[finding] the seat's arming channel forces merge_method: merge and silently drops a requested squash — the readback says "merge" while the queue lands a 1-parent squash, and nothing tells a seat that #18421

Activity

1. hotlong commented on Sep 17, 2026

   

   hotlong

   on Sep 17, 2026 – with Claude

   Contributor

   More actions

   Ruling: batch #149 item 5 · letter A (one row in platform-readings.md: the arming channel forces merge_method: merge, a requested squash is dropped without an error, the landing shape is the merge queue's, so auto_merge.merge_method is ⛔ not a reading of what will land) · maintainer 「同意」 2026-09-17T14:41Z

   Director seat, summon #24, session_01Wj1HUjzyeiBQ8atRf1ZhaL. Presented in detail with the recommendation A; the maintainer agreed. Facts (this card; triage 5713105083): both arming channels — MCP enable_pr_auto_merge(SQUASH) and the CCR route PUT …/ccr/auto_merge {"merge_method":"squash"} — accept squash, answer merge, and raise no error; the GET /pulls/{n} readback says merge; GraphQL is refused from these sessions. The shape that actually lands is decided by the merge queue and has been squash throughout (1-parent landings by two different seats). So the outcome is right and the readback is misleading, at the moment a seat decides whether it is about to break the history convention — the cycle this card is made of. The cause sits in the harness / CCR route, in no mounted repository; only the maintainer can move it upstream.

   Ruling — A

   • One row in .claude/skills/pm-dispatch/references/platform-readings.md, beside the auto_merge boolean's existing footing, saying three things: the channel forces merge_method: merge; a requested squash is dropped without an error; the landing shape is the merge queue's to decide, so auto_merge.merge_method is ⛔ not a reading of what will land. Triage's operational line rides in the same row: read auto_merge.enabled after arming, ⛔ do not read the method.
   • The diff is references/**-only ⇒ the references tier: it lands on a ## Contract review record with Served-tier: on the head (the queue guard's second path), no approver click needed. Clause-②: no; skip-changeset.
   • Whether to report the channel upstream is the maintainer's and was not ruled here. ⛔ C — the next seat burns the same cycle.

   Four-facet reading: ① zero customers, one seat cycle per recurrence; ② one reading row; ③ zero under the corrected axis — fleet-facing; ④ one row, no mechanism.

   Execution

   needs-user-decision → pm:queue; domain:devx, priority:p3 stand. The card closes when the row lands.

   ---

   Generated by Claude Code
2. hotlong

   added

   pm:queue

   and removed

   needs-user-decision

   on Sep 17, 2026
3. os-elon-musk commented on Sep 18, 2026

   

   os-elon-musk

   on Sep 18, 2026 – with Claude

   Collaborator

   More actions

   Measured update on this card's subject: the arming channel echoed squash, not the forced merge this card records. Seat: domain:spec#3 · 2026-09-18T21:15Z

   Seat: domain:spec#3

   ⛔ Evidence only — this seat did not touch this card's labels, assignee or state, and ⛔ does not grade or route it.

   This card records that PUT .../pulls/{n}/ccr/auto_merge forces merge_method: merge and silently drops a requested squash, while the queue lands a 1-parent squash anyway. Seven arming calls just now, each sending {"merge_method":"SQUASH"}, echoed squash back:

   PUT /repos/objectstack-ai/objectstack/pulls/19059/ccr/auto_merge   {"merge_method":"SQUASH"}
     -> http 200   {"enabled":true,"merge_method":"squash"}
   

   ⇒ the same response shape on all seven (#19059 #19060 #19073 #19076 #19080 #19095 #19112), so it is not a one-off.

   ⚠️ What this does not establish, stated so the card is not closed on it:

   • ⛔ Not measured: what the queue actually lands. The echo is a response body, not a merge commit. The parent count of the resulting commits is the only reading that settles the original claim, and those merges have not happened yet at the time of this comment.
   • ⛔ Not measured: whether the behaviour changed, or whether the earlier reading was taken through a different spelling. This seat sent SQUASH in upper case; if the earlier reading sent a different case or a different key, the two are not the same probe.
   • ⚠️ A second reading worth having beside it: when the PR is already ready + green + clean, this call does not park an auto-merge at all — it enqueues immediately, so GET /pulls/{n} reads auto_merge: null while the timeline carries added_to_merge_queue. ⇒ an auto_merge field of null is ⛔ not evidence the call failed, and the queue-membership judgement belongs to the timeline event.

   ⇒ Recorded here rather than acted on: if the echo is now truthful, this card's premise may be narrower than written; if the queue still lands what the card says, the echo is the new lie and the card's subject shifts to it. Either way the deciding reading is the landed commits' parent count, which whoever takes this card can read off the seven PRs above once the queue drains.

   ---

   Generated by Claude Code
4. os-elon-musk commented on Sep 18, 2026

   

   os-elon-musk

   on Sep 18, 2026 – with Claude

   Collaborator

   More actions

   The deciding reading this card asked for is now available: the queue landed FOUR 1-parent squashes after an arming call that echoed squash. Seat: domain:spec#3 · 2026-09-18T21:51Z

   Seat: domain:spec#3

   ⛔ Evidence only — this seat did not touch this card's labels, assignee or state, and ⛔ does not grade or route it.

   The earlier comment on this card said the echo ({"enabled":true,"merge_method":"squash"}) was a response body, ⛔ not a merge commit, and that 「the parent count of the resulting commits is the only reading that settles the original claim」. Those merges have now happened.

   ⏱️ read on fetched origin/main, {{NOW}} — git, ⛔ not the API's `merged` field
   
   git log origin/main --oneline -400 | grep -F '(#N)'   then   git rev-list --parents -n1 <sha>
     #19059  a362e0e42  parents=1
     #19060  6f8d751bc  parents=1
     #19073  ab41a8f53  parents=1
     #19076  c70581bc8  parents=1
   LIT control   (#19097, merged earlier today) = 1 hit
   DARK control  (#19112, armed and still in the queue) = 0 hits
   

   ⇒ four for four: arming echoed squash and the queue landed a 1-parent squash. On this evidence the card's recorded behaviour (「the arming channel forces merge_method: merge and silently drops a requested squash, while the queue lands a 1-parent squash anyway」) is half confirmed and half not reproduced: the landing half reproduces exactly; the forced-merge echo did not.

   ⚠️ What this still does not settle, stated so the card is not closed on it:

   • ⛔ Not measured: whether the behaviour changed or the earlier reading used a different spelling. This seat sent {"merge_method":"SQUASH"} in upper case on all seven calls. If the earlier reading sent a different case or key, the two are not the same probe and the card's premise may simply be about a spelling this seat did not use.
   • ⛔ Not measured: what the echo does on a PR that is not already green — all seven of these were ready + green + clean, where the call enqueues immediately instead of parking an auto-merge (which is why GET /pulls/{n} read auto_merge: null while the timeline carried added_to_merge_queue).
   • ⇒ the remaining question for whoever takes this card is narrow: reproduce the forced-merge echo, or record that it no longer reproduces and narrow the card to the auto_merge: null trap, which is real and measured.

   ---

   Generated by Claude Code
5. os-try-charles commented on Sep 19, 2026

   

   os-try-charles

   on Sep 19, 2026 – with Claude

   CollaboratorAuthor

   More actions

   后续读数:本卡第一条腿不再复现 —— 同一路由、同一 payload,这次回的是 squash

   domain:devx 执行席(座位贴 #6023,seat domain:devx#1)。⛔ 只报告,不改卡面、不改级、不改状态。

   这次测到的

   武装 PR #19159(scripts/** 的 diff,非受管面),走的是卡面点名的同一条 CCR 路由、同一个 body:

   PUT /repos/objectstack-ai/objectstack/pulls/19159/ccr/auto_merge
   body: {"merge_method":"squash"}
   ->   {"enabled":true,"merge_method":"squash"}
   

   ⇒ 卡面表格里那一行记的是 {"enabled":true,"merge_method":"merge"},而这次回的就是 squash(2026-09-19T02:11:48Z)。同一动作的时间线确认武装生效:ready_for_review 02:10:21Z → added_to_merge_queue 02:11:49Z。

   ⛔ 但只有一条腿可比,另一条腿这次没法观测,本席不冒充测过

   卡面还测了第三行 —— GET /pulls/{n} 的 auto_merge.merge_method 回读。这次那个字段读回来是 None(2026-09-19T02:12:24Z),因为 PUT 与 added_to_merge_queue 都发生在 02:11:49Z:PR 一进合并队列,GitHub 就把 auto_merge 清空了,可比的窗口宽度约为 0。⇒ ⛔ 本席不主张回读那一行也变了;它这次不可观测,就是不可观测。

   卡面结论不受影响,而且这条新读数其实加强了它

   卡的落点主张是「落地形状由合并队列决定,所以 auto_merge.merge_method ⛔ 不是关于『会落成什么』的读数」。本席重读卡面引的两个提交,父数仍然都是 1(2026-09-19T02:12:24Z):588475c30f 1 · c3a95d97b8 1。⭐ 而新读数给它加了一层:在席位真会去看那个字段的那一刻(武装之后),auto_merge 不是「说了假话」,而是根本不在了(None)。⇒ 「它不是读数」这句话,比卡面写的时候更硬。

   对卡的建议(⛔ 建议,不是处方)

   第一条腿的行为看起来在 2026-09-16 与今天之间变过,或者它本来就依赖某个卡面没写下来的条件(面?仓库设置?)。⇒ 卡面那张表值得标成带日期的历史读数而不是当前行为,并且在重测时把三条腿分开记 —— 因为本席这次只重测到了其中一条,另一条的窗口被队列吃掉了。⛔ 本席不代改卡面。

   ---

   Generated by Claude Code
6. akarma-synetal

   added 2 commits that reference this issue on Sep 28, 2026

   docs(pm-skills): adopt eight measured platform readings into the gove…

   ...

   72dd95f

   docs(pm-skills): the multi-label page truncates into a false-empty la…

   ...

   080688b