Docs: architecture.md notes exact-time sampling and the layer cache for Particle World (#1373)

docs/architecture.md

@@ -371,7 +371,10 @@ GPU, like `SimCache`; the CPU simulation stays the oracle (tests compare ids and
 CC Particle World samples its Birth Rate at every step (`EffectHost::param_at`, #1373), so a
 keyed rate only decides the births of its own time; the rate is left out of the simulation key,
 and a checkpoint is resumed only with the same rate history (the CPU state's `hist`, the GPU
-checkpoint's births).
+checkpoint's births). The rate's keyframes are read at the step's exact layer time
+(`EvalCtx::value_at_layer_time`; mapping it to comp time and back rounds under a fractional
+stretch), a rate without keyframes or an expression isn't sampled, and a Particle World layer with
+expressions isn't layer cached (its key can't see what an expression returned at earlier times).
 `effectcraft-cli bench --gpu` reports CPU vs GPU ms/frame, speed-up, Auto's ms/frame, pixel
 agreement and the steady-state upload / readback MB per frame for every comp plus an
 adjustment-layer comp.