Analyses that did not produce a number, and why.
Per-instance-set batching cost, ascending by set id.
Per-model geometry cost, ascending by model id.
Owners of each metric's cost, ascending by owner id.
The report bounds this into a short dominant-contributor list; the inventory keeps the complete attribution so a consumer can sum it back to the total.
Resident upload variants, environment derivatives and runtime targets/lookups, sorted by asset, variant and owner keys; several rows may share one asset.
Per-metric counts, bounds or estimates; null means no admitted measurement for that metric.
Inventory format.
Compiled-artifact preflight counts, submission bounds and resource estimates for render-budget evaluation.
The inventory is separate from budget verdict and live renderer observations. Some metrics count exact populations, while others bound submissions or estimate device allocation under their stated input premises. auditAutoMovieRenderObservation can compare available observations with report bounds, but currently has only test callers in this repository; it is not a wired capture audit.
The three per-kind arrays below are not a complete decomposition of the totals. A cloth panel, a planting cluster and a water surface are none of a model, a texture or an instance set, so their cost reaches a reader through totals, owners and gaps alone; summing the arrays would come up short by exactly the things a room is furnished with.