experiment EX-VE-TYP-002

Crowding-Responsive Typography

EX-VE-TYP-002: Crowding-Responsive Typography

Question

Does directly measured visual crowding predict benefit from spacing interventions better than dyslexia diagnosis alone?

Rival models

Model A: diagnosis-only

spacing benefit ~ dyslexia diagnosis

Model B: mechanism-aware

spacing benefit ~ measured crowding + diagnosis + interaction terms

The experiment is successful only if the rival models are compared prospectively.

Participant characterization

Record separately:

  • dyslexia diagnosis, if present;
  • age;
  • corrected visual acuity;
  • reading proficiency;
  • measured crowding;
  • baseline reading speed and accuracy;
  • relevant visual conditions.

Do not use diagnosis as a proxy for crowding.

Crowding measure

Use a psychophysical task with isolated versus flanked character recognition and a preregistered threshold or spacing metric. Preserve raw condition metadata.

Typography manipulation

Compare a bounded set of:

  • letter spacing;
  • word spacing;
  • line spacing;

including a baseline, modest expansion, and settings near the point where grouping is expected to degrade.

Avoid assuming "more spacing" is better.

Outcomes

  • reading speed;
  • word/character errors;
  • comprehension;
  • line-loss or rereading indicators;
  • subjective effort.

Falsification

HY-VE-PE-001 is supported only if measured crowding materially improves prediction of spacing benefit over diagnosis-only baselines on held-out participants or preregistered cross-validation.

It is weakened if diagnosis predicts equally well, crowding is unstable, or the spacing manipulation does not interact with crowding.

Important negative result

If expanded spacing helps high-crowding readers but harms low-crowding readers, that is strong evidence for Perceptual Envelope theory and against a universal accessibility preset.

Ethics

This is human-subject research. Obtain appropriate review/consent before execution. The public repository stores protocol, analysis code, and de-identified aggregates, not raw identifiable data.