[loopx-agent]

Goal And Delivered Outcome

A saved report from a read-only research source was readable but displayed “Analysis is unavailable; refresh to retry.” ResearchAnalysisPanel combined pack availability with execution permission. The same conflation advertised ineffective generation/retry actions for empty or failed read-only results and an enabled cancel button for a stored running receipt.

This change uses the existing pack availability to describe the source and the existing controller permission to gate actions. Saved reports remain readable; empty readers offer refresh, failure records retain their usage, and cancellation stays disabled on read-only sources. Genuine source changes and read failures still appear, with the last saved report preserved.

中文：修复“报告可读却提示分析不可用”的误报；只读空结果可刷新，失败与用量保留，待完成记录不再提供无效的取消操作。

• Outcome basis: the reproduced reader defect and existing source/permission contracts; complete within this bounded repair.
• Intended base: main, based on f094b2cfb2acde7a4a0d4d875bba52f64ac31ba9; a single author commit.

Author Declaration

• Written by: model_agent — OpenAI Codex, GPT-6 family.
• Specification: existing analysis controller guards and pack availability, plus AGENTS.md, docs/development/frontend-delivery.md and docs/development/design.md at f094b2cfb2acde7a4a0d4d875bba52f64ac31ba9.

Criterion	Disposition	Owner / path	Evidence
Saved results are readable independently of execution permission	implemented	ResearchAnalysisPanel	Baseline misreport reproduced; saved report in both locales passes
Discover, read, recover and read back through the existing product journey	implemented	Research entry, analysis panel, existing controller	Packaged research-history scenario and seven visual captures
Permission and source failure remain distinct	implemented	Existing readOnly, pack availability, execute/cancel guards	Empty/failed/running/source-change/read-recovery cases; zero UI write requests
Preserve original source and saved evidence	implemented	Existing trajectory pack/HTTP reader	Isolated source-byte readback unchanged; citations and failure usage retained

• Self-check: inspected the controller and renderer, tested the packaged reader with the real isolated HTTP/store path, reviewed the populated report and negative states, and scanned text and image publication boundaries. The existing guards and configuration owner are reused; no new protocol, setting or helper framework is introduced.

Scope And Continuation

The changed entrypoints are the research analysis entry and panel. Ordinary local interactive fixture behavior is retained. This is a source proposal with a packaged synthetic journey; installed adoption, model quality, research causality and the wider release remain unqualified. No successor is needed for this repair. Independent review and maintainer merge remain separate.

Validation

• Tested revision: f7258a72a8fa9810889d5068d9bcde4ea465bf9c; final source blobs match the successful runs below.
• Run state: finished.
• Input classes: synthetic, public_fixture.

Check kind	Result	Evidence / limitation
static	passed	Dashboard npm run build:desktop: TypeScript and Vite; existing bundle-size warning remains
integration	passed	scripts/chat_bundle.py build and verify, including the final committed revision
real_entrypoint / real_backend	passed	Packaged research-history, 35.7s: disposable HTTP server and real file-backed trajectory pack store, with synthetic provider outputs; no model service
regression_parity	passed	Saved/empty read-only cases in Chinese and English; failed/running/source-change/read-error recovery in Chinese; existing interactive generation/retry/cancel fixture cases retained
integration	passed	Saved results/citations use actual HTTP/store data. Source identity and negative transport states are explicit browser overrides. Backend readback reports unchanged source and zero UI mutation attempts
static	passed	Public text scan, DCO, diff/bilingual checks; no RFC pair changes. Semantic advisory found no supported new vocabulary carriers; its scope excludes arbitrary local strings
static	passed	Final premerge automatic checks: four direct, seven catalog and one public-boundary check; zero automatic failures
manual	blocked	Seven image privacy/visual holds remain for independent exact-head review. Author inspected pixels, synthetic provenance and PNG metadata; this does not clear those holds

The final main combination includes the previously merged first-frame CSS repair. The earlier baseline suite also passed. A wrong-working-directory packaging invocation failed before the script ran, then passed from the worktree root; no test failure is relabeled. No API, persisted-state, runtime routing or provider implementation changed, so backend promotion suites are outside this delta.

Frontend / Visual Evidence

• UI impact: changed. Source-packaged synthetic reader, not an installed App. The disposable capture servers were stopped; there is no public live preview.
• First viewport: desktop 1200×900 and narrow 390×844, report aligned to its opening. Before uses the prior packaged renderer whose source matches the baseline; after uses the committed renderer. The main CSS integration removes initial opacity animation only and does not change the shown report layout.
• Source data: synthetic. The following empty, failed and stored-running states use explicit synthetic transport observations over the existing HTTP fixture.

Viewport	Before	After
Desktop
Narrow

Empty / refresh only	Failed attempt / usage retained	Stored running / cancellation disabled

• Attention review: the report opens with its findings instead of a false warning; versions, sources and next experiments remain available. Empty and failed states expose reading recovery and the actual attempt record. Desktop comparison columns and narrow reflow remain readable without horizontal overflow. The pre-existing narrow source-chip truncation is outside this repair.
• Craft review: existing report hierarchy, typography, controls and status styles retained; no new grouping cards, colors, motion or configuration. The only shared interpretation change separates readable source state from execution permission.

Type of Change

• Bug fix
• Test update

LoopX Area

• Public docs or presentation surface (dashboard)
• Direction: existing Goal research / benchmark-toolkit readback.
• Shared-authority RFC fixture impact: N/A; no authority, provider or compatibility projection changes.

Boundary Checklist

• Public-safe text and synthetic images; no credentials, private state, raw traces, internal links or machine paths.
• Existing benchmark work and scoring are unchanged.
• Scoped to the reproduced reader defect.
• No paired RFC changed; bilingual check passes.
• Actual visual evidence is embedded above.
• Signed-off-by trailer on the author commit.

[loopx-agent]

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

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

动机

用户在只读研究来源打开已保存报告时，需要读到结论、失败与原文，并明确哪些操作可执行。
此前报告可读却提示分析不可用，空结果和失败提供无效分析或重试，旧运行回执还提供无效取消；现在保留报告、刷新与失败用量，只读来源不再承诺这些写操作。
真实隔离 HTTP/存储上的独立页面对照复现四种旧缺陷，当前 head 修复；读取失败保留报告，重试读取恢复，整个浏览测试没有写请求。
不改变分析权限或模型启用，不重新分析历史证据，不证明真实模型质量、研究因果、完整费用、安装态采用或整个研究发布完成。

改动思路

把来源可用性与执行权限分别从既有 pack 和 controller 读取，是现有渲染 owner 内的最小修复，无需新状态、设置或辅助框架。
本 PR 完成只读报告的误报和无效操作修复；原有来源身份、作业、重试、费用、模型和安装验收继续由各自 owner 负责。

未修改会继续复现四种误导；只删除警告又不能解决空结果、失败重试和取消入口，因此这几个分支属于同一个完整修复。useResearchAnalysis 的执行与取消 guard 原本已经阻止只读写入，本次让 ResearchAnalysisButton 与 ResearchAnalysisPanel 准确表达该事实。

规范依据：docs/reference/trajectory-assessment.md，固定修订 f094b2cfb2acde7a4a0d4d875bba52f64ac31ba9。按 existing-isolated-goal-sources 核验来源读取与显式执行，按 whole-history-presentation-language 核验失败/历史结果保留，结合既有前端 task-first-delivery 验证普通阅读、真实失败恢复与窄屏完整原文。它们都先于本修复；作者新增断言或图像不定义验收。经验建议仅用于追踪真实可读结果，不继承以往结论或科研效益。

具体改动

完整九路径：一个渲染文件 +7/-6、既有浏览器场景新增 84 行、七张合成图片。sourceAvailable 只看 pack 可用性，真正可执行仍需非只读且无读取错误；空结果和失败保留刷新与用量，已记录运行的取消控件遵从只读权限。原来源身份、原始报告、引用和失败回执均由现有 owner 维护，没有第二套配置或新公共词表。

独立同输入对照	基线	当前 head
已保存、来源可读	错报分析不可用	报告可读，无假警告
只读空结果/失败	无效分析/重试承诺	刷新读取，失败用量保留
已记录运行	取消看似可用	明确禁用取消
来源变化/读取失败	需保留真实诊断	警告仍在，报告保留，可恢复读取

本轮独立真实 HTTP/存储页面对照覆盖以上四个旧缺陷及真正来源变化，读取 503 后重试恢复。普通本机路径的保存、空、失败、运行、来源变化五种状态，其完整可见文本、控件与禁用状态逐字一致，无归一化。完整打包 research-history 场景及 16 项分析 API 测试通过，UI 写请求为零，来源字节保持。

对主干的风险

最重要的反例是修掉假警告时隐藏真正来源错误，或让已存作业状态冒充执行许可；同输入两向反例与既有 API/controller guard 均已核验。此次只改 UI 投影，不变更模型权限、存储、调度、配额或 API。当前相关 #7326 修复测试启动的包来源，不替代本次界面修复；两者属于不同 owner 边界。

本地构建、最终 Chat 打包校验通过。premerge 四项直接检查、七项 catalog 和一项公开边界自动检查通过；原始父命令仍为 exit 1，明确保留七个 PNG 人工 hold。本轮已检查七图像素、合成来源与 metadata：只有四图附带 ICC 数据，未见私人路径、账户或凭据。自动结果与人工判断分开，均不构成合并授权。未查询或等待 CI。

首次复用目录中的旧打包产物被校验拒绝，重建后原命令通过；独立对照第一次过早观察空视图，保留失败后仅修正等待完整 pack 的条件，全部语义断言保持，重新对照通过。没有将失败改记成通过或删除产品检查。

我的整体评价

没有阻塞发现，批准当前 f7258a72a8fa9810889d5068d9bcde4ea465bf9c 的完整范围。宽屏报告直接显示结论与竞争解释，390px 按原顺序阅读；正文保留 16px，实验、完整方法与引用仍可访问，空、失败和未核验运行状态各有对应操作。
未验证安装态 App、真实模型或独立研究者理解；窄屏来源标签原有截断仍在，未将合成材料的阅读通过视为科研或费用收益。

长期工作连续性保持：历史报告、实际失败、未知保留和原文仍能在重读/恢复后访问。当前修复完成了有限的阅读任务，不能以测试或截图数量结清研究能力、正式发布与安装采用。架构上复用既有来源和执行 owner；无需新增 helper、配置或兼容分支。源码打包、真实回环后端与合成负例各自的证据范围已经区分；平台的 COMMENTED 结论与正式自批准限制、任何合并权仍分别保留。

English verdict: APPROVE f7258a7. Readable source state and execution permission are now distinct. The packaged suite, 16 API cases, paired real-HTTP readonly counterexamples and five verbatim local parity cases pass; real source/read failures remain visible and recoverable. Seven binary holds were inspected separately; installed use, real-model quality and cost remain unqualified.