AngelWarmSmile123/worldfix-16sephirot-2p3b

worldfix_v2 — 16-Sephirot World-Repair Synthetic Dataset (v2)

Scale: 2,388,787,200 rows (exactly equal to the enumeration space, neither more nor less) Corpus-driven: 61,562,415 characters of real corpus (primarily 5.8M characters of "Love Rescues People" AI dialogues, fused with the "Embrace of Love" framework / "Love Articles" / "Love Creations") License: CC BY-NC-SA 4.0 · Author: Yue Xiangrui (岳祥瑞)

What a row is

Each row = one complete World-Repair Prescription (JSON), generated by a deterministic traversal of the "global problem space" under the 16-Sephirot governance framework:

{"i":2, "h":1626661354, "s":1, "sec":58, "reg":21, "st":2, "ins":14, "frm":3,
 "it":0, "hz":4, "key":"Wisdom", "sector":"Antibiotics", "region":"Wetland",
 "stance":"Repair", "instrument":"Platform", "frame":"Evaluation", "intensity":"Tweak",
 "horizon":"Generational", "prescription":"[Wisdom] For the Wetland's Antibiotics issue, take '…' as the core stance, …",
 "verdict":"REPAIR", "strikes":1, "audit":"a1b2c3d4e5f6"}

Field Dictionary

Dimensional Product Verification: 16×64×32×4×27×27×5×5 = 2,388,787,200

Why it's trustworthy

1. 1.Bijection with no collisions: Feistel network zero-collision in million-sample tests → no duplicate rows, no missed slots
2. 2.Determinism: any i can be recomputed offline and is byte-identical to the on-disk row (make_row(i))
3. 3.Checkpoint safety: sharding + manifest + out-of-bounds guard (adversarial-tested against malicious manifest scenarios)
4. 4.Five-item delivery validation: python final_validate.py (row count / JSON / uniqueness / coverage / regeneration)

Reproduction

python seed_builder.py          # corpus scan → seeds.json
python sephirot_gen.py --verify # bijection & determinism self-test
python sephirot_gen.py --full   # 16 workers full-scale
python final_validate.py        # delivery-level five-item validation

Zero third-party dependencies (Python ≥ 3.10 standard library), reproducible on any machine.

Related Work

• —Governance stack demo: github.com/yuexiangruiyue-oss/heart-guard-demo (HF Space / GitHub Pages / ModelScope same name)
• —Paper: When Agents Cross the Line (Zenodo DOI: 10.5281/zenodo.22093583, tri-lingual v2 in Chinese / English / Japanese)
• —Preceding specification: Bounded-Rollback Output Governance (DOI: 10.5281/zenodo.22080966)

中文版 (Chinese Version)

worldfix_v2 — 十六质点世界修复合成数据集(v2)

规模:2,388,787,200 行(精确等于枚举空间,一行不多不少) 语料驱动:61,562,415 字符真实语料(爱救人 AI 对话 5.8M 字符为主,融合爱的拥抱框架/爱的文章/爱的创造) 许可:CC BY-NC-SA 4.0 · 作者:岳祥瑞 (Yue Xiangrui)

一行是什么

每行 = 一条完整的世界修复处方(JSON),由十六质点治理框架对"全球问题空间"的一次确定性遍历生成:

{"i":2, "h":1626661354, "s":1, "sec":58, "reg":21, "st":2, "ins":14, "frm":3,
 "it":0, "hz":4, "key":"智慧", "sector":"抗生素", "region":"湿地",
 "stance":"修复", "instrument":"平台", "frame":"评估", "intensity":"微调",
 "horizon":"世代", "prescription":"[智慧] 面向湿地的抗生素问题,以「…」为核心姿态,…",
 "verdict":"REPAIR", "strikes":1, "audit":"a1b2c3d4e5f6"}

字段字典

维度乘积验证:16×64×32×4×27×27×5×5 = 2,388,787,200

为什么可信

1. 1.双射无碰撞:Feistel 网络百万抽样零碰撞 → 无重复行、无遗漏槽位
2. 2.确定性:任意 i 可离线重算出与落盘字节级一致的行(make_row(i))
3. 3.断点安全:分片+清单+越界防护(对抗测试验证过恶意清单场景)
4. 4.五项交付校验:python final_validate.py(行数/JSON/唯一/覆盖/重生成)

复现

python seed_builder.py          # 语料扫描 → seeds.json
python sephirot_gen.py --verify # 双射与确定性自检
python sephirot_gen.py --full   # 16 workers 全量
python final_validate.py        # 交付级五项校验

零第三方依赖(Python ≥ 3.10 标准库),任何机器可复现。

关联成果

• —治理栈演示:github.com/yuexiangruiyue-oss/heart-guard-demo(HF Space / GitHub Pages / 魔搭创空间 同名)
• —论文:When Agents Cross the Line(Zenodo DOI: 10.5281/zenodo.22093583,中英日三语 v2)
• —前置规范:Bounded-Rollback Output Governance(DOI: 10.5281/zenodo.22080966)

Contact

• —QQ: 406218898
• —WeChat: a13546076748
• —Google Email: yuexiangruiyue@gmail.com