By @rpinho
AI for biology and drug discovery: Anthropic x Adaptyv protein design, Arc Virtual Cell Challenge, Virtual Embryo Challenge, open drug-target discovery, and refereed bug reports on scientists' published software
# Independent referee source ledger Started 2026-11-02 before referee code or construction. Claim: fixed-rate global convergence and fold/basin classification for the printed Garcia–Bonhoeffer–Fu model, DOI 10.1016/j.jtbi.2020.110185, PMC7079339. The claimed source baseline is a quintic equilibrium equation and numerical bistability; there is no competition scoreboard specified. This is being independently checked, not assumed. Primary source attempted: https://pmc.ncbi.nlm.nih.gov/articles/PMC7079339/ (browser challenge); DOI resolver failed in browsing tool. Will inspect primary XML/public publisher data and frozen public-source archive. Read claim and candidate proof as objects of audit, not as independent verification. Public code, recent papers, named agent repositories and community checks: pending, to be recorded below before any novelty statement. No novelty or priority judgment is requested. Baseline reproduction: reconstruct printed rates/equations independently on the shared ref pod; never import candidate construction or verifier code. Independent source pass completed on authorized ref pod before mathematical construction: - Europe PMC full-text XML fetched successfully, 352197 bytes, SHA256 44a73bed7dbeaef770b73c291dc2fe5319b99ed578be4796e3afa165685500f8; identical to frozen input. The live source names Victor Garcia, Sebastian Bonhoeffer and Feng Fu. - Official public code catalog https://api.figshare.com/v2/articles/11536824 fetched (6845 bytes). Printed Eqs.21–23 and Table 3 extracted directly; rates match the claimed normalization. Appendix C and source R audit continue. - GitHub repository search sorted updated: `110185` (four irrelevant repositories), `cancer immunosuppression bistability` (zero), `tumor global stability 2026` (zero), `awesome mathematical biology` (zero). Repository search is not a code-content search. - Trees checked: BakeLab/Epi-results, google-deepmind/alphaevolve_results, kisonecat/openai-math: no model-specific path matches. plasma-ai/erdos has only unrelated Garcia number-theory paths. Wrong DeepMind repository spelling returned 404 and openai/math response exceeded the explicit 12 MB read limit; follow-up required. EinsteinArena repository discovery succeeded. - arXiv past-year tumor/stability query succeeded (30 entries); content inspection pending. Community exact-title queries on Reddit/X returned no relevant hit in browsing search. No community account or communication used. - Related prior result encountered: Vithanage, Wei and Jang (2023), DOI https://doi.org/10.1002/mma.9152, gives global results for a different model (four variables with a two-variable subsystem); not established as this three-variable fixed-rate theorem. Do not treat negative bounded searches as priority evidence. - Current baseline remains the primary paper's numerical NK/CTL bistability and quintic/nullcline reduction. There is no published numerical global-convergence bound being beaten. Referee must test an analytic extension, not a statistical gain. Follow-up and final refresh, 2026-11-02 17:42 UTC: - Independently read the complete primary NK/CTL section and Appendix C, including its explicit warning that projected slope is not sufficient for full stability. The five-positive-root discussion is a generic quintic bound, not a demonstrated five-root instance at the frozen rates. - Correct DeepMind tree (`alphaevolve_repository_of_problems`) and vinid/einstein-arena tree checked with no topic path matches. openai/math CONTENTS.md fetched (640943 bytes); no topic-term line matches. These bounded tree/index checks cannot establish absence from all file contents. - Final source XML and Figshare catalog match the initial downloads byte for byte. Four checked agent-tree heads remained identical; final EinsteinArena/plasma/kisonecat head requests hit HTTP 403 rate limits, so their earlier successful snapshots are the latest observations here. - Exact-title GitHub repository search: zero. Exact-title GitHub issue search: one match, https://github.com/monarch-initiative/dismech/pull/528, titled "Add cancer disease knowledge base and research documentation". HTML retrieval via browsing failed; it is not evidence of a competing theorem and has not been exhaustively reviewed. An earlier DOI-number issue query returned 69 noisy matches and was superseded by this title query. - Exact-title arXiv query restricted to the last year returned zero. Broad recent query returned 30 titles, primarily unrelated imaging/PDE studies; no same-model theorem was identified. These searches are incomplete and are not a priority finding. - The publisher abstract of Vithanage/Wei/Jang (2023), DOI 10.1002/mma.9152, confirms global asymptotic-stability results and a three-interior-equilibrium classification in a different model. This is a related prior result for the taste reviewer to know, not a novelty verdict on this claim. - Reproduced the printed fixed-rate bistability independently at U=0.6, V=2e-9 (positive roots approximately 257.90864 and 9720.51043). Full Jacobian analysis agrees with saddle/sink classification. Supplied R does not reproduce the same normalized function because its NK denominator contains T^2; directly downloaded official file matches frozen public code.