test(release): align inventory and onboarding validation with contracts

[huangruiteng]

The 1.2.1 release sweep found stale validation assumptions after source simplification and prompt compaction. The live-inventory tests assumed a new fork always exceeded the fixed production ceiling, and searched only the first slack token. The onboarding smoke required an obsolete sentence despite already checking the native Goal activation, visible-Goal spend source, and non-automation classification.

Keep every production threshold and error assertion. Pin the fork test's registry and matching anchor to its measured baseline so the additional fork actually overflows; inspect complete comma-separated slack items so exact zero/two slack remains enforced regardless of sibling order. Remove the redundant prose assertion while retaining all existing typed activation/quota/classification, stop, blocked, and user-resume assertions. No product, permission, state, scheduler, or default behavior changes.

Validation: all116 semantic-vocabulary tests pass; the real onboarding CLI smoke passes, including its existing positive/negative cases. The initial clean-source sweep's three inventory assertion failures are retained as failure evidence. Full repository preflight continues independently; final release qualification and public artifacts are not claimed yet.

[huangruiteng]

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

Reviewed exact head: 5149@c2f0e11bf46a860fe4be0f65b98fc3442153e262; base: 6669c178bfeb2269d5d67c6203b9f5a535e6e180.

动机

1.2.1 的真实发布检查复现了3处 inventory 测试假设过时，以及1处 onboarding smoke 要求已被压缩掉的旧提示句。这些失败把源码简化产生的预算余量、合法的列表顺序和已有结构化约束误认为回归，阻塞发布。

改动思路

保留原 overflow 错误断言，把 fixture 的 registry 与匹配 anchor 同时固定到其测量基线，使新增 duplicate 必然真正越界；slack 读取完整 token 而不假设排在第一位。onboarding 已验证 native activation、运行 profile、visible-Goal spend source 和非 automation 分类，因此删除重复旧句断言即可。没有生产预算、source classifier、状态、权限、调度、默认行为或 API 改动；没有新 helper 或抽象。

具体改动

完整差异为两个原有验证文件、10行增加和4行删除。当前116项 semantic-vocabulary pytest 通过，实际 onboarding CLI smoke 通过，包括既有正反路径；原 guard 对 missing／invalid snapshot 下的越界 duplicate 仍拒绝，零 slack 与协调增加后的恰好2仍必须披露，native Goal 的 quota 事件仍必须判为非自动化。停止、blocked 与仅用户恢复的原断言保留。Ruff、diff hygiene、3项原生预合入／公开边界检查和精确差异质量回执通过。持续运行和用户体验保持：没有生产代码变化，修复现有验证消费者；最终发布全量资格和真实模型／产物读回仍由当前发布操作负责，当前全库预检未结束，不声称已经发布。

对主干的风险

风险是用放宽测试来掩盖真实退化。本 head 没有改生产上限，保留原拒绝断言，fixture 只在测试范围内协调收紧基线及 anchor；完整 token 检查保留零／恰好2的独立期望。删除的句子不再是 Host 边界的唯一证据：原 activation_method、host_surface、runtime profile、CLI 参数解析、spend source 和 is_automation_run=false 都保留，原停止／恢复行为也保留。没有新的 prose 分类、默认启用、泛化义务或 permission grant。完整2文件差异和原 consumers 已读；局部验证不替代最终干净发布提交的全量资格。

我的整体评价

该修复消除已经复现的验证漂移，完整覆盖同一维护问题，且保持生产门禁。未来改动便利性检查已应用到 fixture 基线和完整 token 断言，移除重复 prose 检查，无需额外框架。当前精确 head 可以按低风险测试／窄清理规则合入；没有当前差异 blocker，最终发布仍须完成独立资格和远程读回。

English verdict: APPROVE