Douvras Labs datasets and models
Collection
Synthetic Douvras Labs artifacts. Cards list license, splits, baseline and limits. M01 weights and D20 stay unpublished. β’ 342 items β’ Updated
Transparent baseline for D22. It uses a bounded deterministic beam search and verifier feedback, recording every score and rejection. It is a deterministic search procedure, not a neural model and not a claim of mathematical discovery. It was evaluated on the six synthetic problem instances in the dataset; the toy optimum is known to be four.