# DSH Session Insights 0.5.2 candidate review

0.5.2 continues the 0.5.x host-adaptation and repair series. It requires DSH `>=0.2.0-rc.2`, refuses RC.1, and requires a new analysis run after upgrading from 0.5.1. Existing reports remain on disk. A separately installed Python CLI or Skill requires its own update.

The unpublished development candidates used version 0.6.0. Their [independent repair and qualification record](v0.6.0-candidate.md) remains historical evidence; their packages and receipts are not relabelled as 0.5.2. The release version was revised before tag, npm or GitHub Release publication. No new command, tool or report module was added in this repair round.

The final version change updates package and Python metadata, the analyzer version displayed in reports, current install instructions and packaging assertions. V4 input identity remains `v4-dsh020rc2.1`; report and semantic schemas, model validation, selection logic and the dashboard implementation remain unchanged. Compatibility admission does not establish that a future host has been tested.

Source/CI checks and the commit-bound artifact are recorded separately. The final artifact must pass native installation, strict second-install no-op, reinstall, full loaded-file identity, tool/command mounting and populated report-version checks on macOS and Windows Web, Headless and official Desktop. Prior runtime, GUI, model and page results are reused only with exact input hashes and explicit scope; version and installation results are freshly checked.

Real-provider positive output is labelled as schema-guided. The actual Headless negative test deliberately submits invalid output, attempts one still-invalid repair and generates a marked fallback. It does not establish naturally occurring model failure. HTTP observations, static self-containment checks and the user's offline local-file observations remain separate evidence.

This page records the candidate process. Publication and public artifact readback remain separate gates. Complete receipts and any residual acceptance limits are retained by the coordinating review and attached to the release when publication is verified.

## 中文

0.5.2 延续 0.5.x 的宿主适配与修复系列，要求 DSH `>=0.2.0-rc.2`，不再支持 RC.1；从 0.5.1 升级后须新建分析运行，已有报告保留。独立安装的 Python CLI 或 Skill 需要分别更新。

未发布的开发候选曾使用 0.6.0。[独立修复与资格化记录](v0.6.0-candidate.md)保留为历史证据，旧安装包和回执不会改名冒充 0.5.2。版本选择在 tag、npm 和 GitHub Release 发布前修订；本轮没有新增命令、工具或报告模块。

最终版本调整覆盖 npm/Python 元数据、报告显示的分析器版本、安装说明和打包断言。V4 输入身份仍为 `v4-dsh020rc2.1`；报告与语义 schema、模型校验、选择逻辑和页面实现未变。宿主按范围准入，不代表未来版本已实测。

源码/CI、唯一制品和各平台原生验收分别记录。最终制品需在 macOS、Windows 的 Web、Headless、官方 Desktop 上重新检查安装、严格幂等、重装、完整加载字节、工具/命令挂载及非空报告中的版本身份。此前运行逻辑、GUI、模型和页面证据仅在输入摘要与复用范围明确时保留，不能冒充新执行。

真实 provider 的正向结果标记为 schema-guided；真实 Headless 负向测试故意提交无效结果并做一次仍无效的修复，随后产生明确标记的 fallback，不冒充自然模型错误。HTTP 观察、静态自包含检查和用户的离线文件观察分别保留。

本页仍为候选过程记录。发布及公开安装包读回是独立 gate；完整回执与残余限制由协调复核保留，在公开发布核实后随 Release 交付。