Skip to content

A gate set holds a floor per criterion, read from each case's expected criterion (#692) - #699

Merged
don-petry merged 17 commits into
mainfrom
gate-criterion-floors-692
Nov 10, 2026
Merged

don-petry merged 17 commits into
mainfrom
gate-criterion-floors-692

Conversation

@don-petry

@don-petry don-petry commented Nov 10, 2026 •

Copy link
Copy Markdown
Contributor

Closes #692, "A gate set holds a floor per criterion, read from each case's expected criterion", a precondition of gate-v1.

Record 0180: docs/decisions/0180-a-gate-set-holds-a-floor-per-criterion-read-from-each-cases-expected-criterion.md, the Factory Builder's number.

What changes

  • A gate set's manifest may give criterion_bars, {criterion: [n, d]}, for criteria its Role's rubric names, with 1 <= n <= d. A development set may not.

  • A floor counts only the cases that expect a Finding on that criterion, read from each fetched case's expected (the Factory Builder's second ruling). A reviewer blind to a criterion fails its floor even though both kinds clear their bars.

  • eval-aggregate/1 gains criteria, counts only, for the floored criteria alone, and criterion_bars.

    • aggregate_verdict checks these counts against the kind's cases, passes and failures and the manifest's floors. It never fails a single report on a floor.
    • pooled_verdict sums each criterion over the pool and holds it to its floor (criterion_failed).
  • The least, as in decision 0164: each floored criterion needs FLOOR_LEAST (ten) finding cases. The finding cases on no floored criterion must be none or at least ten, since the kind's count minus the floors' count publishes them. Otherwise fetch stops with eval_floor_short before any case is asked (the Factory Builder's first ruling; the security read).

  • Residuals written down:

    • a count still moves by one case;
    • a floor is only as strong as its manifest;
    • the rubric read is the job's own checkout's;
    • the least holds within one manifest, not between two;
    • the least holds for each group, not where two overlap.

    The last two follow the Factory Builder's ruling. Differences between a gate set's manifests across commits can isolate a group under ORIGIN_LEAST #695, "Differences between a gate set's manifests across commits can isolate a group under ORIGIN_LEAST", takes both up after gate-v1, and Protect each eval set's manifest in the hooks, so a floor or mark edit asks first #696 protects the manifests in the hooks.

Built against the schema and synthetic fixtures only; no gate-set case, count or composition was read.

Reads

Reader Head read Findings Outcome
Claude review 0a3b046 major: findings beyond the floors told by subtraction; minor: record wording; info: temp folders Fixed (c25d477)
Claude security 0a3b046 blocking: the same subtraction Fixed (c25d477)
Argon (gemini-4-argon-high, 11 parts) 0a3b046 major: floors' failures beyond the kind's Fixed (8d747e6)
major: short criteria give count_differs, not eval_split_short No change: fetch stops a short floor before any report is written
Muse (muse-wonder-full-1.0) c25d477 minor ×2: a whole floor [1, 1] untested; the sentence for two floors untested Fixed (8d747e6)
Claude review and security 8d747e6 blocking: the cases-sum mutant was made equivalent Fixed (7e1aae5): line, mutant and mention dropped
major: differences between manifests across commits Residual, by the Factory Builder's ruling (91fc275); #695, #696
minor: the same for natural marks, older than this change Named in the same residual and in #695
info: the pool doesn't hold each kind's cases equal Left: older than this change, and same-set reports agree
Argon 8d747e6 major: a floor of [0, d] was held Fixed (714be8f): 1 <= n <= d
major: a single report is not failed on a floor No change: by design, as each kind's bar (decision 0164)
minor: the redundant cases sum Already fixed (7e1aae5)
Muse 8d747e6 major: floor_short untested with two floors (any→all) Fixed (714be8f): test and mutant
major: the equivalent cases-sum mutant Already fixed (7e1aae5)
Astra (gpt-6-astra) c5ccd1b major: a report checked on its first floor alone Fixed (c69c58c): test and mutant
major: every floor checked at the first floor's fraction Fixed (c69c58c): test and mutant
major: a floored criterion and the natural cases can overlap in fewer than ten Residual, as the ruling on manifests (c69c58c); added to #695
CI mutants 714be8f, c5ccd1b two mutants of the no-kinds checks hidden by the new criteria check Fixed (fa6afb1, 51ef73f): those cases null criteria

Checks

🤖 Generated with Claude Code

don-petry and others added 11 commits November 9, 2026 16:39
…d criterion (#692)

A gate set's manifest may give criterion_bars, {criterion: [n, d]}: of the case-runs of
the cases whose expected criterion is that criterion, at least n/d pass. Only a gate set
may give it. The job reads each case's criterion from its fetched case (criteria_of), and
eval-aggregate/1 gains criteria, {criterion: {cases, passed, interval}}, counts alone,
and criterion_bars, the floors applied. aggregate_verdict holds the counts to the set's
and the floors to the manifest's; pooled_verdict holds every run to the same criteria,
sums them, and holds each named criterion to its floor, criterion_failed. bars_said adds
the floors to the record's sentence.

The record is NNNN until the Factory Builder numbers it at enqueue.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#692 holds (#692)

Lane d had built #692 before it moved to this lane; its hand-over named one rule this
build lacked: a floor on a criterion the rubric does not name would only show later, as
count_differs, since no case could cite it. The manifest now refuses it at once.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…iling case's criterion, a case with no criterion (#692)

The security pass and the Claude review of d5f056a:
- a criterion passing more than it holds is count_differs, held by a case and a mutant;
- the job's pass count per criterion is now tested with a run that passes none;
- a fetched case that names no criterion, or one not shaped as one, leaves the criteria
  unread, held by a test and a mutant;
- the record states that a floor is as strong as its manifest, which no person need
  approve.
The least per criterion and whether met cases count toward a floor wait on the Factory
Builder's ruling.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…blished, each of ten cases at least (#692)

The Factory Builder's two rulings of 2026-11-09:
- eval-aggregate/1 counts only the criteria the manifest floors, and no other; a
  floored criterion with fewer than FLOOR_LEAST, ORIGIN_LEAST's ten, cases that expect
  a Finding on it stops the dispatch at fetch, eval_floor_short, before anything is
  asked, so that no published count is one case's verdict;
- a floor counts the cases that expect a Finding on its criterion, its recall, so a
  reviewer blind to it cannot clear it by raising nothing on the clean ones.
The published criteria come from the manifest's floors, which the rubric holds, never
from a case's text. The report is held to the floors exactly, within the kind's
cases and passes. Tests: the blind reviewer, the least at nine and ten through fetch,
and the counts; the record states the least, its reason, and the residuals.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Lane d's Muse read of its own build: a test with one floor cannot tell a pool that holds
only the first. A second floor, PR-4 at four of ten, now fails the pool with PR-3 clear,
and the mutant "the pool holds the first floor alone" is killed by it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…less the floors tells them (#692)

The security and Claude reads of 0a3b046: each case expects one criterion, so the
kind's counts less the floored criteria's are the counts of the findings on no floored
criterion, published on every dispatch with no least. One such case was its verdict.
fetch now stops at one to nine of them (eval_floor_short), and aggregate_verdict says
count_differs for a report leaving one to nine. Tests at 1, 9 and 10; two mutants;
the record says so. The floor fixture cleans up its folders.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…un, and several are said in order (#692)

Argon's read of 0a3b046: a report could say every finding passed and PR-3 four of
ten; aggregate_verdict now holds the floors' failures within the kind's. Muse's read
of c25d477: a floor of every case-run, [1, 1], is held and pinned, and bars_said's
sentence for two floors is pinned to the criteria's order. Three mutants; the record
names them. Argon's other finding needs a short floor at aggregate, which fetch stops
first.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…failures (#692)

The Claude read of 8d747e6: with each criterion's passes within its cases and the
floors' passes and failures within the kind's, their cases are within the kind's
cases too, so that line could not be told from its removal and its mutant survived.
The line, its mutant and the record's mention go.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… floors and marks alike (#692)

The security read of 8d747e6: floors on PR-3 at one commit and on PR-4 at the next,
the cases unchanged, leave the findings on neither published by difference, and a
natural mark moved from ten cases to eleven names the eleventh, as before this change.
The Factory Builder ruled it a residual: a manifest edit is reviewed and only the
Maintainer dispatches a gate set, while a composition rule on unfloored criteria could
stop a set none of us can see. Issue 695 takes it up after gate-v1; issue 696 protects
the manifests in the hooks.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…stops the fetch (#692)

Argon's read of 8d747e6: a floor of [0, 10] was held, a floor that asks nothing; a
floor is now n of d with 1 <= n <= d. Muse's read: floor_short was tested with one
floor alone, so any() could become all(); a test of PR-4 of one beside PR-3 of ten
holds it. The floor's fraction is held by three mutants on the new line, and the
record says so.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@don-petry
don-petry requested a review from a team as a code owner November 10, 2026 00:01
@don-petry don-petry added the factory:hands-off A person's mark: the Factory writes nothing on this pull request (kernel/terms.py LABEL_HANDS_OFF) label Nov 10, 2026
@chatgpt-codex-connector

Copy link
Copy Markdown
Codex Review Summary: ❌ Failed

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ❌ Failed 2026-11-10T00:01:04.965192Z
We couldn’t complete this review. Comment @codex review to try again.
714be8f PR opened
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@coderabbitai

coderabbitai Bot commented Nov 10, 2026 •

Copy link
Copy Markdown

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

Run configuration

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • Trigger review
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

don-petry and others added 4 commits November 9, 2026 18:18
…no criteria either (#692)

CI's mutants shard 4 at 714be8f: "a gate set's report says the split held where no
kind was read" survived its named test, as the case still carried the good report's
criteria and the criteria's own check refused it first. The two cases of no kinds
that hold the split and the blocking words null the criteria too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…er (#692)

"a gate set's kinds may be unread while its words are said" survived its named test on
the merge of main: that case kept the good report's criteria, which their own check
refused first. It nulls them now, as the split and blocking cases do.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…d overlaps a residual (#692)

Astra's read of c5ccd1b: the pool's floors were tested at one fraction alone, and a
report's per-criterion counts on its first floor alone; a test of PR-4 at nine of ten
beside PR-3 at a half, and one of PR-4's interval wrong, hold them, with two mutants.
Its third finding, a floored criterion and the natural cases overlapping in a part of
one, is a residual beside the differences between manifests, as the kinds and the
natural cases overlapped before; issue 695 takes it up with them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
don-petry and others added 2 commits November 9, 2026 20:18
NNNN is 0180 in the record's name and title. The overlap residual names its issue by
the one row beside it rather than by a bare number.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@sonarqubecloud

Copy link
Copy Markdown

@don-petry
don-petry added this pull request to the merge queue Nov 10, 2026
Merged via the queue into main with commit fe90c64 Nov 10, 2026
26 checks passed
@don-petry
don-petry deleted the gate-criterion-floors-692 branch November 10, 2026 03:00
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

factory:hands-off A person's mark: the Factory writes nothing on this pull request (kernel/terms.py LABEL_HANDS_OFF)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

A gate set holds a floor per criterion, read from each case's expected criterion

1 participant