Working notes
Longer pieces on why the app is built the way it is — the measurement problems, the design decisions that cost something, and the defects found along the way. Written for people who want the argument rather than the summary.
Five ways the puzzles leaked their own answers
An automated audit measures how much each surface feature of a problem reveals about its answer, and fails the build past 0.05 bits. It has caught five real leaks — including one that printed the answer on the item.
One dial and one number
Dual n-back gives you a single difficulty knob and a single metric. Matching a relation rather than an item changes what the task demands — a structure to update instead of a list to hold.
The test we hope doesn't improve
Tanren Sensei measures one ability it deliberately never trains. Without that comparison, an improvement has four explanations and no way to choose between them.
There is no newsletter and nothing to subscribe to.An RSS feed exists, which asks nothing of you and tells us nothing about you.