They tested it on working doctors. It held at six months.

Back in issue 03 the honest catch was: that evidence came from students. Here is the professional version. A randomized controlled trial put practicing pediatrics and ER residents through topics taught two ways: some restudied, some tested and spaced. On a delayed exam six months later, the tested-and-spaced topics won by a significant margin. The study.

These were busy experts under real clinical load, not undergraduates with time to spare. The finding survives the move from classroom to working professional, which is the move you care about for your team.

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

Expertise is not a credential your senior engineers earned once. It is a live thing that fades without use, the same as anyone else's. The senior who has not touched the auth system in eight months has a weaker model of it than the title suggests, and the org chart will not warn you. So when you staff next quarter, you are leaning on a seniority that your senior's idle systems no longer fully back: the title is current, the model has quietly decayed. The defense is the same as for the new hire, pull the knowledge back out briefly and often. Seniority buys a slower decay, not an exemption from it.

One for your next review:

A user updates their profile, the save succeeds, and the next screen still shows the old name. Refresh a few seconds later and it is correct. An engineer says it is a caching bug and starts clearing caches.

The cache-clearing will start fast; what matters is whether anyone interrupts it with "read replica" first. The self-healing-in-seconds behavior is the clue, and reading it means asking whether the write hit the primary while the read came off a replica still catching up. Here is the part that is useful to you: whoever reads that clue is on your real map of who still understands the data layer, and that map is rarely the one the org chart draws.