chore(pm): retire half-state row H34; H2 reports every card it named (#29654)

<!-- claude-projects-attribution: {"github_login":"huangyiirene"} -->
_Requested by **Irene** · [project
thread](https://claude.ai/code/project/chan_01TJXtQ1tVD26ecLDZxFS67x?thread=cmsg_01TJXtQ1tVD26ecLDZxFS67xCjGpvHD4tRyZ72DtPifmqL)_

Part of #29471

Clause-②: no
Metadata-protocol: none

## 维护者速读

- 改了什么:删除半态巡检行 H34(认领评论用了非冒号分隔符)。它只在 H2 已经报告的卡上出现,所以由 H2 接着报。
- 为什么改:#29471 的逐行预筛(评论 6591407246)里只有 H34 三项全过:无独占规则、线上 0
行、被留下的行完整覆盖。席位认领 6591424775。
- 风险与回滚:只动 `scripts/pm/check-half-states.mjs`;198 种状态逐一比对,H34 报的 42 种 H2
全报,H2 本身逐字不变。可整 PR revert。
- 先看:删掉的 H34 段落与巡检循环里的两行调用。
- 两分钟验证:`node scripts/pm/check-half-states.mjs --self-test`,预期 5166
条全过(主干 5202)。
- 需要你做的:无,席位落地。

## What changes

- **H34 is deleted**: its rationale banner, `CLAIM_NEAR_MISS_MARKER`,
`looksLikeClaimContent` and `nearMissClaimSeparators` (all read only by
H34), `h34ClaimShapedNonCanonicalSeparator`, its band-registry row, its
two-line call beside H2 in the sweep loop, and its self-test block.
- **H2 is untouched**: `h2AssigneeNoClaimComment` and its call are
byte-identical.
- **Prose pointers** that named H34 now name H2 (an assigned card with
no claim the marker reads) or H1 (an unassigned one): the
`CLAIM_COMMENT_MARKER` docblock, the list-item marker note, the
near-miss census overlap note, H33's blind-spot paragraph and its
self-test comment, H50's population note, the H44/H64
artefact-recognition notes and the H60 band comment. Three test names
changed (one H50 case and two BATTERY68 cases) with the same assertions.
- One H60 case compared H60's band to H34's (`familyBand('H60') ===
familyBand('H34')`). With H34 gone, it now pins `familyBand('H60') ===
'state'`, the value both had.
- No gate, row, constant or battery is added. No battery shrank, so no
pin moved.

## Why: every card H34 named is an H2 row

H34 returned null unless the card was pm-tracked (`pm:queue` or
`pm:dispatched`), was assigned, had a readable thread and had no comment
matching `CLAIM_COMMENT_MARKER`. Under those same conditions,
`h2AssigneeNoClaimComment` returns true. So by construction, every H34
row sat on a card H2 already reported. H34 only added the separator's
codepoint as a diagnosis. The sweep called both on the same thread in
the same loop.

Measured by importing both trees (`main` at d280265 and this head) and
judging 198 states: 6 label sets × 3 assignee sets × 11 threads (em
dash, en dash, hyphen, fullwidth colon, blockquoted dash, canonical
colon, dash plus colon, prose control, a null body, several near misses,
and an empty thread):

- H34 fired on 42 of the 198 states, and H2 fired on all 42 (0 misses).
- H2's verdict is identical on all 198 states before and after.

Live on objectstack: 0 H34 rows in a full sweep finished
2026-11-02T23:43Z (code at d280265), per the #29471 pre-screen.

## Verification

- `node scripts/pm/check-half-states.mjs --self-test` on 9fcce36:
**5166 cases pass** (5202 at d280265, so 36 fewer). All 36 are the H34
block.
- Every battery reads the same `n/floor` as at d280265, so nothing was
re-pinned.
- Net lines: **+33 / −381 (−348)**, one file.

<details><summary>Local gates on 9fcce36 (`dispatch-gates
--commands`)</summary>

- On head 9fcce36, `node scripts/pm/dispatch-gates.mjs --commands`
derived 47 commands. All 47 ran, each with a 900 s timeout, with exit
codes recorded before any pipe. **All 47 exited 0.**
- The `--ran` reconciliation printed: `47 derived famil(ies) accounted
for — 47 run, 0 NOT-MEASURED`.
- `node scripts/check-commit-card-trailers.mjs --range d280265..HEAD`:
exit 0.
- Prerequisite: `pnpm install --frozen-lockfile`.
- Families CI runs beyond the derived list are read from this PR's check
runs, not re-run locally.

</details>

## Acceptance notes

- What is lost is the diagnosis H34 added: the separator's codepoint and
the "re-post with `Claim:`" remedy. The card is still reported, by H2's
row "assignee set but no claim comment on the thread".
- The ownership-marker near-miss census still names a dash-written claim
as the `separator` form in its summary clause, so the spelling stays
visible on the summary line.

## Human Minimum

- [x] 1. Schemas start as Zod; `packages/spec` holds no runtime logic.
(not touched)
- [x] 2. Every name I added uses the casing for its kind. (no names
added)
- [ ] 3. No formatter ran; `pnpm lint` passes. (no formatter ran; `pnpm
lint` was not run locally, so the CI lint job decides it)
- [x] 4. One issue, one branch off an up-to-date `main`, one pull
request.
- [x] 5. A change that publishes carries a changeset, and both
declaration lines are filled in here and in the changeset. (publishes
nothing: `scripts/pm/**` only; `skip-changeset`)
- [x] 6. `packages/spec` touched: its generated artifacts are
regenerated and committed. (not touched)
- [x] 7. No hand edits to `content/docs/references/`,
`packages/*/CHANGELOG.md` or `content/docs/releases/`.
- [x] 8. Title, body and comments are in English. (except the 维护者速读
block, which the seat contract writes in Chinese)
- [x] The *Before Submitting* list in CONTRIBUTING.md is done.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

https://claude.ai/code/session_01J8VVAmM3BRQQcQSA67VE3L

Co-authored-by: Claude <noreply@anthropic.com>

Author: objectstack-fleet[bot]