PosterHarness reframes scientific poster generation as an auditable instruction-following benchmark. Its placeholder-first contract assigns models responsibility for typography, layout, reading path, color, and background, while forbidding them from generating data-bearing figures. A deterministic compositor inserts source-paper figures into labeled empty regions. In a pilot covering 12 papers, six from HEP and six AI/ML-adjacent, a counterfactual probe reduced VLM-counted synthesized figures from 34 to 0 across three papers. The paper also defines failure categories and compares the harness with Paper2Poster, reporting a trade-off rather than superiority.
No heat snapshots are available in the last 24 hours.