Skip to content

Design: what P-33 and P-34 showed of closing references, an issue's content time and a timeline's count #240

Closed

Description

@don-petry

Part of #3. What trials P-33 and P-34 (trials/P-33.json, trials/P-34.json, #184) showed that the design does not yet say, each a line for §4.1 or §3.10, none a change to the Kernel:

  1. A pull request to a base that is not the default branch carries no closing reference at all (P-33: the merged and the open case on trial/p-33-base both read closingIssuesReferences empty, and the issue's timeline said willCloseTarget: false for them). §6's fallback is taken: the issue Lifecycle serves the default branch only. §4.1 already says GitHub serves a stacked pull request when it retargets it at the default branch; one sentence should say that a pull request to another base is never linked, and that GitHub enforces it, not the port.

  2. The reverse read from the issue is no substitute for the pull request's own closingIssuesReferences. CrossReferencedEvent.willCloseTarget is false for a closed-unmerged pull request that keeps its reference while the issue is open, false for every pull request once the issue is closed, and a keyword link raises no ConnectedEvent. §4.1 reads the join "at every pass": the port reads it from each pull request (the forward read), and §3.10's facts port should say so.

  3. lastEditedAt and editor move for the body alone (P-34); the title's change is only the timeline's RenamedTitleEvent. content_changed_at is the newer of the two, which the port computes; and the first body edit leaves two userContentEdits nodes, the older with editedAt equal to the issue's createdAt (the original content). §4.1's "when the title or body last changed" should name the two reads.

  4. timelineItems.totalCount ignores itemTypes (P-33: pull request 37 reads totalCount: 2 with no node of the two types asked). The Kernel requires complete.timeline; the port must read pageInfo.hasNextPage, never judge a timeline read whole by totalCount. A fixture of a timeline read short, in the way trials/P-29.truncated.json shows a diff read short, is the gap; it belongs with the ports' stories.

  5. The issue's close trails the merge by a second or two (mergedAt 21:18:36Z, closedAt 21:18:37Z, the ClosedEvent 21:18:38Z), which §4.1's "whether or not GitHub has yet closed the issue" already allows; the fixture is its first evidence.

Done when §4.1 and §3.10 say 1 to 3 in a sentence each, 4 is a row in the facts port's story, and 5 is a citation. No decision record: nothing is decided, five facts are read.

Activity

  1. don-petry commented on Oct 20, 2026

    @don-petry
    ContributorAuthor

    Working rulings (overturnable in review)

    Solution Architect, delegated. On main at 9bfc813 (#241). Read from the issue, trials/P-33.json and trials/P-34.json, design §4.1 (the issue Lifecycle: the stacked pull request at line "GitHub serves a stacked pull request when it retargets it at the default branch", the join "at every pass", "when the title or body last changed", row 1 "whether or not GitHub has yet closed the issue"), §3.10 (what the ports read whole, and the complete flags), §6 (P-29's truncated read as the form of a short-read fixture), sim/grounding.py (the comments #241 left on the four rules), and kernel/issue.py (complete.timeline, is_linked, content_changed_at).

    What this story decides. Five facts two trials read that the design does not yet say, each a sentence where the design shapes the read that found it. Nothing is decided; the design is brought to what GitHub does.

    1. §4.1 gains three sentences. After the stacked-pull-request sentence: "A pull request to a base that is not the default branch carries no closing reference, whatever its description says, so it is never linked; GitHub enforces this, not the port (P-33)." At the join read "at every pass": "The join is read from each pull request's own closing references, never from the issue's timeline, whose cross-reference events say a pull request will close the issue only while it is open and unmerged, and nothing once the issue is closed (P-33)." At "when the title or body last changed": "The body's last edit is the issue's lastEditedAt; the title's is the timeline's rename event; the content's time is the newer of the two, which the port computes, and the first edit of a body leaves an edit of the original content's time beside it (P-34)."

    2. §3.10 gains one sentence, where the facts port reads a list whole: "A timeline's totalCount counts every item whatever types were asked, so a timeline read whole is told by its page's end, never by that count (P-33)." The fixture of a timeline read short, in the form of trials/P-29.truncated.json, is owed by the facts port's story in slice 1; this story names it there as a row and does not run it.

    3. §4.1's row 1 gains a citation: "(P-33: the issue closed a second after the merge, and its closed event two)".

    4. Nothing else changes. sim/grounding.py already says each of the five in its comments; the rules' words stand. kernel/issue.py already requires complete.timeline and reads the join from each pull request. §6's rows do not record outcomes. No schema, no example, no test.

    5. Record. None: five facts are read, nothing is decided. A row on decision 0028, "the issue Lifecycle serves the default branch's pull requests alone, as GitHub enforces", held by trials/P-33.json through tools/check_trials.py and the content-pinning test #241 added.

    6. Files. Edited: the design (§4.1, §3.10), decision 0028 (one row). Nothing new.

    7. Gates. Export verify (tools/check_context.py holds every path the sentences name); a fresh Claude review and a security pass; no AGY and no Muse (a docs change: Muse reads Kernel and hook pull requests only, the Maintainer, 2026-10-20, which the PR body says); the PR says "Closes #240".

    8. Out of scope. The facts port and its short-read fixture (slice 1); any change to the Kernel; a rerun of either trial.

    Needs the Maintainer? None.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions