The board will ask if AI is hollowing out the team.

At some point the question arrives from above: we spent on AI, is it making the team better or quietly worse? Velocity charts cannot answer it. Issue 01 named why: the speed is real, that was never in doubt, but it says nothing about how long the team can keep shipping this well on skills it has stopped practicing. A dashboard climbing tells you they are fast today, not whether they could still reason through what they shipped. You need a second axis. One of the field's largest meta-analyses, 272 controlled comparisons across 118 articles, found retrieval practice beats restudying by about half a standard deviation and beats no practice by nearly a full one. The study.

That is the shape of an answer to the board: not "we shipped faster," which is true and beside the point, but "we can show the team's ability to reason about its systems held or grew while we adopted AI." Capability trend alongside velocity trend.

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

This is the whole series in one line: the debt is invisible to every dashboard you currently run, because they all measure output, and output is exactly what AI inflates. The board-grade move is to measure the thing output hides, whether the people can still reason about what they ship. Do that and "are we adopting AI responsibly?" stops being a vibe and becomes a trend you can put on a slide. That is what a capability metric buys you that a velocity metric never will.

One for your next review:

A "claim your free credit" endpoint checks whether the user already claimed, and if not, grants the credit. Under normal use it is fine. During a promo, a handful of users end up with two credits.

The reassuring answer reaches for the database. Two requests can both clear the "already claimed?" read before either writes, so the fix that holds is a unique constraint or an atomic update, something that serializes the write itself. Reaching instead for one more check before the write is the answer to worry about, because it shrinks the window without ever closing it. That instinct, a database-level guarantee over one more hopeful if-statement, is the difference between a bug that is fixed and one that is merely rarer.