Adaptive Vision Testing

Triad

Tri-dimensional Rapid Interleaved Acuity Dissociation

A letter chart measures one thing. TRIAD measures three — visual acuity, contrast sensitivity, and crowding resistance — in a single adaptive session of 45 trials, with three Bayesian staircases interleaved by uncertainty and an AI interpretation of the result.

45 Trials3 ChannelsQUEST+ Bayesian10-AFC Sloan LettersIncluded In Every Lane
Live three-channel score radar

The Trial Screen

One screen. Three staircases.

Channels aren't tested in blocks — every trial, the engine picks whichever channel has the most uncertainty left and presents a single Sloan letter. The patient answers from ten choices; the posterior updates; repeat.

A · AcuityB · ContrastC · Crowding
14 / 45
Channel A — White letter · size adapts

Which letter did you see?

The Three Channels

Three dimensions of sight.

Channel A

Visual Acuity

The familiar dimension — the smallest letter the patient can resolve. A white Sloan letter on black, sized adaptively along the logMAR axis.

Stimulus rangelogMAR −0.2 → 1.1 (≈ 20/12.6 → 20/251)
RoleAnchors letter size for channels B & C
Normal≤ 0.10 logMAR
Channel B

Contrast Sensitivity

The dimension letter charts miss. Letter size stays fixed at 2.0× the running acuity estimate while the grey level fades toward the background.

Stimulus rangelog₁₀ contrast −2.0 → 0.0 (luminance ramp)
AnchorLetter fixed at 2.0× acuity threshold
NormallogCS ≥ 1.70
Channel C

Crowding Resistance

How close can flankers get before recognition breaks down? Bars above and below the letter move inward adaptively — the psychophysics of amblyopia and cortical processing.

Stimulus rangelog₁₀ spacing ratio −0.3 → 1.0
AnchorLetter fixed at 1.5× acuity threshold
NormallogCSR ≤ −0.20

How It Works

45 trials. Zero wasted questions.

Start
1
2
3
4
Results

Pick the noisiest channel

Before every trial, the engine compares the posterior uncertainty of all three staircases and tests whichever channel it knows least about — so no trial is spent confirming what's already certain.

Fixate, then flash

A 350 ms crosshair centers the patient's gaze, then a single Sloan letter appears — sized, faded, or flanked according to the active channel's next most informative stimulus level.

Ten-alternative forced choice

The patient picks from the ten Sloan letters (C D E F H K N O R S) on screen or keyboard. Guessing is built into the model — a 10% guess rate is part of the psychometric function, not a source of error.

Bayesian update

Each response updates that channel's QUEST+ posterior over threshold and slope. Fifteen trials per channel later, TRIAD reports thresholds with their standard deviations — you know the number and how much to trust it.

45
Adaptive Trials
3
Vision Channels
15
Trials Per Channel
6
Clinical Patterns

Instant Classification

Six clinical patterns.

The three normalized scores classify every result the moment the last trial ends — a triage signal, not a diagnosis.

Within normal limits

All scores ≥ 70%

Acuity, contrast, and crowding all fall in the expected range. Routine monitoring.

Optical / media loss

Acuity < 50% · Contrast < 50% · Crowding ≥ 60%

Acuity and contrast reduced together while crowding is preserved — consistent with refractive error, cataract, or media opacity. Refraction and slit-lamp first.

Early neural sensitivity loss

Contrast < 50% · Acuity ≥ 50% · Crowding ≥ 50%

Isolated contrast loss with intact acuity — the signature that letter charts miss. Warrants retinal and optic nerve evaluation.

Cortical processing deficit

Crowding < 50% · Acuity ≥ 40% · Contrast ≥ 50%

Disproportionate crowding weakness — associated with amblyopia and cortical visual processing differences.

Multi-dimensional loss

All scores < 40%

Every channel reduced. Suggests significant pathology or unreliable responses — comprehensive workup indicated.

Borderline result

Everything else

Mixed scores that don't match a classic pattern. Repeat testing or targeted follow-up on the weakest channel.

The CCR Biomarker

CCR = logCSR / logCS   (healthy baseline ≈ −0.17)

The Crowding–Contrast Ratio compares the two non-acuity channels against each other. A ratio near baseline is typical; drifting above zero means crowding is deteriorating faster than contrast (a cortical/amblyopic direction), while a ratio far below baseline points at contrast-dominant loss (an optical, media, or retinal direction).

After The Last Trial

Numbers,pattern,narrative.

TRIAD hands all ten numeric outputs — three thresholds, their standard deviations, the normalized scores, and the CCR — to Claude, which returns a two-to-three sentence clinical narrative plus three or four concrete considerations. It's prompted to respect uncertainty (an SD above 0.15 is called out, not glossed over) and to stay non-alarmist: language for a clinician's note, not a scare.

Run TRIAD In Your Lane ↗
Channel A · Acuity
0.10 logMAR≈ 20/25 · ±0.07 · 80%
Channel B · Contrast Sensitivity
1.60 logCS±0.09 · 75%
Channel C · Crowding · CCR −0.07
−0.11 logCSR±0.10 · 85%
✦ AI Interpretation — example · Powered by Claude

All three channels fall within normal limits, led by strong crowding resistance. The acuity threshold of 0.10 logMAR (≈20/25) and contrast sensitivity of 1.60 logCS are unremarkable for a routine exam; the CCR of −0.07 sits slightly above the −0.17 baseline but within measurement uncertainty.

  • Routine recall interval appropriate based on this result
  • Contrast SD of ±0.09 is acceptable; repeat only if symptoms suggest otherwise
  • Consider a dedicated contrast workup if night-vision complaints arise

The Science

Built on
psychophysics.

QUEST+ Adaptive Estimation

Each channel runs a full Bayesian QUEST+ staircase (Watson, 2017) over a Weibull psychometric function — every stimulus is placed where it's expected to be most informative about the threshold.

p(x, t) = γ + (1 − γ − δ) × (1 − exp(−10^(β·(x − t))))

Crowding, Quantified

Channel C operationalizes Bouma's law — recognition collapses when flanker spacing falls below a critical fraction of eccentricity. TRIAD measures that critical spacing directly at fixation.

References

  • Watson, A. B. (2017). QUEST+: A general multidimensional Bayesian adaptive psychometric method. Journal of Vision, 17(3):10.
  • Bouma, H. (1970). Interaction effects in parafoveal letter recognition. Nature, 226, 177–178.
  • Pelli, D. G., Robson, J. G., & Wilkins, A. J. (1988). The design of a new letter chart for measuring contrast sensitivity. Clinical Vision Sciences, 2, 187–199.
  • Levi, D. M. (2008). Crowding — an essential bottleneck for object recognition. Vision Research, 48(5), 635–654.
  • Strasburger, H., Rentschler, I., & Jüttner, M. (2011). Peripheral vision and pattern recognition: a review. Journal of Vision, 11(5):13.

Where You Get It

Two distances. One test.

Included in every lane

TRIAD Distance

Runs on every activated exam lane at your calibrated testing distance — no extra hardware, no add-on fee. Open it straight from the lane's chart menu, on the same screen your acuity charts run on.

Get started free →
Ocular Edge Clinical

TRIAD Near

The same three-channel engine at near working distance (33 cm default), part of the Ocular Edge Clinical near-testing suite alongside PHOTOSTRESS recovery and WARP metamorphopsia mapping. Enabled per-account — talk to us.

Contact us to enable →

See all three
dimensions.

20/20 tells you one number. TRIAD tells you three — with the uncertainty, the pattern, and the narrative to act on them. Free account setup, no credit card required.