Not every observation carries the same weight. Evidence strength is a property of the individual signal, before anything is aggregated.
What makes a signal strong
- Directness. Something stated or shown outright beats something implied.
- Specificity. A concrete detail beats a vague impression.
- Repetition. A pattern across a conversation beats a single instance — though repeated instances of the _same_ event are de-duplicated so one incident cannot be counted several times.
- Category. Safety-category signals outrank communication, reciprocity, and investment signals. If a safety signal is present, it dominates the read even when everything else looks good.
Why the weights are fixed. Weights live in a versioned table, not in the model's discretion. This is deliberate: it means two identical inputs produce two identical outputs, and any change in behaviour is traceable to a version bump rather than to model drift.
What you see. In a Decision Brief, each contributing signal is listed with the observation behind it, so you can check the reasoning rather than take it on faith. If you disagree with a signal, you can see exactly which piece of your material produced it.