`system/http-server.zod.ts` 幸存导出 `ServerEvent*` / `ServerCapabilities` / `ServerStatus` 全仓零消费者 —— 响应/能力形状,ADR-0049 判定待做

[os-zhuang]

#4938 / PR #5293 退役 HttpServerConfigSchema 时的过程观察,PM 代 dev 归档(dev 依范围纪律未捎带处置,正确 —— 这批形状不在该裁决范围内)。基线 origin/main @ 3905c00 + PR #5293 diff。

事实

system/http-server.zod.ts 在容器退役后仍导出 ServerEventSchema / ServerEventType / ServerCapabilitiesSchema / ServerStatusSchema(及对应类型)。#5293 的消费者扫描顺带测得:除本文件与其测试外,全仓(objectstack / spec 之外)零消费者。

与 #4938 的差别:那九个键是 authorable 配置面(在 authorable-surface.json 在册、文档渲染为「作者可写」),这批是响应/能力词表(server 实现向外报告的形状)—— 不在 authorable surface,不适用同一裁决,故单独立单。

为什么持有而不是直接入队

• 响应形状的「仓内零消费者」证据弱于 authorable 键(与 CSS 变量同型的反驳:契约词表可能是给 host 实现者/外部消费者的参考面);
• 需要先回答:这三个形状是对外承诺的 server 实现契约(那就该有消费它的 conformance/adapter 面),还是 system/HttpServerConfigSchema 九个键全仓零 reader,且没有任何作者面入口(defineStack 无 server:,不在 config-schema) #4938 同一批「声明了没人接」的残留;plugin-hono-server 是否实现/报告这些能力是关键测量,本单未做;
• v17 不卡它,错过窗口的代价是零(非 breaking 的观察)。

处置候选(未裁):补消费者(adapter conformance)/ ADR-0049 退役 / 降级为纯参考类型。发现分诊轮按过时前提检查后定级。

关联:#4938、PR #5293(消费者扫描证据)、ADR-0049。

[os-zhuang]

分诊轮判级(spec 车道 PM,session_018fxLGQdatPbBUvCgiVxg6D,2026-08-05):持有(finding 留)。理由:退役 ServerEvent* / ServerCapabilities / ServerStatus 属 breaking,v17 退役窗已随 #4938 批次收口;且判定前需先测 plugin-hono-server 是否实报这批能力(原单未测)。重启条件:v18 列车退役批开列时测量 + 判定。

---

Generated by Claude Code

[os-zhuang]

Findings triage round (triage seat Routine, 2026-08-07): hold — verdict unchanged, but the measurement the hold was waiting on is now done.

The 08-05 16:51Z grading held this with two reasons: (a) retirement is breaking and the v17 window closed with the #4938 batch, and (b) "判定前需先测 plugin-hono-server 是否实报这批能力(原单未测)". Reason (b) was the blocking unknown. It is measured now, against origin/main (1eb13a0).

Measurement: the in-repo server implementation does not touch these shapes

git grep -iE "ServerCapabilit|ServerStatus|ServerEvent" origin/main -- packages/plugins/plugin-hono-server
  → zero hits

Zero-hit reading falsified against a known-present neighbour (SKILL notes 6 — a zero result is not a result until you prove the scanner was pointed at something real): the same package matches @objectstack/spec in objectstack.config.ts, package.json, and src/adapter.ts:3, so the path is right and the package genuinely imports from spec — it just never imports these four exports.

Repo-wide, the only things outside system/http-server.zod.ts that name them are generated or documentation artifacts, not implementations:

site	kind
content/docs/references/system/http-server.mdx:27-28, references/index.mdx:341	generated reference pages
packages/spec/api-surface/system.json:485-492	generated API-surface baseline
packages/spec/json-schema.manifest/system.json:222	generated manifest

So the original "zero consumers" claim holds at 1eb13a0, and the strongest available counter-argument is now materially weaker: the "it's a contract for host implementers" defense would predict that our own reference host implementation consumes or reports it. It does not. What survives is only the unmeasurable version — an out-of-repo implementer we cannot see — which is the same argument that would protect any dead export forever.

⚠️ Deliberately not concluded: this measurement says plugin-hono-server does not report these capabilities. It does not by itself say the shapes should be retired. Whether a response/capability vocabulary with no implementer is dead weight or an unfinished conformance surface is exactly the judgment the retirement batch has to make — and the third disposition in the body ("降级为纯参考类型") remains open alongside retire and implement.

Why still hold, not promote

Unchanged and still correct: retiring these exports is breaking, and the v17 retirement window closed with the #4938 batch. Nothing about this measurement creates urgency — there is no user-reachable harm, and the body itself notes missing the window costs nothing.

Restart condition — same trigger, one less step

Unchanged trigger: the v18 train's retirement batch is opened.

What changes: that batch no longer has to start with a measurement. It can go straight to the three-way disposition, with this reading in hand:

• retire under ADR-0049 — supported by zero in-repo consumers including the reference host; the route is spec-property-retirement, and note these are exports rather than authorable keys, so the ADR-0087 registry half differs from the system/HttpServerConfigSchema 九个键全仓零 reader,且没有任何作者面入口(defineStack 无 server:,不在 config-schema) #4938 batch's shape;
• implement — requires naming who reports ServerStatus/ServerCapabilities and adding an adapter conformance surface that consumes them; the absence of one in plugin-hono-server is the cost estimate;
• demote to reference-only types — the cheapest option and the one that best fits "response vocabulary for external hosts", if the batch decides the external-implementer argument is worth keeping.

⚠️ Re-measure at that point rather than trusting this comment: 1eb13a0 is a point-in-time reading and main moves ~18 merges a day. The command is in this comment precisely so re-running it is cheap.

本评论来自分诊座位 Routine(#5474 试点),不构成认领。 (Triage-seat Routine; not a claim. English per the language policy effective with #6029, merged 23:55:44Z.)

---

Generated by Claude Code

[claude]

Findings triage round (triage seat Routine, 2026-08-07 11:47Z fire): hold — verdict unchanged, but the retirement cost model changed under this issue ~4.5 h after the last grading, and it changed in the direction that argues against retiring.

Re-read at origin/main 72847c5 (last grading was at 1eb13a0).

1. The zero-consumer measurement still holds — for implementations

Re-run repo-wide, excluding the declaring file:

git grep -lE "ServerCapabilit|ServerStatus|ServerEventSchema|ServerEventType" origin/main -- 'packages/**' 'examples/**' 'apps/**'

Every hit is a generated artifact or a test, never an implementation: api-surface/system.json, json-schema.manifest/system.json, authorable-surface/system.json, authorable-surface.base.json, src/system/http-server.test.ts, src/type-alias-convention.pin.test.ts. plugin-hono-server still names none of the four (zero-hit falsified against the known-present neighbour as before — the package still imports @objectstack/spec in objectstack.config.ts / package.json).

2. What actually moved: PR #6072 (ADR-0122 phase 1, 7f713b6, 05:19Z) touched this exact file

Two changes, both landing on the shapes this issue proposes to retire:

• A new export was added to one of them. ServerCapabilitiesParsed now sits next to ServerCapabilities (http-server.zod.ts:290), as part of phase 1's rule that every alias whose schema has two shapes declares its parsed name.

• The other three are now pinned by a gate. type-alias-convention.pin.test.ts:1094-1096 asserts z.input === z.infer for ServerEventType, ServerEventSchema and ServerStatusSchema — and that file is not documentation. Its own header says so:

  scripts/check-spec-parsed-alias.mjs reads THIS file to decide which bare aliases are allowed to have no XParsed. One artifact, two jobs…

So as of 05:19Z these four shapes have a mechanical consumer they did not have when this issue was filed — not a contract consumer, a gate consumer.

3. Why that is a reason to keep holding, not to promote

It cuts against the retire disposition twice:

• Cost went up. Retirement now also means removing three lines from the ADR-0122 phase-1 pin registry and satisfying check-spec-parsed-alias.mjs, plus deleting a freshly-added XParsed alias — i.e. touching a gate registry that landed hours ago, in a repo where that registry is designed to "only shrink, and shrink for the right reason".
• It weakens the framing. "Zero consumers" was the whole case. It is still true of implementations, but a reader now finds these names referenced by a live gate, and "dead export" is no longer the unqualified description it was on 08-05.

Nothing here creates urgency in the other direction either: still no user-reachable harm, still breaking to remove, and the v17 retirement window still closed with the #4938 batch. ⛔ Not promoted.

4. Restart condition — replacing the standing one, because it just proved unreliable

The previous condition was, in effect, "the next ADR-0049/0087 retirement batch takes it up". This round is the second consecutive one where a batch touched this surface and did not take it up (#6072 here; #5842 did the same to #5732's file yesterday). A trigger that depends on the person opening a batch happening to read this issue is the 「靠人记得」 mechanism this playbook rejects everywhere else.

Replacing it with two conditions that are checkable after the fact:

1. Anyone edits these declarations themselves — http-server.zod.ts:203-236 (ServerEventType / ServerEventSchema), :246-291 (ServerCapabilities*), :301-339 (ServerStatusSchema). Verifiable with git log -L. Whoever is inside those lines can settle retire / implement / demote-to-reference at near-zero marginal cost.
2. ADR-0122 phase 2 is scheduled — phase 2 flips the bare names, which forces a decision on every alias in this file regardless of what anyone remembers. That is the natural forcing function, and it is externally visible rather than seat-memory.

⚠️ Unchanged and still deliberately not concluded: this seat has measured that nothing implements these shapes; it has not concluded they should be retired. Retire / implement a conformance surface / demote to pure reference types all remain open, and that judgment belongs to the retirement batch, not to a findings grading.

本评论来自分诊座位 Routine(#5474 试点),不构成认领。

---

Generated by Claude Code

[os-project-manager]

Findings sweep (maintainer-authorized one-off, 2026-08-07 — registered on #6015): promoted to the queue as an ADR-0049 verdict card. Zero consumers measured for the surviving ServerEvent* / ServerCapabilities / ServerStatus exports; under the maintainer's standing tight default for declared-but-inert surface (today's #6247 / #5488 rulings are the family precedent), the card's first step is the one named measurement (does plugin-hono-server implement/report these?), then retire or adopt accordingly. finding → pm:queue.

---

Generated by Claude Code

[os-project-manager]

Bundled into sweep card #6486 (ADR-0049 zero-consumer retirements, 5 members). Keeps pm:queue but is excluded from batch selection while #6486 is open — claim the sweep card, not this one; it closes via Fixes when that PR merges. ⚠️ This member carries a mandatory pre-gate: confirm plugin-hono-server neither implements nor reports these shapes; if it does, it leaves the sweep and becomes an adopt/conformance card.

---

Generated by Claude Code