An estimate is not a commitment

Jonas, a backend engineer at a logistics company, is asked how long order-splitting will take. He answers honestly: "Six to ten weeks — and I haven't looked at the migration yet, so low confidence." His team lead writes "~8 weeks" in the planning notes. The department review slide says "Order splitting — end of Q3." Three weeks later, a sales call mentions September. Nobody lied at any step. Somewhere between Jonas and the customer, a range with a warning label became a date with a witness.

The hardening is structural, not moral

It is tempting to read this as someone's dishonesty, but every actor behaved reasonably inside their format. Planning notes want one number, so the range became its midpoint. A review slide wants a quarter, so the number became a date. A customer call wants something to say, so the date became a promise. Each artefact on the path has a smaller field than the one before, and uncertainty is the first thing each format drops — because uncertainty is precisely the part that does not fit in the cell. The information lost on the way up is exactly the information that mattered: how sure anyone was.

Two different speech acts

An estimate is a claim about size: this looks like six to ten weeks, and here is how sure I am. Recorded properly, it is a figure in FTE-months with a confidence level attached — a two on a five-point scale says "I have not looked at the migration" in a form that survives being written down and passed along.

A commitment is a promise: a named owner, a scope, and a date someone else may now build on. The difference between the two is not precision — it is accountability. An estimate can be wrong and the estimator blameless; that is what "estimate" means. A missed commitment needs an explanation from its owner. Blur the two and you get the worst of both: nobody ever promised September, yet somebody is late. (Scrum fought the same battle at ticket level and renamed the sprint commitment a forecast — right at that level, wrong as an organisational conclusion.)

The recorded act that stops the hardening

The fix is not better slide discipline; it is a state. Between "the team estimated this" and "we are delivering this" there has to be a separate, recorded act of promising — performed by a named owner, at a specific moment, on the record. Until that act happens, the demand has estimates and no date, and anyone quoting a date upstairs is quoting something the record can contradict. After it happens, there is exactly one date, and it is the one the owner put their name to.

The act also has to be version-guarded: if the demand changed since the owner last looked — the estimate revised, a dependency added — a commit against the stale version is rejected, and the owner re-reads before promising. That sounds bureaucratic until you consider what it prevents: a promise made against last month's facts, which is not the promise the owner thinks they are making.

What changes for the engineer

The under-appreciated effect is on Jonas. When the estimate is treated as the commitment, giving an honest range is a career risk, so engineers learn to pad — and padded figures are still quoted as dates, so the padding buys safety for nobody while degrading every number in the system. Separate the acts and the range becomes safe to say: Jonas's "six to ten, low confidence" is input to a decision, not a promise extracted from him mid-sentence. Teams that trust the separation give better estimates within a quarter, because the incentive to hedge is gone.

The question that proves the record

Mid-September, order-splitting is not done, and a director asks the useful question: when did we commit to September? With the recorded act in place, the answer is checkable — we didn't. The estimate said six to ten weeks at low confidence; the commitment, made three weeks later after the migration was actually examined, says October 20 and carries the team lead's name. The director is not delighted, but she is arguing with a record instead of a rumour — and the next sales call quotes the commitment, because now there is one to quote.

An estimate that travels with its confidence, and a promise that exists only as a deliberate act: that is most of what a capacity practice has to get right. The rest of the argument — what to measure, in what unit, and what to refuse to measure — lives in the capacity overview.