[BigDataDZ]

On native Windows with PYTHONUTF8=0, the repeated-registration regression decodes the UTF-8 Goal state with the GBK locale and fails before it can verify replay preservation. Its later writes also use the locale, so non-ASCII owner narrative cannot be safely tested.

Keep all state IO in that existing replay test explicitly UTF-8, and extend its existing current/historical writer matrix with Chinese, an accented character and an emoji. The existing byte-preservation, equivalent-header, later narrative, paused status and acceptance-drift assertions remain. No production registration or decoding rule changes.

Validation at signed head 2f422f98f, based on main 218aa087b:

• Original matrix: 1 failed / 7 passed on native Windows / CPython 3.12.10 with UTF-8 mode disabled; the historical Unicode-separator fixture fails in the test's locale read.
• Updated matrix: 10 passed / 65 deselected under the same native mode, through the real CLI and canonical storage.
• Full module: 73 passed / 2 failed. The strict-registry rebinding assertion still expects a different error message (also seen on the previous unchanged main). A distinct-registry concurrency case hit an owning-lock timeout in this full run; its cause is not established and no fix is claimed. These results are preserved, not counted as passing.
• Changed-file Ruff and diff hygiene pass. Only one existing test file changed; no private paths, raw logs, new state vocabulary, product UI or API changes.

Author declaration: authored with OpenAI Codex under contributor direction. The related future-facing pass keeps UTF-8 ownership local to this existing replay test rather than adding a new fixture or changing global Python behavior. No model/benchmark job, active Goal mutation, system upgrade or self-merge. Independent review and current-head CI remain pending; maintainer merge required.

[loopx-agent]

Reviewer: model_agent | gpt-6.1-sol | OpenAI | runtime_reported | reasoning_effort=max

动机 / Motivation

维护 project register 回归的贡献者会在非 UTF-8 的 Windows 环境下触发这个问题。原测试在重复注册后用本地编码读取已经按 UTF-8 保存的 Goal 文件，可能先解码失败，也可能在后续写入时损坏非 ASCII 正文；改后测试明确按 UTF-8 读写，再验证原状态没有被重复注册覆盖。独立受控 CP936 注入下，基线既有矩阵为 1 失败、7 通过；精确 head 的 10 个当前及历史 writer 场景全部通过，完整注册模块 75 项通过。本 PR 只修复既有回归测试的编码并扩展输入，不修改注册实现、状态格式、权限或界面，也不认证全部原生 Windows 行为。

改动思路 / Approach

这个切片修的是测试自身的文本边界。注册实现本来就按 UTF-8 写状态，测试应使用同一文件契约读取和构造后续呈现，再由原有字节断言判断重放有没有破坏它。把整个测试环境强制成 UTF-8 会失去问题的判别力；删除特殊输入也会减少保护。这里直接修复原函数内五处 I/O，新增一个同时含中文、重音字符和 emoji 的输入，当前/历史 writer、转义等价、暂停状态和验收漂移继续走实际注册入口，没有另起恢复 owner 或测试框架。

具体改动 / Changes

关键代码讲解

• test_project_register_repeated_identical_request_is_a_noop 在现有参数矩阵加入混合 Unicode。这里的 legacy 是历史展示 writer 形状，并不代表绕过规范存储的旧 authority provider。
• 同函数后半段 的五个文本读写显式指定 UTF-8；read_bytes 比较、changed is False、storage_selection.status == "replayed" 和 drifted 正文相等断言都保留。测试仍调用 loopx.cli.main，没有以 mock 返回成功替代产品执行。

规范依据为 docs/reference/source-cli-entrypoint.md，固定 spec_revision=bafbafc258de72029b54ea016103c0a7bd2409c9：已接受正文。UTF8-FILE-IO → implemented：文件文本 I/O 固定 UTF-8，由受控 CP936 反例和新矩阵验证；REGISTER-OWNER-PRESERVED → implemented：原注册/持久状态/回执 owner 未改，完整模块和原字节保护断言通过。这是对已有义务的覆盖，不新增语义词汇或产品行为。

对主干的风险 / Risks

未发现阻断问题。独立验证使用 macOS CPython 3.14.8：普通完整模块 PASS，75 passed；只对目标状态文件的隐式 Path.open 注入 CP936，基线 218aa087b66bb99038b1ce7d782110741a5e42ae 在历史 writer 分隔符输入处出现 UnicodeDecodeError（1 failed / 7 passed），head 相同控制条件 PASS，10 passed。这个负对照针对测试编码，不是原生 Windows 仿真；显式 UTF-8 产品 I/O 没有被替换。标准 canary premerge --from-git-diff 的 5 项直接检查通过，选择器没有额外 smoke，未将这个结果当完整产品资格。

作者此前 native Windows 全模块的 strict-registry 诊断和并发锁两项失败仍是历史 UNVERIFIED 范围；本次完整 macOS 绿测不证明它们修好了。此补丁不修改这些诊断或锁的因果路径，也没有放宽相关断言。没有查询、轮询或等待 GitHub CI。还核对了同作者 #7454/#7455：分别处理依赖供给和另一报告生命周期，不是重复注册 smoke；无需合并成一个测试框架或发起贡献限制。

我的整体评价 / Verdict

建议 APPROVE，结论绑定 2f422f98fc5a2c782b3db3795a88ccc4b881503d。这是完整且可回退的测试维护修复：长期工作语义 preserved，覆盖仍经过真实重复注册、状态漂移和持久字节回读；产品体验 not_applicable，没有改变用户操作或呈现。视觉检查 not_applicable：仅测试文件变化，未改变任何渲染页面或可见呈现。未验证视觉范围：none：此 diff 没有视觉变更。没有新增兼容层，继续在原函数维护是当前最小边界；原生 Windows 全面资格、合并决定和安装采用分别保留，不由本批准代替。

English verdict: APPROVE — the existing replay test now preserves its UTF-8 contract without weakening its state-preservation assertions; native Windows qualification remains explicitly separate.