Skip to content

A re-review is given none of its Lens's open Findings, so a thread closes on the silence of one more review #856

Open

Description

@don-petry

What happens

Since the fix for "A gating thread the reviewer opens is closed only by a person's hand: no Role answers it, no Intent rules on it and no queue App resolves it" (#833), a Lens's thread is the Kernel's to resolve once a later review by that Lens, of a head pushed after the thread opened, records nothing on it. That review is not told of the earlier Finding: harness/read.py hands a review no Findings, so its silence is one more sample of the same Lens and not a judgement of that Finding.

So a blocking or major Finding holds only while each later review raises it again, and a push that leaves the diff as it was, a merge of the base, buys one more review. The Run ceiling bounds it at one Run a push, and a person's action renews that count. A security Finding's thread is never closed this way.

The Claude security read of the fix raised it, as minor, for the Maintainer to weigh. The fix already requires a push between the thread and the review, so a Run that failed and is tried again on the same head drops nothing.

What is asked

Whether a re-review should be given its Lens's open Findings on the item, to say of each whether it still stands, so that a thread closes on a ruling and not on silence. That changes what a review Task carries, which is a model Lens's input and takes the Maintainer's own approval.

Beside it, for the same hand: kernel/gate.py's closed() says in its docstring that the queue App closes a superseded thread "for a thread whose Run's record is past retention". Its code reads the mark alone, so since the fix it closes a dropped thread too, and the words want the second case. The file is protected and was left as it was.

🤖 Generated with Claude Code

Activity

  1. don-petry commented on Nov 13, 2026

    @don-petry
    ContributorAuthor

    Beside the note on closed(), from the second Claude security read of the fix (#859), rated info: since the fix the Kernel's act never names a security Finding's thread, whatever ruling the facts carry, but kernel/gate.py still reads a Lens's thread the queue App resolved as closed on an accepted ruling whatever its category. A security Finding's thread with ruling accepted, resolved by the queue App, with an approval, answers converged, approved, act enqueue. It was so before the fix, adapters/applied.py gives a security Finding's thread no ruling, so nothing reaches it today. It is the gate's half of the same guard, and the file is protected.

    🤖 Generated with Claude Code

  2. don-petry commented on Nov 14, 2026

    @don-petry
    ContributorAuthor

    Ruled, 2026-11-14

    The Maintainer chose the full fix and to keep the release as it is meanwhile; the Factory Builder ruled its steps.

    The gate. These three steps land before author_role is named, beside "The simulator has a pull request the Factory did not open wait for its author, before author_role is named (slice 2, after S9b)" (#824), "The implementing rounds' bound is a brake, and moves into kernel/brakes.py before the author Role is named" (#874) and "The facts port reads a Factory-opened draft's required-check rollup" (#878).

    Step What Route
    1 adapters/texts.py reads the first comment of each of the reviewing Lens's own open gating threads. A thread's text is outside text: it reaches a Task only as trust-marked data, never a decision and never a reason code Factory
    2 harness/read.py hands a re-review its open Findings, by id and text; harness/apply.py refuses a Result that gives no disposition on a Finding it was shown; the reviewer's instructions gain the ask. Development-set cases exercise it, never held-out content, and the pull request says what stays unevaluated. A Release is cut after it merges By hand, one pull request
    3 After that cut: a thread closes on a later review's accepted ruling alone, and silence releases nothing. A record takes the place of decision 0212's row on a dropped thread Factory

    Until step 2 is live, a later review's silence releases a thread as decision 0212 has it. Step 3 does not land before step 2 is cut: nothing but a person would close a thread in between.

    Done when

    • Step 1 merged.
    • Step 2 merged, and a Release cut that carries it.
    • Step 3 merged, with its record.
    • author_role is still null when the third lands.

    🤖 Generated with Claude Code

  3. added 20 commits that reference this issue on Nov 14, 2026
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