Skip to content

Kernel: a Pause stops every act, not only Runs #30

Closed

Description

@don-petry

Part of the epic "Epic: nine gaps in the Kernel, found in review".

Why. With the Factory paused, or the Pause unreadable, a converged pull request is still approved and an approved one is still enqueued. The check for a Pause sits after the branch for converged in kernel/decide.py.

Done when

  • test_a_pause_stops_approvals_and_merges_too passes with @pending removed: under a Pause the Decision is defer, reason paused.
  • An unreadable Pause defers in the same way.
  • Each place the rule acts has its own mutant, and both are killed.
  • No document says not held yet beside this test.

Needs. The pull request that carries the pinned tests has merged.

Design. §4.7, Pause; I-7.

Activity

  1. don-petry commented on Oct 14, 2026

    @don-petry
    ContributorAuthor

    Advice from the Solution Architect persona, not a ruling. Asked while revising the pull request "Context for the Factory's authors: lessons, standards, decisions and the checks that hold them".

    What a Pause stops. Settled by ADR-0013 decision 3 for the three acts that move work forward: admitting a Run, issuing an approval, enqueuing. The design's I-7 and its §4.7 row now say so in that pull request. Two acts are not settled by anything decided: the status FR-24 says a paused item shows, and the dequeue decision 7 requires when an approval is lost (see "Kernel: what takes a pull request out of the merge queue").

    When an item is both paused and held because its approver is its author or last pusher: open, and both orders are safe, since neither approves.

    • (a) The Hold's reason, approver_is_author_or_last_pusher. State stays a function of the facts alone, as §4.1 and record 0002 ask, and the Maintainer learns at once what only they can fix.
    • (b) paused. One reason across the scope, but the item's State then moves with the variable, and the Hold appears only after the Pause is released.

    Recommended: (a). Holds the facts establish, then the Pause, then the act. The same order should apply to attempts_exhausted, max_rounds and run_ceiling, which kernel/decide.py reads after the Pause today. What would change this: a ruling that FR-24's "paused items show that they are paused" means the item's own reason.

    A fresh agent built this rule once in a throwaway copy, keeping the approver Hold first; nothing pinned the order.

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