MetaGenesis 核心 – 计算声明的离线验证
MetaGenesis Core – offline verification for computational claims

原始链接: https://www.metagenesis-core.dev/

该服务提供防篡改、可验证的“包”,用于计算结果,解决跨领域中的信任和可重复性问题。它允许独立验证输出,*无需*访问底层模型、代码或专用硬件(如GPU)。 主要应用包括:证明机器学习/人工智能模型准确性(ML_BENCH-01),确保制药/生物技术领域的法规合规性(DATA-PIPE-01 - FDA 21 CFR Part 11),促进金融/风险领域的独立模型验证(Basel III/IV),以及保证研究与数字孪生/有限元模拟的可重复性(SYSID-01, DT-FEM-01)。 验证速度快——通常在标准笔记本电脑上60秒内完成——并侧重于根据预定义标准(例如,准确性、相对误差)验证输出。 此外,也接受自定义声明,为验证几乎任何计算结果提供解决方案。

## MetaGenesis Core:计算声明的离线验证 MetaGenesis Core 是一种新协议,旨在*离线*独立验证计算结果,解决在不重新运行复杂流程的情况下信任结果的问题。该协议由 Yehor Bazhynov 独立开发,提供两层验证系统——完整性(SHA-256 哈希)和语义——确保文件真实性*和*声明本身的有效性。 现有工具如 MLflow 侧重于跟踪,但缺乏强大的第三方验证。MetaGenesis Core 通过验证关键语义元素的存在,防止“绕过攻击”,即修改后的数据仍然可以通过基本的完整性检查。 该协议领域无关,适用于机器学习、材料科学和金融等领域。它不需要 API 密钥或网络访问,使用 Python 3.11+ 运行。目前,该协议已通过广泛测试(91 个测试通过)和一个可证明的对抗场景验证,但正在寻求外部生产团队的验证。 该项目是开源的 ([https://github.com/Lama999901/metagenesis-core-public](https://github.com/Lama999901/metagenesis-core-public)),发明者欢迎反馈并识别与现有工具的潜在重叠。
相关文章

原文

ML_BENCH-01 Your model claims 94% accuracy. Your client wants proof they can verify in 60 seconds — no model access, no GPU. We package your result into a tamper-evident bundle. mg.py verify --pack bundle.zip → PASS

DATA-PIPE-01 · MTR series FDA 21 CFR Part 11 requires electronic records with audit trails. We package your ADMET predictions or calibration outputs into a verifiable bundle any regulator can check offline — no proprietary model access required.

ML_BENCH-01 · DATA-PIPE-01 Basel III/IV SR 11-7 requires independent model validation. We package your VaR or credit scoring outputs into a verifiable evidence bundle — your model validator gets proof without touching your proprietary code.

MTR-1/2/3 · SYSID-01 · ML_BENCH-01 Your reviewer can't reproduce your simulation without your exact environment. We package your result into a bundle they verify in 60 seconds on their laptop — no environment, no GPU, no trust required.

DT-FEM-01 · MTR-1 · DRIFT-01 Your ANSYS, FEniCS, or OpenFOAM output needs machine-verifiable proof against a physical reference. We package your displacement result as a tamper-evident bundle — rel_err ≤ 0.02 certified, verifiable offline by any engineering auditor.

Custom claim Any computational output can be verified. Describe your result below — we'll assess which claim domain fits and build a bundle for it. If your domain isn't covered yet, we'll tell you honestly.

联系我们 contact @ memedata.com