What parity means, and what it doesn't

kirby-cost is validated against Hero Designer on 655 of 655 characters: every value the corpus compares matches, on every object and on every character total. That’s the headline number, and it’s accurate. It’s also easy to over-read, so it’s worth being precise about what it does and doesn’t claim.

It claims that for the characters in the corpus, the two engines agree. It does not claim that the corpus contains every shape a HERO 6E build can take. The corpus grew twice over the life of this project, and both times it grew because a character came along that exercised something the existing set hadn’t — a framework, a modifier combination, a template wrinkle. Each growth pushed the denominator up, and the percentage stayed at 100 only because each new case got fixed before it was allowed to count. A parity number without knowing how many characters it was measured against, and how they were chosen, isn’t something you can actually check — it’s just a number someone is asking you to trust.

The instrument matters as much as the sample. “Validated against Hero Designer” means validated against a specific oracle: a reference implementation that itself has been wrong and been corrected. For four months, that oracle silently ignored the template a character had declared and costed it as if no template were in play. Every comparison run during that window was comparing against a quietly broken reference, and every “match” from that period inherited the same blind spot. Finding it meant distrusting the oracle’s silence, not just the engine’s output — a mismatch is a signal something is wrong, but which side is wrong is a separate question every time.

So the honest form of the claim has three parts, not one: a percentage, a denominator, and an instrument. 655 of 655 is real and it’s the current state. It is also a number that has moved before, in both directions, as the corpus grew and as the oracle itself got fixed. Reporting it without the other two parts would be reporting less than what’s actually known.