Every score is computed per (carer, participant) over a 28-day window — never per shift. A single data point is neutral, not accusatory. A 🔴 flag needs both a high score and signals firing across ≥2 independent families — that convergence rule is why Bluey scores ~0 while Reece hits 0.71. Suppressions are logged: the system can prove it actively cleared an honest carer, not just failed to flag them.