Public Visual Documentation

Trace publishes two intentionally separate visual collections.

The release gallery manifest describes 22 examples rendered from the current public package: two examples from each visual domain. Run python scripts/generate_release_gallery.py to regenerate that gallery.

The paper domain-montage manifest describes the 11 exact reviewed PNGs selected for the TRACE paper: one per domain, in the paper's 4/4/3 order. Those PNGs are frozen, content-addressed documentation inputs. The manifest records each task, query, instance seed, source index, source data/image hashes, path, and dimensions without relying on an internal checkout or machine path. The README-facing composite places those lettered panels into the first 11 cells of a 4-by-3 grid and uses an unlettered Trace brand card in the final cell. The manifest records the brand inputs separately; the card is not a paper panel or a twelfth domain. Run python scripts/generate_paper_domain_montage.py to validate the inputs and regenerate the README-facing montage.

Keeping these collections separate is deliberate. The current task package can continue to become more deterministic or portable without silently changing the exact examples shown in the paper.