SynCraft dataset for few-shot reasoning and unsolved molecules

syncraft_data_release_v1.tar.gz - `cliff_pair_corpus/cliff_pairs.json` — (source, edits, reasoning) tuples used as the in-context exemplar pool for RAG inference. - `test_sets/` — three unsynthesizable molecule pools - `multi_llm_inference/` — eight JSONL files of raw inference outputs from four LLM backends (Gemini-2.5-Pro, GPT-5.2, DeepSeek-V4-Pro, DeepSeek-V4-Flash) on both source pools (~30,000 inference records). - `plip_retention/` — per-target binding-mode retention tables: 191 RIPK1 (PDB 7YDX) pose pairs and 189 Mpro (PDB 7L12) pose pairs, plus per-source Vina records. - `mpro_skill_demo/` — end-to-end agent-skill rescue workflow trace on 47 TargetDiff-generated SARS-CoV-2 Mpro candidates: per-mode edit outputs, retro-validated analogs, and paired source/edit Vina records.