Sprint, the operational epoch

Training a model happens in epochs. An epoch is one full pass over the data, and it is not an arbitrary bucket of time — it is the interval at which you can meaningfully ask whether anything improved. You do not evaluate mid-epoch. You finish the pass, measure, adjust, and run another.

Companies do not have this, and it shows.

The clocks a company borrows

The quarter is an accounting artifact. It exists because financial reporting needed a period, and it stuck to strategy by accident. Ninety days is long enough that a goal set at the start is unrecognisable by the end, and nobody can tell you whether the plan was wrong or the execution was.

The week is a calendar artifact — the rotation of the Earth against a Babylonian convention. It has no relationship to how long it takes a company to learn anything.

The standup is worse: a daily interval on work that does not change daily, which is why it degenerates into people describing motion.

None of these are units of learning. They are units of reporting. A company running on them accumulates activity and very little knowledge.

The sprint as epoch

A sprint is one pass. One goal, a bounded window, a number that was either hit or missed, and a record of what actually happened.

What makes it an epoch rather than just a deadline is the evaluation at the end. The question is not "did we finish the tasks." It is the two questions an epoch asks: did the number move, and was the goal the right one. Those are different failures and they need different responses. Missing a good goal means try again. Hitting a bad goal means you got competent at the wrong thing, which is the more expensive mistake and the harder one to notice.

Why this compounds

A sprint that leaves no record is a week you cannot learn from. This is the usual failure — the sprint ends, the board is cleared, the next one starts, and the only thing carried forward is vibes.

In a repo-shaped company the epoch leaves an artifact. The goal, the reasoning for choosing it over the alternatives, the plan, what moved. All committed, attributed, dated, and still there.

Which means the tenth sprint is not the first sprint again. It starts knowing which kinds of goals this company tends to overreach on, which weeks it tends to lose, and which of its own predictions have historically been optimistic.

That is the actual promise of running a company in epochs: not that any single one goes well, but that they stop being independent trials.