{"name":"hypoarena","repo":"OpSafari/hypoarena","url":"https://midorreal.com/project/opsafari-hypoarena","repository":"https://github.com/OpSafari/hypoarena","description":"Scientific hypothesis-discovery workbench: grounded hypothesis-evidence graphs, synthetic literature with planted causal chains, generate-debate-evolve loops over pluggable offline adapters, Elo tournaments recovering planted skill order, paraphrase dedup, Bayesian evidence accumulation, reproducible reports. NumPy core, CPU-only torch extra.","facts":{"license":{"value":"MIT","source":"https://github.com/OpSafari/hypoarena","checked":"22 Nov 2026"},"language":{"value":"Python","source":"https://github.com/OpSafari/hypoarena","checked":"22 Nov 2026"},"last_release":{"value":"v0.1.4, 9 Oct 2026","source":"https://github.com/OpSafari/hypoarena/releases","checked":"22 Nov 2026"},"releases":{"value":"1","source":"https://github.com/OpSafari/hypoarena/releases","checked":"21 Nov 2026"},"contributors":{"value":"1","source":"https://github.com/OpSafari/hypoarena/graphs/contributors","checked":"21 Nov 2026"},"archived":{"value":"no","source":"https://github.com/OpSafari/hypoarena","checked":"22 Nov 2026"}},"stars":525.0,"writeup":{"what_it_is":"A fully offline scientific hypothesis-discovery workbench that decomposes AI co-scientist pipelines into testable components. It generates synthetic literature with planted causal chains, runs generate-debate-evolve loops, verifies grounding, deduplicates paraphrases, ranks hypotheses via Elo tournaments, and accumulates Bayesian evidence.","audience":"Researchers studying hypothesis discovery and AI co-scientist mechanisms.","claims":[{"kind":"specific","claim":"Depends only on NumPy at runtime; PyTorch is optional","excerpt":"At runtime the package depends only on NumPy. PyTorch (CPU-only) is an optional extra","status":"not_checked"},{"kind":"specific","claim":"Produces byte-identical artifacts across platforms and Python versions","excerpt":"run outputs — and the golden digests pinned over them — are byte-identical across platforms and interpreter versions","status":"not_checked"},{"kind":"specific","claim":"Includes deduplication using TF-IDF, MinHash LSH, and n-gram Jaccard","excerpt":"TF-IDF cosine, and MinHash LSH with documented false-positive/false-negative behavior","status":"not_checked"},{"kind":"specific","claim":"Recovers planted causal chains in synthetic corpora","excerpt":"planted links: 6  recovered: 6  rate: 1.0","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-10-11"},"why_now":null,"owner_supplied":null,"score":{"verdict":"mid","hype":0,"reality":8,"gap":-8,"momentum":-0.3,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-10-09T16:34:27+00:00","text":"Latest release v0.1.4","source":"https://github.com/OpSafari/hypoarena/releases"},{"date":"2026-09-30T03:35:45+00:00","text":"Repository created","source":"https://github.com/OpSafari/hypoarena"}],"cite":"hypoarena is a GitHub project at https://github.com/OpSafari/hypoarena written mainly in Python, described there as \"Scientific hypothesis-discovery workbench: grounded hypothesis-evidence graphs, synthetic literature with planted causal chains, generate-debate-evolve loops over pluggable offline adapters, Elo tournaments recovering planted skill order, paraphrase dedup, Bayesian evidence accumulation, reproducible reports. NumPy core, CPU-only torch extra\".","attribution":"Data from Mid or Real, https://midorreal.com/project/opsafari-hypoarena, checked 22 November 2026. Attribution with a link is required.","still_real":null}