What a score is

What a number here means, what it does not mean, and what you must not use it for.

A quality score is one number, 0 to 100, for one conversation. It is what the rules in force found, weighted by how much each failure matters, with an automatic fail able to hold the whole number down on its own. Every score opens onto the turns it came from, and that is the only reason to believe one.

What a score is not

Evidova produces an opinion, not a ruling. A score is evidence for a person to decide with: it is not a certification, an audit opinion, an accreditation, or legal, regulatory or professional advice, and the report you can export certifies nothing. It must never be the sole or deciding basis for a decision about a person, which covers hiring, pay, discipline and dismissal, and also credit, insurance, housing, education, healthcare and any legal right.

A score compares only with its own version

One version of the rules produced it. Change what we check and the next score stands on a different footing, so the version travels with the score and a trend is drawn only across scores from the same one.

Movement smaller than the noise is not movement

How much a score moves on its own when nothing about the conversation changed. A change smaller than this is not real movement.

So a score that moved two points is not news if two points is inside the measured movement for your workspace. The agreement page says where that figure is published.

A blank is not a zero

We have not measured this yet. The reason lives in the sentence next to it, not the label. A blank is never a zero.

The AI can be wrong

The scoring runs on large language models, whose answers can be incomplete, inconsistent between runs, or confidently wrong. Invented detail is a known trait of the technology and not a defect peculiar to ours. That is why a score is built to be opened rather than trusted, why failures are read again by other judges, and why the agreement figures are published where you can find them.