Brainback
researchdata

The attempt-gate in numbers

What changes when the AI physically can't respond until you've tried.

Jul 25, 20265 min readBrainback Research
abstract

A short data note on what happens to session behavior when the attempt-gate is enforced server-side.

During beta we ran a small internal experiment. We loosened the attempt-gate — reduced the minimum attempt length from 20 characters to 5 — for a subset of users. The change was silent. Users didn't know.

Two weeks later, the pattern was clear. Users in the loose-gate group averaged 40% more messages per session but 30% shorter session length. They also averaged fewer hints requested (they didn't need them), fewer explain-backs completed (they didn't finish the concept), and lower Brain scores at the following weekly compute.

In other words: they were getting more answers, doing less work, and learning less. The gate was doing exactly what it was designed to do.

We restored the 20-character minimum for all users and haven't loosened it since. The strong version of this rule is a good one: the AI cannot help you until you have shown you have started.


Written by Brainback Research. Published Jul 25, 2026. Filed under Data.

if this landed, share it with the friend who’s outsourcing their brain

if the research landed

Put it into practice this semester.

Every essay here is downstream of a design decision inside Brainback. Come see the design.