Same class, same material. 13 points apart.

The last two issues were the cost. This one is the way out, and it is almost boring how well it works. In a five-year classroom study (N=1,400+), the only change was a handful of brief, low-stakes quizzes dropped into normal lessons. The quizzed material scored 94% on the real exams. The same material, taught the same way but not quizzed, scored 81%. The gap held into the semester finals, months later. The study.

These were students, not engineers, so hold the exact figures gently. What transfers is the mechanism: pulling an answer out of your own head, briefly and often, is what makes it stick. Re-reading the doc does not. (The professional version of this, run on working doctors, is coming in a later issue.)

Comprehension debt: the gap between what gets shipped and what the team can still reason about.

You cannot buy your way out of this with more content. A bigger course library is more re-reading, and re-reading is the thing that does not work. The lever is small and repeated, not large and one-time. And the edge you are budgeting for is the one your estimates quietly assume the team still has: the day that assumption is wrong, you find out inside a missed forecast, not a training report. So spend on the practice that keeps the edge, not on a one-time push that has decayed by the next quarter.

One for your next review:

A nightly job calls a payment provider. It sometimes times out, so an engineer adds a retry: on failure, call again. The provider later reports some customers were charged twice.

The fix is not the point here; the reasoning is. A retry is safe only if doing the thing twice equals doing it once. Some engineers have a word for that, idempotent, but what matters is whether they ask what the provider saw before the timeout, knowing a charge can land even when the response never arrives. Without that question, every flaky call in the codebase is a double-charge waiting for enough traffic to find it.