Five evidence groups — regime, momentum, structure, liquidity, event risk — each produce a 0-100 score. The confluence engine combines them with configurable weights, not a simple average, so a strong signal in a heavily-weighted family matters more than a weak signal in a lightly-weighted one.
If an evidence group's data isn't genuinely available (for example, no recent liquidity observation), it's excluded entirely and its weight is redistributed across the remaining groups — never defaulted to a neutral 50 or a fabricated number.
A hard event-risk blocker (major breaking news) can override everything else and force a BLOCKED status regardless of how strong the rest of the evidence looks. A counter-trend setup (proposing LONG in a BEAR regime, for instance) requires a meaningfully higher score before it's treated as a real candidate at all.