NAMEFRAME Commercial Preview Apply for Pilot
Documentation

Distributions

How many objects, and where inside the zone. Two different questions with two different mechanisms, and an execution boundary between them worth understanding.

Numeric distributions#

Declarative, for quantities:

uniform       flat between two bounds
normal        mean and standard deviation
lognormal     skewed, for counts with a long tail
triangular    minimum, mode, maximum
weighted      explicit values with weights
curve         an authored response curve

Writing 4,24 for a count means "sample uniformly between 4 and 24 for every frame", not "place exactly twelve". A dataset whose object density is identical in every frame has told the model that density is a constant.

Where a distribution is actually sampled#

This is the distinction the execution model page exists for, and it matters here more than anywhere else.

The native spawner supports adopted distribution semantics for supported population quantities. Other recipe fields are sampled by the planner into deterministic per-frame values before the runtime sees them.

a planner-sampled value
    ≠
every runtime subsystem receiving an untouched distribution object

Both are deterministic and both are reproducible. They are not the same execution path, and the generated field reference exposes the owner for each field so you can tell which one you are configuring.

Spatial profiles#

How placements are spread inside the zone, once the count is decided:

ProfileSpread
legacyThe original shuffle behaviour.
uniform_areaEven across the zone's area.
poisson_diskEven with a minimum separation, so nothing clumps.
gaussianConcentrated around the zone centre.
center_biasedDenser in the middle, thinner at the edges.
edge_biasedThe reverse: hugging the boundary.
ringAround an annulus rather than filling the volume.
Poisson, precisely

The Poisson profile is deterministic bounded sampling with a minimum separation. It is not a maximal Bridson implementation, and describing it as one would be claiming a property of the output that has not been implemented.

Choosing one#

Match the profile to the real thing you are imitating. Pedestrians in a plaza are not uniform; they are clustered near entrances and sparse in the middle, which is center_biased inverted or a set of cluster rules. Stock on a warehouse floor is closer to grid with jitter. Uniform is the right answer surprisingly rarely, and it is the default people reach for first.