The input-modeling pillarDistribution, trace, bootstrap and fitted sources all feed the same input field. Distributions are drawn in the trial; traces are replayed; bootstrap and fitted rows are generated once per replication and shared by every design point. inputs.c gives each input its own random stream and applies its exhaustion policy. Four interchangeable sources inputs.dist.* inputs.dist.* distribution: drawn inside the trial, in C, on the input's own stream inputs.trace(data) inputs.trace(data) recorded: replayed exactly as recorded, shared read-only by all trials inputs.bootstrap.* inputs.bootstrap.* resample: rows generated in Python once per replication inputs.fitted.* inputs.fitted.* fitted model: rows generated in Python once per replication row buffers row buffers read-only memory one row per (replication, source) shared by every design point (CRN) regenerated 2× longer on exhaustion inputs.c inputs.c one InputSlot per input field in the trial record kind · policy · parameters stream state · data · cursor series step · origin · history own random stream exhaustion policy model code model code x.next() d.now() d.at(t) identical for every source parameters data* 1 C call on exhaustion on exhaustion fail · wrap · end_trial · extend Swap or sweep sources without touching the model or recompiling. Click a box for details.