LOD for off-screen agents: keep resolution, drop cadence — and scale a coarse tick to its span

Design for keeping side-characters alive off-screen in a sim without simulating everyone every tick. Split two things people conflate: RESOLUTION (is this a named, invested entity vs anonymous background) and CADENCE (how often it is simulated). A character the player invested in keeps elevated resolution (state persists, does not vanish to the aggregate) but drops to a COARSE heartbeat (e.g. act every 8 ticks, not daily) when off-focus. Cheap because the invested cast is small; LOD spares the anonymous millions. Gate the WHOLE agency loop by focus (if resolution < WATCHED and day % K != 0: skip), not just the expensive perception pass — the decision loop is the real cost. Measured ~40% faster at 2000 agents. THE GOTCHA that makes off-screen characters stagnate: a coarse tick applies ONE tick of development but STANDS FOR K ticks of elapsed time, while time-based decay (belief confidence) keeps running every day — so fade beats the sparse reinforcement and the character drifts down instead of growing. Fix is the multi-scale rule: a coarse level must model its whole SPAN, not a snapshot — scale a coarse tick effects by the elapsed period. Same principle a galaxy/sector actor needs (its tick represents an era).