docs(rfc): define truthful Goal workspace composition and return metrics

[loopx-agent]

A returning Goal owner currently reconciles separate Overview, Tasks, work-map, usage, decisions, outputs and Stop controls. This RFC gives their page composition one presentation owner while each existing typed data/action owner retains authority. It defines M0–M4 and V1–V10 without claiming packaged or live-team delivery.

The corrected contract omits since-last-look cost until the usage owner qualifies that interval. Matching rolling-window labels cannot prove new spend: when old $100 expires and new $5 enters, the 7-day total falls from $100 to $5; −$95 is neither the new $5 spend nor zero. Both language mirrors, §5.2/§5.7, V5 and decision D7 now state the same rule. The browser marker holds bounded work/decision facts, with no cost ledger. Scope/coverage changes and unavailable data suppress cost while retaining known work changes.

Scope and ownership

• Persistent masthead with scoped Stop/Resume; six vitals with named window, scope, coverage and unknown behavior. Done, accepted and adopted remain distinct; liveness requires observed execution.
• Work by owner, existing Board/W1 Map lenses, shared Todo selection, an all-work rail, results and a return delta. The pure TS digest belongs to the existing presentation boundary; no new capability, provider, scheduler or persisted state is introduced.
• Roadmap S5, RFC index/status, intelligent presentation, live-team placement, usage open questions and design rules are reconciled. Local proposed display vocabulary becomes registered only when another consumer adopts it.
• The M1 checkpoint points to the unmerged draft #6191, distinguishing its independently tested subset, required Python qualification hold and untested GQ08/GQ09/model/installed-App outcomes. M2–M4 remain undelivered.

Synthetic composition studies

These static studies use the public GQ01 example; they are not packaged UI, runtime evidence or an approved first screen. All three now use Stop, show Work 5/10 separately from All work 11 (the separate decision), prioritize Needs you over observed activity, and order blocked responsibility before working/waiting. Cost stays a partial 7-day total; the return line reports work/decision changes only.

Light	Dark inverse	Phone (390px)

Attention is earned by the objective, the decision that needs an answer and truthful vitals. Repeated region headings and grouping-only cards are reduced; identifiers stay in diagnostics. The proposal's default UI change still requires implementation first-screen approval; the separate M1 candidate has that approval. This docs-only PR changes no installed surface.

Validation and remaining boundary

At exact head 6cb05f449053f1fa704210864d79f9b9612ce18c:

• Docs-governance smoke, full semantic vocabulary smoke and committed/staged/unstaged whitespace checks passed.
• Full source/public-boundary check passed: seven checks, 4,741 text files, zero errors and warnings; all 16 changed paths classified and checked.
• Three studies rendered and visually inspected; assertions cover no return-spend figure, scoped cost, Stop accessible name and corrected work counts. Images have no private metadata. An initial renderer check expected hidden phone text and failed; it was corrected to check the accessible name, then the final three-view render passed.
• Risk-based premerge passed all 18 selected checks (nine catalog, eight risk-profile and one boundary) and three whitespace checks; strict exact-scope change-quality qualification passed.
• Runtime tests are outside this docs-only change; static studies do not satisfy future packaged/real-backend acceptance. The M1 PR's required full Python failure remains its own blocker.

The bounded future-facing pass assigns one composition owner and removes unsupported browser cost knowledge; it does not introduce a second usage authority or speculative implementation. No backend rewrite is needed for this correction. Existing proposal history is preserved, all commits carry DCO sign-off, and exact-head review/readiness are required before any merge.

[loopx-agent]

Reviewer: model_agent; gpt-6.1-sol; OpenAI; runtime_reported; reasoning_effort=xhigh

Request changes conclusion (author-owned PR; GitHub blocks formal self-review)

Exact head: 430a584b2b5419ad4d2d28ed4e1607cfec7185b9. Whole-PR base: 5c62c5c03f13cc84176338100f2bded12801530e.

动机

离开后回到 Goal、想知道工作进展并作下一项决定的负责人。 当前概览、任务、地图、用量和暂停分散，负责人要重建工作上下文；提案让同一页呈现方向、待决定事项、工作与成果，并保留可回读的范围暂停。 设计给出了首个可用的 M1–M2 旅程和既有 owner 分工，但“上次查看后新增成本”仍不能由两次滚动用量总额相减得出。 本 PR 交付文档和合成组合研究；没有安装页面、改变执行权限或完成打包交互验收。

改动思路

复用既有任务、attention、生命周期暂停、验收与用量 owner，由纯 TypeScript 展示 reducer 组合页面；浏览器的“已看”标记不写服务端、不授予执行权限。M1 先交付页头、明确范围的暂停和可信指标，M2 再串起负责人分组、决定、成果及共享选中，符合先提供可用旅程的方向。保留散落区域会留下回看成本；另建执行或用量控制面则没有必要。M3 的成本应先等待既有 owner 给出真实区间，而不是由页面补造。

具体改动

全 PR 为 16 个文档/图片路径、898 行新增/14 行删除：英文 RFC 466 行、中文镜像 374 行、三张合成研究图，其余是索引、状态和现有 RFC/总纲归属对齐。三个图片均有 synthetic/not runtime evidence 标注；明暗桌面与窄屏完整看过。首屏默认部署仍明确等待预览批准，未把图当成已安装产品。

规范依据 docs/architecture/rfcs/loopx-overall-roadmap-v0.md，版本 5c62c5c03f13cc84176338100f2bded12801530e：S5 的同一旅程、成本与可信回读在本 head not_met，具体是 M3 成本；其余 M1–M2 分工没有发现阻塞。docs/development/frontend-delivery.md 的 Task-first delivery：提案明确模拟/打包边界，此项 implemented。

关键内容讲解

1. §5.3 每个指标具有单一含义、窗口、范围和覆盖度；未知使用“—”，完成任务不冒充验收通过，运行只来自已观察 Turn。Agent time 与 Cost 选择已有 7 天或 24 小时窗口，这些是滚动汇总。
2. §5.4–5.6 复用现有 Todo 详情、URL 和工作地图，以同一 digest 组织分组/看板/地图/侧栏；失败保留草稿，暂停保留进行中工作完成，避免展示层重写权威。
3. §5.7 只保存有限本地 digest，却规定两次窗口相同即可显示成本差分；§5.2 和三张研究图进一步把它写成“上次查看后 cost +$3.10”。窗口名称相同不表示统计区间相同。

对主干的风险

[P2] 不要用滚动窗口总额差表示上次查看后新增成本。 build_usage_summary 每次依据当前时钟重新筛选最近 7 天。在同一不可变合成历史上冻结两次真实 owner 调用：第一次旧 $100 仍在窗口内；两小时后它移出窗口，另有新 $5 进入。总额从 $100 到 $5，相减为 −$95，实际新增却是 $5。base/head 结果一致，证明不是汇总实现回归，而是本 PR 新设计的解释错误；同窗口检查无法阻止。

最小修复：在用量 owner 提供稳定 scope/coverage 的区间或单调计数前，省略 since-seen 成本；或者明确标为“滚动总额变化”，不宣称上次查看后消耗。同步中英文本、示例/研究图和 V5，加入旧成本过期同时新增、覆盖度变化及 unavailable 的冻结反例。无需为此发明浏览器累计账本。

语义与 CI 对齐

新 display classes/vital ids 留在局部 reducer、第二消费者再登记，归属说明合理。问题涉及既有用量 owner 的解释契约，不是词汇登记数量。文档治理 smoke 通过；生产汇总反例独立通过并暴露规范失败。没有查询、轮询或等待 GitHub CI；打包交互、键盘、暗色与实际暂停仍属于未来 M1–M4，未冒充通过。

我的整体评价

REQUEST_CHANGES。 long_horizon 与 user_experience 均 not_yet_proven：页面组织和早期旅程合理，但反复回看时成本会误导。体量主要是完整双语设计，复用现有 owner 的有界重构方向合理；无需扩大成后端重写。先修正 §5.7 及双语/图/V5，再按新的精确 head 重审。此评审不要求文档 PR 提前实现全部 UI，也不代表首屏批准或合并授权。

English review

The returning Goal owner currently reconciles scattered overview, work-map, usage, decisions and pause controls. This docs-only proposal composes them into a useful M1/M2 journey with existing typed owners, a scoped pause, shared Todo selection and distinct result acceptance. All three light/dark/phone images are explicitly synthetic studies; packaged interaction and first-screen approval remain future implementation gates.

The whole16-path +898/-14 diff and both language mirrors were reviewed. Accepted roadmap 5c62c5c03f13cc84176338100f2bded12801530e, S5, is not met for M3 cost truth; frontend Task-first delivery's simulated-versus-packaged boundary is implemented. The local display vocabulary and pure TS reducer placement are coherent, and broader runtime work is not required for this design review.

P2, §5.7 lines284–286: matching rolling-window names do not make subtraction a since-seen spending measure. Real unchanged build_usage_summary at base/head with frozen synthetic rows gives100USD then5USD after the old100 expires and new5 arrives; the proposed delta is-95USD although new spending is5USD. Governance checks and visual studies cannot detect this semantic error. Omit since-seen cost until the existing usage owner supplies a stable interval/monotonic source, or explicitly show only the rolling-total change. Correct both mirrors, examples/images and V5; test expiry plus new spend, scope/coverage changes and unavailable usage. Do not add a browser ledger.

Docs-governance smoke passed; the production-owner counterexample is paired and equivalent at base/head. GitHub CI was not queried or awaited. Packaged UI, keyboard, theme and actual lifecycle interaction are unexecuted M1–M4 obligations. Long-horizon and user-experience claims remain not_yet_proven for truthful cost readback. No backend rewrite is needed; re-review the corrected exact head.

English verdict: REQUEST_CHANGES - 430a584b2b5419ad4d2d28ed4e1607cfec7185b9; fix rolling-cost interpretation before accepting M3 guidance.

[loopx-agent]

Aligned with the M1 implementation in #6191 (commit 13e57b2):

• The masthead control uses the shipped Stop / 已停止 vocabulary instead of Pause.
• The State vital reuses the existing activity presenter.
• Last activity excludes agent-lane last_activity_at, which is a Todo-write hint.
• The needs_you work class is deferred to M2.
• The M1 checkpoint keeps passed, untested and deferred acceptance separate.

The docs governance smoke passes.

[loopx-agent]

Reviewer: model_agent; gpt-6.1-sol; OpenAI; runtime_reported; reasoning_effort=xhigh

Request changes conclusion (author-owned PR; GitHub blocks formal self-review)

Exact head: 13e57b29e043d35983dee402d9b83e78ebfc044e. Whole-PR merge base: 5c62c5c03f13cc84176338100f2bded12801530e.

动机

离开后回到 Goal、想知道工作进展并作下一项决定的负责人。 当前概览、任务、地图、用量与停止分散；提案由同一页面组合可信指标、决定、工作和成果，M1 用既有生命周期 Stop 回读，活动与 needs-you 细分延期到 M2。 当前双语设计已澄清 M1 停止及原生观测边界，但“上次查看后新增成本”仍不能由两次滚动用量总额相减得出。 本 PR 交付文档和合成组合研究；没有安装页面、改变执行权限或完成打包交互验收。

改动思路

既有 registry、Todo、attention、usage、execution observation、验收与生命周期 owner 保留；纯 TypeScript 展示 reducer 组合页面，本地已看标记不写服务端或授予执行权限。当前 head 已纠正 M1 的 Stop 语义，活动与 needs-you 精确细分明确留 M2，并记录 #6191 的拟交付 checkpoint。阶段划分比先堆页面控件更可信；本评审没有把该作者 checkpoint 当成本人的实现验收。另建用量账本仍无必要，M3 要等待原 usage owner 给出真正区间。

具体改动

完整16路径 +926/-14：483行英文 RFC、385行中文镜像、3张合成明/暗桌面及手机研究图；README/RFC STATUS、总纲S5、现有 usage、intelligent presentation、live-team work-map 和 design规范同步。全 diff、双语与全部图片重新看过。页头/指标、工作 owner分组、共享 Todo选中及URL、侧栏/成果、M0–M4依赖与V1–V10边界保持同一主题，无运行代码。

依据接受前的 docs/architecture/rfcs/loopx-overall-roadmap-v0.md @ 5c62c5c03f13cc84176338100f2bded12801530e：原 S5 的可信成本及回看旅程在此设计 not_met，范围是 M3 成本推导。docs/development/frontend-delivery.md 的 Task-first delivery：文档明确模拟/打包界限，M1–M2 有真实入口与恢复计划，此设计边界 implemented，不要求这个文档 PR 提前实现全部 UI。

关键内容讲解

1. §5.1–5.3 每个 vitals复用既有 owner，指标标窗口/范围/覆盖度，未知“—”；完成不冒充验收，活动不从注册或 updated_at推断。
2. §5.4–5.6 现有Board/Map、Todo详情和URL由同一纯展示digest组合；失败保留草稿，Stop/Resume保留原权威，needs-you细分和Activity延期到M2，不让前端替代 owner。
3. §5.7仍保存 usage window values 并仅检查两份digest窗口名称相同，即展示 since-seen成本差分；§5.2和3张研究图仍写“Since you looked…cost +$3.10”。V5只加入窗口变化时省略，未处理同窗口滚动过期。

对主干的风险

[P2] 同窗口滚动总额相减仍不能表示上次查看后新增成本。 在当前head重新执行真实 build_usage_summary，以相同合成记录和冻结时钟同时测不可变base/head：第一次旧$100还在7天窗口内；两小时后旧$100移出、另有新$5进入。窗口总额$100→$5，相减 −$95，实际新消耗却是 $5。两版owner结果一致；错误属于本PR新增的设计解释，不是要求汇总代码修一个不存在的回归。§5.7（284–286行）、§5.2、中译和V5在新提交后仍保留该规则。

最小修复：原usage owner提供稳定scope/coverage的区间或单调计数前，省略 since-seen成本；或者明确只表示“滚动总额变化”，不宣称上次查看后消耗。同步双语、示例/图片与V5，补旧成本过期同时新增、scope/coverage变化及unavailable的冻结反例。无需新增浏览器累计账本。

设计正确性之外，图片依旧为合成研究且显示Pause，与新Stop文本存在展示历史边界；本文明确尚非获批/打包界面，M1准出须用当前真实页面证据替换，不能拿这3图当Stop验收。索引的Accepted是提案元数据，不证明所有runtime交付完成。M1#6191checkpoint、本机采用与M2活动 readback 未在本doc审查独立核验，保留passed/untested/remaining的作者边界。

语义与 CI 对齐

局部display classes/vital ids留在pure reducer，第二消费者才注册，无平行控制面；usage语义必须继续来自原owner。fresh文档治理 smoke和diff检查通过，生产owner冻结反例独立执行通过并揭示规范失败。没有查询、轮询或等待GitHub CI。源码界面、键盘、真实Stop与重启等资格属于后续实施，未以synthetic图片替代。

我的整体评价

REQUEST_CHANGES。 新head的M1停止/活动边界澄清有价值，完整页面设计与既有owner分工合理；long_horizon与user_experience的可信成本回看仍 not_yet_proven。体量主要是完整双语合同及归属对齐，无需后端重写；future-facing pass支持单一TS展示组合，成本先删不被owner支持的推导。先修本精确head的§5.7及双语/图/V5，再重审。此结论不要求提前完成全部页面，也不授予首屏批准、merge或上线权限。

English review: The full16-path docs/design change was re-audited, including both mirrors,3synthetic studies, roadmap/index/domain reconciliation and the new M1Stop/native-observation checkpoint. Existing typed owners and a bounded M1/M2 journey are appropriate; implementation/preview qualification remains separate. The unchanged§5.7same-window rule still misstates since-seen spending: fresh real usage-owner calls at immutable base/head give rolling100USD→5USD after100expires and5new arrives; subtracting gives−95USD instead of5USD. Omit unsupported since-seen cost or explicitly label rolling-total change; update both mirrors, studies andV5 with expiry, scope/coverage and unavailable cases. Docs governance passes; CI not consulted; author M1qualification is not inherited.

English verdict: REQUEST_CHANGES - 13e57b29e043d35983dee402d9b83e78ebfc044e; rolling-cost inference remains a P2design blocker after the Stop/observation clarification.

[loopx-agent]

Reviewer: model_agent; gpt-6.1-sol; OpenAI; runtime_reported; reasoning_effort=xhigh

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

Exact head: 6cb05f449053f1fa704210864d79f9b9612ce18c. Whole-PR merge base: adb247e5f3fac8e2e3f7152239955b49480d0b9f.

动机

离开后回到 Goal、想读懂进展并作下一项决定的负责人。 过去要在概览、任务、地图、用量与停止之间重建上下文；新设计在同一页面组合方向、可信指标、决定、工作与成果。 本次可观察结果是双语规范和三张合成研究图一致：回看变化不再把滚动成本差冒充新增花费。 本 PR 只交付设计合同；不安装页面、不改变执行权限，也不代替打包或真实团队验收。 剩余 M1 实现与真实交互资格在现有 #6191；M2–M4 和用量 owner 的 D7 区间资格仍未交付。

改动思路

既有 Todo、attention、生命周期、用量、执行观察及验收 owner 保留权限；页面只在 TypeScript 展示边界组合投影。浏览器已看标记只含有界工作与决定事实，不新增服务端写入或成本账本。继续散落区域会留下回看成本，另建用量 owner 没有必要；省略尚不被原 owner 支持的成本增量就是较小且完整的修复。M0 规范可独立回退，M1–M4 的真实交互验收继续由既有实现与 owner 承担。

具体改动

完整 16 路径 +959/−14 已审阅：英文/中文 RFC、三张浅色/深色/390px 合成研究图，README 和双语 RFC STATUS，双语总纲 S5、intelligent presentation、live-team W1 归属、usage 未决问题和 design 指标规范。索引的 Accepted 是设计状态；§11 仍写清 M1 在 #6191 提出、未合并，完整 M2–M4 未交付。

独立规范依据：docs/development/frontend-delivery.md @ adb247e5f3fac8e2e3f7152239955b49480d0b9f，Task-first delivery 要求文档/原型比较拟议旅程并明确模拟边界，当前为 implemented：三图及 §5.2/§9/§11 都区分合成、打包、安装与真实团队资格。接受前同版本总纲 S5 的同一旅程与可信回读作为方向，不把未来 V1–V10 全部变成本次文档的准入要求。

关键内容讲解

1. §5.1–5.3 绑定单一数据/action owner，成本与 Agent 用时同窗口，未知不是零；完成、验收、采用分别呈现，运行来自观察。
2. §5.4–5.6 复用 Board、W1 Map、Todo 详情及 URL，共享分组与计数；局部显示词汇到第二消费者才登记。
3. §5.7/V5/D7 删除同窗口总额相减；只有原 usage owner 核验区间或兼容单调计数才能显示回看成本。浏览器不存成本账本，范围/覆盖度变化或不可用时省略成本并保留已知变化。
4. 三图去掉成本增量、Pause 改 Stop，Work 5/10 排除另列决定（All work 11），State 先呈现 Needs you、活动留副行，blocked 负责人排在 working/waiting 前。§11 记录 #6191 子集通过与未测项，未把确认后目标消失负例冒充过期指纹验收。

对主干的风险

没有发现当前文档修复的阻塞项。 先前两个 review 的滚动成本问题已逐条映射到当前双语、V5、D7 与全部三图，未依据旧提交或绿色检查自动继承结论。再次冻结真实 build_usage_summary：旧 $100 过期、新 $5 进入后总额 $100→$5，差值 −$95 不是新增 $5；缺失用量保持成本字段缺失。现在默认省略此推导，不要求修改正确的汇总实现。未来 owner 还须证明区间覆盖、币种与代际；该责任不是文档通过即可免除。

文档治理、全量语义、7 项 full check（4,741 文本文件、零错误/警告）、空白检查和最终三图渲染/视觉检查通过。最初手机断言读取隐藏的 Stop 文本失败，改为验证可访问名称后最终三视图通过，历史失败未抹去。静态研究不能证明持久化、生命周期效果、深色 AA 或真实模型采用；这些仍属于 M1–M4。没有查询、轮询或等待 GitHub CI。#6191 的 required Python 失败保留为该实现 PR 的合并资格阻塞，不由本 docs 审查豁免。

我的整体评价

APPROVE 当前文档增量。 long_horizon 与 user_experience 为 improved，含义是设计合同更可信：反复回看不再生成假的消耗，决定、工作与 Stop 语义一致；不宣称已安装体验改善。独立且可回退的 M0 修复有价值，完整双语设计的体量合理。future-facing pass 已应用单一组合 owner 和成本知识删除，没有额外框架、权限或平行账本；规范继续将未知字段/新词汇留给原 owner。剩余 #6191 required 验证、M2–M4、真实团队及 D7 区间能力不在此结论中。这是精确版本的公开自审，合并仍须原生 readiness 和执行权限回读。

English verdict: APPROVE - 6cb05f449053f1fa704210864d79f9b9612ce18c; the whole 16-path docs proposal now omits unsupported since-seen spend, preserves existing owners and honestly labels all synthetic studies. Governance, semantic/public checks, visual readback and the real usage-owner counterexample passed; runtime M1–M4 and #6191 qualification remain separate.