Axiom-0

Current bilingual reading note / 当前双语阅读说明（2026-11-13）
中文：Axiom-0 的 CODE、规范与固定样例可验证的性质必须分别说明，不能推断出通用 Agent、自主安全或绝对零熵能力；本段是文档解释补充，不替代当前实现、具体测试或历史证据
English: Axiom-0 CODE, specifications, and fixed-example checks establish distinct bounded properties, not autonomous agency, universal safety, or zero-entropy proofs. This is an interpretive guide, not a replacement for current code, individual tests, or dated evidence

Latest archived checkpoint: v2026.10-30-evidence-snapshot · Version DOI. The DOI badge above identifies the software-family concept.

Open Research / 开放科研 · Research Template / 科研模板

A dependency-free Python reference for explicit data contracts, measurable transitions, and reproducible repository checks. / 一个以显式数据契约、可度量状态转换和可复现仓库检查为核心的无第三方运行时依赖 Python 参考实现；

1. 仓库目的与非目标 / Repository purpose and non-goals

[CN] Axiom-0 演示规范化 JSON、SHA-256 摘要、概率分布校验、以 nat 为单位的 KL 散度、带校验钩子的事务式状态转换，以及一个十阶段事件记录样例；它是可检查的参考实现，不是基础模型、自主智能体、安全系统、授权层、沙箱、分布式调度器或数据库；

[EN] Axiom-0 demonstrates canonical JSON, SHA-256 digests, probability-distribution validation, KL divergence measured in nats, transactional state transitions with validation hooks, and a ten-stage event-recorded fixture. It is an inspectable reference implementation, not a foundation model, autonomous agent, safety system, authorization layer, sandbox, distributed scheduler, or database.

2. 已实现能力 / Implemented capabilities

3. 十阶段参考流程 / Ten-stage reference flow

[CN] AxiomOrchestrator.run_continuum(input) 的成功样例按 T-01 到 T-10 记录有序事件；T-04 使用注入的指标；当前 T-09 从内部状态读取分布，但 _dehydration_pipeline 并未把调用者的分布写入该状态，因此实际比较固定的默认向量与同一基线，虽然独立的 kl_divergence(p, q) 能计算显式输入，T-09 目前不能作为调用者输入的有效一致性门控；配置的 KL 阈值在现有数据路径上不会因输入分布变化而触发，输出事件包含变化的时间戳，run_id 因而是事件轨迹身份而不是稳定的输入摘要；请使用 input_digest 比较按本仓 JSON 合约规范化后的相同输入，阶段名称不表示认知、对齐或安全保证；

[EN] A successful AxiomOrchestrator.run_continuum(input) fixture records ordered events from T-01 through T-10. T-04 uses injected metrics. At T-09 the current data path compares a fixed fallback distribution with the same baseline: _dehydration_pipeline never places a caller-supplied distribution in the state consumed by T-09. The separately callable kl_divergence(p, q) is input-sensitive, but T-09 currently is not a coherence gate for the caller's distribution, and its configured divergence limit cannot fail due to changes in that input field on this path. Timestamp-bearing events enter run_id, which is a trace identifier rather than a stable input digest; use input_digest for the canonical input identity. Stage names do not imply cognition, alignment, or safety guarantees.

4. 验证与运行环境 / Verification and runtime

[CN] 以下命令是本地可运行的标准库验证入口；仓库当前没有自动执行 Python 测试的 GitHub Actions 工作流；只有实际记录了提交、解释器版本、命令和退出码的运行，才构成该环境的验证证据；成功结果只适用于被测提交、环境与样例；

[EN] The commands below are locally runnable verification entry points for the standard-library implementation. The repository currently has no GitHub Actions workflow that automatically runs the Python tests; only a run that retains its revision, interpreter version, command, and exit code is evidence for that environment. A successful result applies only to the tested revision, environment, and fixtures.

python -m compileall -q CODE tests *.py
python -m unittest discover -s tests -v

5. 前端与 Pages 边界 / Frontend and Pages boundary

[CN] FRONTEND/ 是独立的 React/Vite 展示层，不是 Python 参考库的运行时依赖；当前 Pages deployment workflow 使用 Node 24 和已提交锁文件执行安装与构建，并在部署过程中写出 docs/ 构建产物后上传 GitHub Pages；这个 deployment build 不是 Python runtime 或研究结论的测试证据；

[EN] FRONTEND/ is a separate React/Vite presentation layer and is not a runtime dependency of the Python reference library. The current Pages deployment workflow uses Node 24 and the committed lockfile, writes the docs/ build output during deployment, and publishes it to GitHub Pages. This deployment build is not test evidence for the Python runtime or research claims.

cd FRONTEND
npm ci
npm run lint
npm run build

6. 规范、决策、方法、证据、安全与复现 / Specification, decisions, methodology, evidence, security, and reproducibility

• 工程规范 / Engineering specification：实现接口、错误行为与仓库兼容边界； / Implemented interfaces, error behavior, and repository compatibility boundaries.
• 架构决策索引 / ADR index：长期架构决策、非能力边界与实现锚点； / Durable architecture decisions, non-capability boundaries, and implementation anchors.
• 方法论索引 / Methodology index：测量、解释与证据协调方法及其实现关系； / Measurement, interpretation, and evidence-reconciliation procedures with implementation relations.
• 证据基线 / Evidence baseline：外部资料、检索日期与本地结论的适用范围；它是有日期边界的校准，不是当前周期状态页； / External sources, retrieval dates, and the scope of local conclusions; it is a dated calibration, not a live period-status dashboard.
• 复现要求 / Reproducibility：提交、环境、命令、样例摘要和未测试边界的最小记录； / Minimum records for revisions, environments, commands, fixture digests, and untested boundaries.
• 发布策略 / Release policy：软件出版、归档身份与后续 main 演化的边界； / Software publication, archival identity, and later-main boundaries.
• 安全策略 / Security policy：私密报告流程、受支持代码与调用方责任； / Private reporting, supported code, and caller responsibilities.
• 长期维护契约 / Long-term maintenance contract：当前维护与历史保全的公共规则；它不定义运行时能力； / Public maintenance and historical-preservation rules; it does not define runtime capability.

日／周／月科研证据与版本快照 / Longitudinal research evidence versus versioned software

中文：RESEARCH/ 的 Daily A1–A4、Weekly A5、Monthly A6 是带来源、运行条件和时间边界的研究记录；图表、指数或某次 KL=0 的参考输入结果，不是全系统零差异证明；阶段展示应保留 canonical 路径、Git revision、输入、测试范围、实际执行状态与缺口；main 当前状态、历史日志、发布标签和 Zenodo 版本不能混为同一证据

English: Daily A1–A4, Weekly A5 and Monthly A6 in RESEARCH/ are research records scoped to sources, conditions and periods. A chart, index, or KL=0 for selected fixtures is not universal equivalence. Stage views must retain canonical paths, Git revisions, inputs, actually executed checks, and gaps. Current main, historical records, release tags and Zenodo versions are separate evidence identities

7. 限制 / Limitations

[CN] 通过样例和检查不能证明语义真值、模型或智能体对齐、性能、通用正确性或生产安全；规范化摘要证明字节级契约，不证明语义等价；KL 散度描述声明分布之间的差异，不判断事实真伪；阈值属于调用方策略；身份、授权、隔离、网络策略、配额、密钥管理、持久幂等和事件响应必须由集成方提供并单独验证；

[EN] Passing fixtures and checks do not establish semantic truth, model or agent alignment, performance, general correctness, or production safety. Canonical digests establish a byte-level contract, not semantic equivalence; KL divergence describes differences between declared distributions, not factual truth; thresholds are caller policy. Integrators must provide and separately verify identity, authorization, isolation, network policy, quotas, secret management, durable idempotency, and incident response.

Current periodic evidence note — 2026-09-23

Research artifacts under RESEARCH/** currently include September Daily pipeline evidence through 2026-09-23 and an open month-to-date A6 manifest.

These artifacts are evidence/history surfaces, not additional implemented capability.

PERIODIC_RESEARCH_PRESENT
!= NEW_RUNTIME_CAPABILITY

MONTH_TO_DATE_A6
!= FINAL_PROTOCOL_SEAL

BOUNDED_RUN_RESULT
!= UNIVERSAL_CORRECTNESS

The implemented-capability table above remains governed by current source and executable contracts.

Current periodic evidence note — 2026-09-24

The merged A1 review preserves the complete September evidence history through 2026-09-23. Current main now also retains the 2026-09-24 A1→A4 Daily pipeline manifest.

CURRENT_DAILY_PRESENT
!= NEW_RUNTIME_CAPABILITY
D_KL_0_ON_RECORDED_CASES
!= UNIVERSAL_ZERO_DIVERGENCE
100_OF_100_SPECIFIED_RUNS
!= UNIVERSAL_CORRECTNESS
INDEX_ALIGNMENT
!= SCIENTIFIC_VALIDITY

Current executable behavior remains owned by implementation and active specification/methodology contracts, not by the existence of periodic research artifacts.

Archived checkpoint — 2026-10-14

v2026.10-fourteen-day-evidence-checkpoint is archived as 10.5281/zenodo.23365534. For the software-family concept DOI, earlier versions and exact-checkpoint citation, see Release and Archival Policy.