estimate-before-build

"It'll take a few minutes" is where scope creep pays rent. The agent starts editing, the request turns out to be three times the size anyone imagined, and stopping now looks like failure. The expensive failure is never the wrong estimate — it is the unmade decision. estimate-before-build turns that silent slide into a choice: this scope for this budget, a smaller scope, or a timeboxed spike first.

The core mechanism replaces hours with four size bands (S / M / L / XL). Hours invite false precision and calendar promises; bands communicate shape. For M and above you must name the uncertainty driver — risk lives where the unknown lives, and that is where a spike belongs. The estimate does not pick for you. It puts full scope, a reduced scope that fits the budget, and a timeboxed probe on the table so the decision is yours to make.

Record the estimate, then compare it against actuals at the finish line, task by task. That reconciliation is the only mechanism that makes the next estimate better. Skip it and calibration never happens.

Built for engineers and teams driving AI agents on coding or build tasks who want the scope decision laid bare before the first line of code is written.