The naming system contradicted its own premise

Naming a company, a product or a fund normally happens in a room. Sticky notes, a spreadsheet of three hundred candidates, a shortlist that narrows as people get tired, and whoever is most confident at four in the afternoon wins.

Six months later somebody asks why that one. Nobody can answer. Not because the decision was bad, but because nothing about it was written down in a form that survives the room.

So we built a system where it could be. Candidates generated seven ways, not one. Rules that reject outright instead of scoring low and letting something survive on the strength of everything else. Twelve independent measures, each computed by ordinary code, each carrying its own explanation. The expensive checks against the outside world run last and only on the leaders. A person ratifies; the system adopts nothing on its own.

It runs offline and replays exactly. Same brief, same result, a year later.

Then we ran a real engagement through it

The output contradicted the premise the whole thing rested on. Not a bug, not a bad input. The system, working correctly, produced a finding against the idea that had justified building it.

We changed the design.

Why this is the useful partA system built to test an idea is worth nothing if the idea is allowed to overrule it. That sounds obvious written down and it is not obvious at four in the afternoon, when the result is inconvenient and there is a plausible reason to re-run it with different settings. The reason we keep the replay guarantee is not elegance. It is that a result we dislike cannot be quietly re-rolled until it agrees with us. The same brief produces the same answer every time we ask.