeep advance widths constant and tune per typeface.
Contrast as three discrete stepsExtrapolation to testEvery state can be contrast-checked in advance.
Letter and word spacing as user tokens within WCAG 1.4.12Practitioner opinionType designers disagree on tracking, so expose it as a setting.
Ship a real light option everywhere, and let the reader chooseSystem defaultNN/g advises offering dark, honouring OS settings and letting people choose.

Three ideas carry the system

Heart 01

Three text tiers

Pure white on black is a default to avoid. Three dark text tiers take its place: soft, default and high contrast.

Why this matters. The harshest pairing becomes a choice the reader makes.

Heart 02

Bounded from both sides

Contrast is something to bound from both sides, with a floor and a cap.

Why this matters. Too little contrast costs acuity. Too much is where halation lives.

Heart 03

Dials the reader sets

Contrast and spacing are dials the reader sets, in discrete steps a designer can check in advance.

Why this matters. Each dial stands on its own, so nobody has to accept a bundled accessibility mode.

06In the details

A shape problem in the eye, made worse by shape decisions on screen

The cornea or lens is shaped more like the back of a spoon than a ball. Light bends unevenly across it, so two images compete for the same spot at every distance, and the eye keeps refocusing to compensate.

Halation: bright text on a dark field scatters

Bright text on a very dark background can scatter inside an eye that already bends light unevenly, producing a perceived glow around letterforms.

Order confirmed for Thu, Oct 2#FFFFFF on #000000 · 21:1 · max contrast, max scatter

The glow is a simulated illustration of an effect that happens inside the eye. Preferences are mixed even among people with astigmatism, so the response is to make it adjustable and default to a softened version.

Letters that already overlap need room

WCAG 1.4.3 sets a 4.5:1 minimum for normal text. It’s a floor, and the standard leaves the ceiling open. Spacing works the same way: a layout keeps working when the reader opens it up.

Below the minimum

Astigmatism means light entering the eye focuses at two points instead of one, so every letterform already arrives slightly doubled.

line-height 1.15 · spacing normal

Meets SC 1.4.12

Astigmatism means light entering the eye focuses at two points instead of one, so every letterform already arrives slightly doubled.

line-height ≥1.5 · letter ≥0.12em · word ≥0.16em

Typefaces drawn to stay legible when they blur

Characters that collide

Il1 O0

In many sans-serifs, capital I, lowercase l and the digit 1 are near-identical strokes.

Atkinson Hyperlegible

Il1 O0

From the Braille Institute: a tail on the lowercase l, a curved base on the 1 and a narrower 0, so each character stays distinct under blur or doubled focus.

07What’s next

Four open questions, and the pilot that answers them

The system ends with the questions it raises.

  • Polarity by eyeHow polarity plays out across astigmatism axis, magnitude and correction.
  • Softened text pairsHow softened pairs compare with #FFF on #000.
  • Weight, size and trackingTheir effect on astigmatic readers in dark interfaces.
  • Halo and bloomHalo from bright-on-dark text, separated from OLED bloom.

The next step is a small pilot with 8 to 20 adults with astigmatism, comparing standard dark with an adjusted dark. These are my projections going in.

Comfort and clarity

Ratings lean towards the adjusted mode.

Reading speed

Holds steady between modes, within about 5%.

Individuals

Responses split, and a sizeable minority do better in standard dark.

At that sample size only very large effects reach 80% power, so the honest claim is feasibility and signal, reported as individual-level counts with confidence intervals.

08Audit checklist

Six checks a design review can run today

Each one is traceable to a source.

  • 01

    Body text meets 4.5:1; large text meets 3:1.

    WCAG 1.4.3
  • 02

    Colour always travels with a second signal for state, error and required fields.

    WCAG 1.4.1
  • 03

    Body copy uses a face with distinct l, 1, I and 0, O.

    Braille Institute, Atkinson Hyperlegible
  • 04

    A softened theme exists alongside pure black-on-white or its reverse.

    This case study
  • 05

    Content stays intact when line, paragraph, letter and word spacing increase.

    WCAG 1.4.12
  • 06

    Text and controls resize to 200% with content and function intact.

    WCAG 1.4.4

09Reflection

What I’d carry forward

When evidence is thin

Grade, don’t guess

Separate what is known from what is inferred, say which is which, and design for choice.

Two outcomes

Comfort and performance

A design can feel better without being faster. Measure both, and let each stand for itself.

On defaults

A default is a decision

Whoever picks the theme for everyone has already chosen who gets the exceptions.

Keep exploring

Explore more work

Civic servicesUX research

Finding what breaks before users do

A heuristic evaluation of the document wallet used by 550+ million citizens.

10 findingsRead case study

DeloitteEnterprise UX

The Invisible Redesign

Two enterprise platforms, rebuilt around the habits expert users already relied on.

2023 – 2025Read case study

HoneywellConversational AI

UX for Conversational AI

One intent, resolved the same way across voice and chat.

2022 – 2023Read case study

View all work