PROJECT · shipped

Blindfold Lab

An audio-guided practice platform for blindfolded perception: once a session starts, everything is voice, keyboard, or swipe.

Next.js · Supabase · audio pipeline

BLINDFOLD LAB
A real phenomenon

Learn to See
Without Your Eyes

Begin Training
Why it has to be right

Trust-based on purpose. It exists for the user's own practice, not to prove anything to anyone.

What it is

Blindfold Lab is a live practice platform for blindfolded perception. The app shows a color or a shape full-screen while you are blindfolded, and you practice perceiving it. Some people report learning to see this way, and the skill apparently responds to consistent practice. I built it to practice myself, made it available to others, and people started using it.

The novel part

You use the app while blindfolded, so the interface cannot assume eyes. Once a session starts, everything is audio: spoken guidance announces each trial, reads out your options ("press left arrow for red, or right arrow for blue"), and confirms what you selected, and all input is keyboard arrows or swipe gestures. There are six recorded voices with a text-to-speech fallback, a repeat key for the last prompt, and a guided breathing phase to settle in before the trials begin.

What I built

Three exercises: contrast detection (black or white), color recognition (red, blue, green, yellow), and shape identification (circle, square, triangle, plus). Sessions run 12 trials across three difficulty tiers, from two options at five seconds of exposure down to four options at two. The app records accuracy and reaction times per trial and suggests moving up a level at 80% accuracy, or down below 50%, measured over your last three sessions.

The trust model

There is no verification here, on purpose. Confirming that someone truly perceived a shape through a blindfold would require video access and would still rest on trusting they are not cheating some other way, so the app does not pretend to measure anything for anyone else. Results are practice logs for the person training. Cheating is not something we engineer against, because the only person a user could cheat is themselves.

What broke and what I learned

Shipping to strangers is a different sport from shipping to yourself. Audio behavior across devices, sessions abandoned mid-trial, users doing things no spec predicted: real users found every gap within weeks. It permanently raised my bar for what "done" means.

Status

Live and running, with a small but real user base.