Each step, every Active-tier entity scores its candidates — intrinsic behaviors (self-directed) plus in-range advertisements (world offers) — using:

score = need × distance × trait × mood × memory × role × cycle × relationship
        + base_score          (intrinsic behaviors only)
        + HYSTERESIS_BONUS    (0.25 when repeating the last action)

The winner is selected by argmax (temperature = 0) or Boltzmann soft-max (temperature > 0). The engine calls the registered callback and applies the returned patch to world state.

Factor	Range	What it does
`need`	[0, 1]	Sum of urgency × magnitude per matching need
`distance`	[0, 1]	`1 - dist/range` — closer providers score higher
`trait`	[0.1, 3.0]	Product of per-need personality multipliers
`mood`	[0.3, 3.0]	Need and axis multipliers scaled by mood intensity
`memory`	[0.15, 1.5]	`1 + outcome × 0.5` from learned Q-table
`role`	configurable	Match bonus (1.5×) or mismatch penalty (0.7×)
`cycle`	[0.5, 1.5]	Day/night and seasonal affinity from need definition
`relationship`	[0.5, 2.0]	`1 + (trust − rancor) / 100 × 0.5`