Trust comes from knowing when not to make a claim
Protu should never sound more certain than its evidence allows. Confidence describes the strength of the evidence behind a conclusion, not the likelihood that a decision will succeed. Where the evidence is strong, Protu can say more. Where it is weak or conflicting, it says less. Knowing the difference is fundamental to how we build it.
Confidence is not prediction
Confidence describes how strongly the evidence supports a conclusion. It does not predict whether a hire, promotion or organisational change will succeed. Outcomes depend on people, timing and circumstances that no behavioural model can fully observe. A confident read can still lead to a poor outcome; a cautious one can still prove valuable. We keep those ideas deliberately separate.
A system that is never unsure is a system that is quietly guessing.
Evidence should match the consequence
Not every decision carries the same consequence if it is wrong. Some changes are relatively easy to revisit; others are costly, disruptive or difficult to reverse. Protu demands stronger evidence as the consequence of being wrong increases. A promising possibility may be worth exploring with limited evidence. A consequential recommendation has to earn more.
Agreement strengthens confidence. Disagreement is evidence too
No conclusion should rest on a single signal. Protu looks for support across multiple behavioural signals and perspectives. Where the evidence converges, confidence can grow. Where it conflicts, that disagreement remains part of the evidence rather than being averaged away. The aim isn’t to force a clear answer. It’s to represent the evidence as it is.
The reasoning should be reproducible
A conclusion should remain connected to the evidence and reasoning that produced it. Protu versions that reasoning so the same evidence, interpreted by the same version of the model, can be reproduced and inspected later. Reproducibility doesn’t make a conclusion infallible. It makes it possible to understand how Protu got there.
We only claim what we can support
There is a difference between something the model can infer, something we have observed repeatedly and something that has been independently validated. We don’t treat those as interchangeable. Claims about Protu should move with the evidence behind them, and where that evidence has not yet been established, we should say so.
A system that knows what it does not know is the only kind worth trusting with a decision about people.
Keep reading