fix(status): print the scope beside the projection-route verdict

[huangruiteng]

The disagreement this fixes

loopx doctor and loopx status both print a healthy value for runtime projection routes. They read different denominators, and neither line said so.

Reproduced on this machine before the change:

$ loopx doctor    # global registry, every route
- runtime_projection_routes_healthy: `False`
  {"healthy": 0, "ready": 2, "single_runtime": 29, "missing": 13, ...}

$ loopx status    # project registry, no goal filter
- runtime_projection_routes: healthy=True

$ loopx status --goal-id loopx-meta
- runtime_projection_routes: healthy=True

Both values are arithmetically correct. doctor collects the shared global registry and marks the fleet unhealthy because 13 goals report source_registry_missing; status scopes the same collector to its own registry, and further to one goal under --goal-id. Pointed at the same registry, status already agreed with doctor:

$ loopx status --registry ~/.codex/loopx/registry.global.json
- runtime_projection_routes: healthy=False

So the operator-visible contradiction was the missing scope, not a wrong number: an owner running doctor after status saw red next to green with nothing to reconcile them.

Change

• runtime_projection_routes in the status projection gains the route count it was computed over.
• The status renderer prints registry, goal filter and route count beside the flag.
• The doctor line prints registry, route count and the full counts map, so a red reading is actionable without --format json.

After:

- runtime_projection_routes: healthy=True (goals=1)
- runtime_projection_routes_healthy: `False` (registry=`<global>`, goals=`44`, counts=`{"missing": 13, ...}`)

The status line carries only the denominator, because the agent-facing markdown
rows allow 32 extra characters each; doctor names the registry and the full
counts map on the unhealthy reading.

Interface budget

runtime_projection_routes grows by exactly one key. This is a deliberately budgeted hot-path payload, so the scope is limited to the denominator the flag needs rather than inlining the full diagnostics. The interface-budget smoke passes with the dashboard status payload at 19155/19500 chars.

Validation

• examples/control_plane/refresh-state-shared-runtime-projection-smoke.py — added the reconciliation invariant: status and doctor must agree on healthy and goal_count over the same registry, and doctor's markdown must name its scope.
• examples/control_plane/shared-runtime-material-projection-smoke.py, examples/control_plane/hot-path-interface-budget-smoke.py — updated the assertions that pinned the previous lossy {"healthy": ...} shape.
• pytest tests/ -k "status or doctor or projection_route" — 311 passed, 1 skipped. Two test_task_graph_topology.py sqlite-current/sqlite-missing failures are environmental and reproduce unchanged on origin/main.
• loopx check — clean.

Boundary

Public-safe: no private state, credentials, local paths or raw evidence. No control-plane decision, permission or scoring behavior changes; only the projection's self-description and the two rendered lines.

[huangruiteng]

动机

本车道在 2026-09-17 的旅程资格里记录了同一会话内的矛盾：`loopx status` 报 `runtime_projection_routes: healthy=True`，`loopx doctor` 报 `runtime_projection_routes_healthy: False`，两个面都不说明各自读的是哪个范围，也没有对账规则。这是典型“读数缺分母”的问题：两者共用同一个 `healthy` 键，但一个读项目内 goal 路由、另一个读共享全局 registry。

改动思路

不做仲裁、不改判定逻辑，而是让每个面把“这个布尔值是在多少条路由上算出来的”一并说出来：`status` 在既有行尾补 `(goals=N)`，受 32 字符 agent 输出预算约束；`doctor` 额外给出 registry 路径、goals 数与完整 counts 映射，便于直接对账。两侧渲染都由 smoke 断言。

具体改动

• loopx/control_plane/status/collection.py：`runtime_projection_routes` 增加 `goal_count`，取自同一路由诊断，缺省为 0。
• loopx/presentation/renderers/status_markdown.py：`healthy={bool}` 后按需附 `(goals=N)`，并加 docstring 说明为何 registry 路径留给 doctor。
• loopx/doctor.py：`runtime_projection_routes_healthy` 行补 `registry=`、`goals=`、`counts=`（JSON、键排序）。
• 三个 smoke 由“等于 {"healthy": bool}”改为断言 `healthy` 与 `goal_count`，并新增 status/doctor 一致性断言与两处渲染文本断言。

对主干的风险

产品行为变化限于新增的展示字段与 renderer 文案：原先断言精确字典形状的 smoke 已同步改为逐键断言，避免把新增字段读成回归。scope 语义没有被重新定义，只是显式化；未改变任何门禁或判定结果。hot-path 输出预算 smoke 已在该 head 通过，说明 `(goals=N)` 未超出 agent 面行预算。

我的整体评价

以“补分母、两边都点名范围”而不是“改一个面附和另一个面”的方式修掉了我此前记录的 operator-truth 缺口，方向和最小性都对，且用 smoke 把两面对账固定下来。建议在该 head 的必过检查转绿后合并；合并后即可关闭依赖它的 todo_845adbcd6703。

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

Head: 0061271

English verdict: APPROVE

[huangruiteng]

动机

loopx doctor 与 loopx status 都会为 runtime projection routes 打印一个 healthy，但两边的分母不同，而且两边都没说自己用的是哪个分母。复现出来的结果就是：doctor 报 runtime_projection_routes_healthy: False（全局 registry，44 条路由里 13 条 source_registry_missing），同一次会话里 status 报 healthy=True（项目 registry、且可再按 --goal-id 收紧）——两个数在各自口径下都对，读的人却会以为其中一个错了。本车道在本地管家旅程里如实记下了这个分歧（todo_845adbcd6703）。

改动思路

让每条打印自己那一层：status 行只带分母 healthy=False (goals=44) / healthy=True (goals=1)，registry 名与完整计数留给 doctor。这样既点名“这是哪一层的健康度”，又不去动 agent-facing 的输出预算——这一点很关键，因为第一版把 registry=、goal=、goals= 都塞进 status 行后，markdown 行涨了 119–125 字符，被 examples/control_plane/cli-output-base-head-differential-smoke.py 以每行 32 字符 / 64 字节的额度拒掉。

具体改动

• loopx/presentation/renderers/status_markdown.py、loopx/control_plane/status/collection.py、loopx/doctor.py：status 侧输出收紧为分母，doctor 侧保留 registry 与完整 counts map。
• examples/control_plane/hot-path-interface-budget-smoke.py、refresh-state-shared-runtime-projection-smoke.py、shared-runtime-material-projection-smoke.py：把新契约与该头的输出预算一起钉住。

对主干的风险

这是面向操作者的输出改动，风险是“为了说清楚而把 agent-facing 行撑大”，而第一版正是踩了这个坑，所以修正版把信息量压到分母上、并用差分 smoke 把额度钉住。本次复审在更新后的 head 上实测（该 head 只是把 origin/main 合进来，六个被评审文件在旧 head 与新 head 之间逐字节一致，已用内容哈希核对）：

• uv run --extra test python examples/control_plane/cli-output-base-head-differential-smoke.py → ok base_ref=origin/main base=102 candidate=102 candidate_only=0 review_required=0
• uv run --extra test python -m pytest -k "status_markdown or cli_output or projection_route" -q → 85 passed, 9708 deselected

我的整体评价

把一个真实存在、且会让操作者得出相反结论的口径分歧变成“各自说明自己那一层”，并用输出预算门把“说清楚”和“别撑大”同时钉住，方向是对的。建议在该 head 的必过检查全绿的前提下合并。

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

Head: 8888a92

English verdict: APPROVE

[huangruiteng]

Superseded by #4636. Retained the useful route count, but a denominator alone cannot distinguish equal-sized scopes. The replacement makes doctor report its actual diagnostic registry and filter metadata, reuses status registry/Goal/activation context, and labels the global drill-down. Real CLI fixtures cover same-route agreement and different registries with identical counts but different health. Output budgets and all 19 premerge checks pass. The route-health computation and write behavior are unchanged; closing this branch to keep one merge target.