fix(diagnostics): distinguish status and doctor route-health scope

[huangruiteng]

Goal And Delivered Outcome

status and doctor can report different route health because they read different registries or Goal scopes. Equal counts do not establish equal scope. This replaces #4594: retain the compact count, expose the actual diagnostic registry/runtime/filter metadata, reuse status's existing scope envelope, and label doctor as a global drill-down.

Scope And Continuation

Complete within diagnostic interpretation. The existing collector still owns the health calculation; no route resolution, projection writes, Goal selection, permissions or provider behavior changes. The related simplification pass avoids a second health authority, scope fingerprint and duplicate scope object on the hot path.

Validation

• Tested revision: da75ad88e0bba892f8b495ca38ad7edf872fbd27
• Run state: finished
• Input classes: synthetic, public_fixture

Check kind	Result	Public-safe evidence / limitation
real_entrypoint / regression_parity	passed	Shared-runtime refresh and material projection smokes exercise real CLI writes/readback. Same routes agree; two registries with equal counts but healthy/lagging results remain distinguishable by actual provenance.
static	passed	Ruff, configured mypy (22 files), diff hygiene and boundary scan of all seven paths; zero hits.
real_entrypoint	passed	Hot-path and full CLI base/head output-budget smokes; no budget changes.
integration	passed	Risk-based premerge: 19 selected checks and 4 direct checks passed, no manual holds. An initial missing TypeScript parser dependency was repaired before this passing rerun.

Exact-diff quality receipt cqr_79a34ae6d4714192de0b is valid for 7 files; no blockers/warnings/advisories, safe-fix allowed but not used. No live model or benchmark jobs were run. Maintainer review and merge remain required.

Frontend / Visual Evidence

UI impact: none. CLI Markdown and diagnostic JSON change. The packaged frontend has no consumer of runtime_projection_routes; no settings or Lark companion is required. Health remains a scoped observation, not Goal acceptance.

Type of Change

• Bug fix
• Test update

LoopX Area

• Control plane (runtime diagnostics)

Shared-authority RFC fixture impact

N/A: read-only diagnostic metadata; no authority-store or runtime-routing change.

Boundary Checklist

• No private state, credentials, raw traces, internal links or machine paths.
• No benchmark execution.
• Scoped to diagnostics; every commit is DCO signed.

[huangruiteng]

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

Reviewed exact head: da75ad88e0bba892f8b495ca38ad7edf872fbd27 (codex/status-doctor-route-health).

动机

status 与 doctor 可能报出不同的 route health，因为它们读的注册表或 Goal 范围不同；而计数相同并不等于范围相同。旧输出只给 healthy，于是操作者无法判断「两个面不一致」到底是范围差异还是真的退化。这正是这类诊断最容易误导人的地方。

改动把「我量的是哪个范围」变成输出的一部分：diagnostics payload 带上 registry / runtime_root / goal_filter / activation_state_filter，status 复用既有 scope envelope 增列 goal_count，doctor 打上 registry/goals/counts 并把自己标成 global drill-down。

改动思路

入口是 collect_runtime_projection_route_diagnostics（唯一 health 计算权威，本次不改其判定逻辑），由 collect_status 与 collect_doctor 各自消费：status 把 goal_count 加进既有 envelope，doctor 在 markdown 里补 provenance，status markdown 把 details=loopx doctor 改成 global details=loopx doctor 并加上 goals=<count>。作者明确说明这次没有引入第二套 health 权威、scope 指纹或热路径上的重复 scope 对象——这几点在 diff 里成立：改动是「把 collector 自己已经用到的输入接到返回收据上」，不是新机制。

具体改动

7 个文件、+58/-8：payload 四个 provenance 字段、status envelope 的 goal_count、两处 renderer 文本，以及三个 smoke 从「== {"healthy": True}」改成断言 provenance 关系。

我跑了三个真实入口 smoke，全部通过：refresh-state-shared-runtime-projection-smoke（passed）、shared-runtime-material-projection-smoke（passed）、hot-path-interface-budget-smoke（ok，top_level_keys=25/25，说明新增字段是嵌套的、没有顶破接口预算）。

其中 refresh-state smoke 的新断言正是这条 PR 的核心语义，我读了实现：

• 同一批路由在 status 与 doctor 上的 healthy / goal_count 必须一致；
• doctor 的 registry 必须等于共享 runtime 的 global registry，而 status 的 registry 等于本次选中的 source registry，两者必须不同——「计数相同、范围不同」因此可被区分；
• 另写一个不相关工程/注册表，其 healthy 结果的 goal_count 与 lagging 的那个相同，但 registry 字符串不同，仍可区分；
• markdown 断言包含 goals=1、registry: ...、goal_filter: ...、global details=loopx doctor，doctor markdown 含 registry=…、goals=、counts=。

关键内容讲解

• runtime_projection_route.py:694：payload 增加 registry（解析后绝对路径）、runtime_root、goal_filter、activation_state_filter（经 normalize_goal_activation_state 归一化）——都是 collector 实际使用的输入，不是重新推导的。
• status/collection.py:159：只在既有 runtime_projection_routes envelope 上加 goal_count，保持「复用 status 既有 scope」而不是新增 scope 对象。
• status_markdown.py:142：输出 healthy=False (goals=1), global details=loopx doctor——global 一词明确区分了「doctor 是全局面板」而不是同范围复述。
• doctor.py:1190：健康行后追加 registry=…, goals=…, counts=…。

对主干的风险

测试-only 意义上没有风险面外的改动：health 判定、route 解析、Goal 选择、权限与 provider 行为都没动，前端也没有消费 runtime_projection_routes（作者已说明，且我在 diff 中未见前端路径）。

一条 P3（F1，非阻塞）：同一个值出现了两个名字。status markdown 打的是 activation_filter:（取自 goal_projection.scope，而 collection.py:198 把该 scope 设为 activation_filter.value），而同一份 diagnostics payload 暴露的字段名是 activation_state_filter（runtime_projection_route.py:700-702）。读者对照 JSON 与 Markdown 会以为这是两个不同的过滤条件——而消除这类 scope 歧义正是本 PR 的目的；smoke 也没有断言那个 markdown 标签，所以不一致不会被发现。建议统一命名（markdown 用 payload 的字段名，或把 payload 字段改成与 envelope 一致），并在某个 smoke 里断言该标签。

另外记录一条不是问题的点：payload 现在带 registry/runtime_root 绝对路径。status 早就在打印 registry 路径（status_markdown.py:60），所以这与既有输出面一致；只是提醒操作者不要把 doctor/status 的 JSON 直接贴进公开产物。

我的整体评价

结论 APPROVE（本 PR 由维护者本人所有，GitHub 不允许正式自我批准，故以 COMMENTED 评审记录批准结论）。改动方向准确：把「范围」作为诊断输出的一部分，让「计数相同但范围不同」这件事在操作面上可区分、可归因；实现上复用了 collector 已有输入与 status 既有 envelope，没有引入第二套权威或热路径开销。我跑了三个真实入口 smoke（含新加的两注册表对照与 markdown 断言），并确认接口预算未被顶破。回退成本是几个字段与两行文本。

唯一 P3 是同一值的两个命名，非阻塞但值得顺手统一。

English verdict: APPROVE - exact head da75ad8; the route-health diagnostics now carry the registry, runtime root, goal filter and activation filter they actually used, status reuses its existing envelope to add goal_count, and doctor labels itself the global drill-down with registry/goals/counts, so "equal counts, different scope" becomes distinguishable. I ran the three real-entrypoint smokes (refresh-state shared-runtime projection, shared-runtime material projection, hot-path interface budget) - all pass, with the new assertions covering same-route agreement, registry inequality between the two surfaces, and an unrelated registry whose equal goal_count is still separable. One non-blocking P3: the same activation value is rendered as activation_filter in status markdown but exposed as activation_state_filter in the payload.

[Job28703]

Independent review at exact head da75ad88e0bba892f8b495ca38ad7edf872fbd27 — English verdict: APPROVE.

Verified by running all three modified smokes plus the budget suite at this head:

• refresh-state-shared-runtime-projection-smoke passed — including the new counter-example arm that matters most here: two registries with equal goal_count but healthy vs lagging outcomes are distinguishable only by provenance (registry field), which is exactly the "equal counts ≠ equal scope" fix.
• hot-path-interface-budget-smoke ok; shared-runtime-material-projection-smoke passed.
• pytest tests/control_plane/test_cli_output_budget.py — 22 passed, confirming "no budget changes": the new (goals=N) suffix and activation_filter line are outside every budgeted surface.
• Scope metadata is additive (.get() consumers; schema_version stays runtime_projection_route_diagnostics_v0); the doctor no-registry fallback dict (loopx/doctor.py:891-901) gained the same four keys, so the shape doesn't fork; status reuses the existing envelope instead of duplicating scope (status/collection.py:164).

One observable-output change to be aware of on merge: runtime_projection_routes: healthy=X lines now render as healthy=X (goals=N) and the unhealthy suffix changed from details= to global details=loopx doctor — in-repo consumers are updated in this diff.