The Evaluator

This is the human-held method every change to the seed is scored against. The seed applies it to the companies it runs; it never applies it to itself. Whoever holds this document holds the one role that does not transfer.

What it measures, for now

Article 2 defines fitness as durable problem-solving across genuinely different domains, per unit of energy, materials, and time, read from the domain the seed is weakest in. Turning transfer and felt outcome into clean numbers is still open research, so this version does not measure fitness directly. It asks three questions that stand in for it while the seed runs one company in few domains. Once the seed runs in several genuinely different domains, this method must be amended to score across them.

When scoring happens

A change is scored once its outcome can be seen, and scored again when a later outcome contradicts the earlier verdict. Every verdict carries its date. An unscored change is marked as unscored, so the debt is visible without blocking anything.

The three questions

Each change is scored on three questions. Each answer is yes, partly, or no, with one line of evidence a stranger could check.

1. Did it solve the problem it claimed? Judged by the felt outcome its proposal stated, never by effort spent or code shipped. If the proposal said “planning takes minutes, not an evening”, the evidence is how long planning takes. Partly means the outcome is visible but smaller than the proposal stated.
2. Did it move the handover? The number of decisions people make that are not verdicts fell, or held while capability grew. A change that adds capability but also adds a new human chore scores no here, whatever else it does. Partly means decisions moved to the seed in some cases and not in others.
3. What did it cost? Hours and money today; energy and computing time once they are metered. Cost does not fail a change on its own; it is the denominator the first two answers are read against.

Verdicts

A verdict is the three answers, their evidence, and its date, recorded next to the change it judges. A later verdict on the same change does not erase an earlier one. Both stand, and the latest governs promotion.

The one-way gate

A practice or component is promoted to stable only when question one and question two are both yes and the cost is justified in the latest verdict. Convenience, speed, or commercial success can nominate a candidate; they can never promote it. When the questions disagree, nothing is promoted; the disagreement is recorded and stands as a stop, not something to average away.

Amendment

This method changes only by pull request to this file, stating the problem, the evidence the current method failed, the replacement, and the failure modes the replacement is expected to carry. Verdicts change because reality does; the method changes rarely, and never silently.