Atlas actions

Each scenario starts with a 10% base rate — the prior probability before any current evidence is considered. As historical observations accumulate, these will be replaced with empirically-derived per-scenario rates.

Each of the scenario's 6–8 indicator conditions is scored −1 to +1. The score reflects both direction (does the value confirm the thesis?) and extremity (how unusual is this reading in the historical distribution?). Scores are weighted and averaged.

Signal strength becomes an exponential multiplier on the prior: 4 ^ strength. Full confirmation (+1) quadruples the prior. Full contradiction (−1) quarters it. Neutral evidence (0) leaves it unchanged.

Correlated scenarios share indicators. Without adjustment, shared signals would inflate both scores simultaneously. A two-pass adjustment dampens negatively-correlated pairs and slightly boosts positively-correlated ones — preventing double-counting.

All nine adjusted scores are normalised so probabilities sum to exactly 100%, preserving relative ordering. The result is a probability distribution across all scenarios simultaneously — not a single classification.