ums: a handed finding's particulars are the half that is wrong - #1014
Conversation
Three findings from the six-PR session that merged #996, #999, #1003, #1004, #1005 and #1008, plus one candidate declined as already covered. metacognitive-monitoring: extends 'A premise you were handed is still a claim' to a premise handed by a reviewer rather than by a person. That section's only detector is a hedge in the source, and a review comment carries none, so the signal is silent exactly where the premise most often arrives. Records the asymmetry that says which half to check: across five findings the conclusion held five times and the particulars were wrong five times, and particulars are what decide the edit list. Adds the two reasons the check feels done when it is not, comprehension and delegation. fully-clean: a third state for the same external reviewer, alongside the fifth case's refusal. Its check run completes success having posted no review at all, so nothing on the PR reports that a configured reviewer never weighed in. Defeats the 'no verdict is its own state' remedy, which reads the job outcome and is scoped to a job that failed. avoid-hardcoding-external-data: the boundary the parameterize rule must not cross. A quoted command that was run, a verbatim error string, and the stated conditions of a measurement are claims about the past, so substituting a parameter falsifies them rather than generalizing them. Declined: that merge-base --is-ancestor reports non-ancestor for every branch in a squash-merge repo. Measured true here, 8 of 8, all merged, but CLAUDE.md already states it flatly at two sites.
|
Self-review while Finding (nit, not blocking)
The enumeration that follows lists five items, so the second number is internally consistent. What is missing is any statement that the two sets differ. A reader hits Either reading is defensible, which is what makes it worth a clause rather than a correction: if five were checked and only four had been prematurely relayed, say so. Fixing it here rather than guessing, since the entry is specifically about not restating particulars you have not verified. Holding the fix until Particulars confirmedThe two case records make four checkable claims, and all four hold:
The Note on the run
Generated by Claude Code |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
Code reviewThis PR adds three new sections to existing Verified accurate:
One finding (posted inline on A few other candidate issues (a "fifth case" cross-reference in VerdictNeeds minor changes — one non-blocking wording nit (self-contradicting sentence, one-line fix); otherwise ready for merge. |
|
Claude finished review — View run Review SummaryReviewed PR #1014 ("ums: a handed finding's particulars are the half that is wrong"), a documentation-only PR adding three new sections to existing Extensive fact-checking passed: every cited PR number, commit SHA, timestamp, and quoted literal in the new content checked out exactly against the actual repo/GitHub state — no hallucinations found. 1 issue found and posted:
3 other candidates investigated and rejected after validation (a cross-reference to "fifth case," a four-vs-five count in a case record, and a missing head-commit filter in the Posted the inline comment and a summary comment with verdict to the PR. VerdictNeeds minor changes — one non-blocking wording nit (self-contradicting sentence, one-line fix); otherwise ready for merge. |
The third-state case record closed by saying nothing anywhere on the PR distinguished a reviewer that had approved from one that never spoke. The sentence directly above it reports the opposite: the login-filtered review list returned eight reviews with none from Copilot, which is exactly the discriminator, and exactly what the section's own Do-bullet sends a reader to fetch. The body prose was already right -- nothing on the PR *reports* the gap on its own -- but the case record escalated that to *distinguished*, which contradicts the check the section exists to prescribe. Naming the query keeps the intended contrast with the green-check signal and makes the point land harder: every signal except the prescribed one was uninformative.
Code reviewRe-reviewed at head This round's fact-checking pass (verifying every specific PR/commit/timestamp/query claim against live repo state, per this corpus's own
All four were independently verified against live GitHub API state and file contents before posting (not just plausible-sounding on a re-read). Findings 1 and 2 are the more substantive ones, since they affect a load-bearing factual claim and a usability point in a section about verifying particulars; 3 and 4 are prose-precision nits in case records, not blocking on their own. VerdictNeeds minor changes — one factual claim in a new case record isn't supported by the cited PR's actual check-run data, plus a related surface/field ambiguity and two smaller case-record precision issues. No CLAUDE.md/lab-manual violations, no code, no security concerns — this is a documentation-only PR and the underlying rules/Do-Don't pairs in all three sections are sound; the issues are in supporting evidence and internal consistency. |
|
Claude finished review — View run I've completed the review of PR #1014. Summary: The prior round's finding was confirmed fixed (the self-contradicting sentence in
Posted a summary comment with verdict Needs minor changes — no CLAUDE.md/lab-manual violations, no code/security concerns, but one load-bearing factual claim needs correction plus three smaller precision issues. VerdictNeeds minor changes |
|
Working on this --- paws off until I'm done. |
…and two self-contradictions Four findings, all correct, all verified against live API state rather than reasoned about. 1. fully-clean.md's case record claimed Copilot's check run on #1008 completed success at 04:50:41Z. There is no such check run. Neither #1005 nor #1008 carries ANY Copilot-attributable check context (8 and 10 checks respectively; filtering either for /opilot/ returns 0), and that holds both where Copilot posted a refusal and where it was silent. Rewrote the mechanism: the silence is invisible on the check surface by construction, not misread from a green Copilot check -- which is a stronger argument for the section's own advice. Left the error visible in the record rather than deleting it. 2. The prescribed lookup named user.login while tool-mappings.md's CLI fallback exposes author.login, and the two surfaces also disagree on whether the value carries [bot]. Measured on #1005: REST returns copilot-pull-request-reviewer[bot] under user.login, gh pr view returns copilot-pull-request-reviewer under author.login. Mixing them returns zero hits and reads as 'did not review' -- the exact false negative the section is about. Now a two-row table. 3. avoid-hardcoding-external-data.md said the judgment was 'never written down' four lines below quoting one of the two files writing it down. Both files state it, in fact. Replaced with the point that survives: an in-file rationale protects a reader, not a grep, so a sweep re-flags them regardless. 4. metacognitive-monitoring.md said 'four findings' then 'the five findings'. Reconciled, and corrected its closing claim: that entry is where 04:50:41Z came from, and it called its own conclusion 'exactly right'. It was not. A verification that adjusts a figure without asking whether the measured thing exists propagated the fabrication into fully-clean.md instead of catching it, so the record now reads five of six rather than five of five. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Code reviewRe-reviewed at head
No new issues found in this round's fact-check of the diff (banned punctuation, cross-reference resolution, and the newly-added prose all check out against live state). VerdictReady for merge. |
|
Claude finished review — View run Posted the review comment: #1014 (comment) SummaryRe-reviewed PR #1014 at head
No new issues surfaced in a fresh fact-check pass of the diff. VerdictReady for merge. |
2 participants
UMS pass over the six-PR session that merged #996, #999, #1003, #1004, #1005
and #1008. Three entries recorded, one candidate declined. All three extend
fragments that already exist and are already wired into
CLAUDE.md, so thereis no new fragment and no
CLAUDE.mdchange.1.
shared/workflow/metacognitive-monitoring.md(+71)Extends "A premise you were handed is still a claim" rather than adding a
section, because that section already covers the general shape. Three
increments it does not cover:
detector is a hedge in the source -- "I think", "probably", a trailing
question mark. A review comment carries no hedge: it states its finding
flatly and cites a file and a line. So the one signal offered is silent
exactly where the handed premise most often arrives.
rather than a scolding, because it says which half to spend the check on.
Across the five findings checked afterward, the conclusion held in five
of five and the particulars were wrong in five of five: a guard whose
failure was broader than reported, a script that was not the one running, a
cited line number with nothing at it, a hardcoded-value scope naming 2 of 6
sites, and a two-command failure whose first command failed with an error
the finding never mentioned. Particulars decide the edit list, so relaying
them unverified propagates a wrong list of lines to change underneath a
conclusion that is right.
from the inside, because the effort goes into understanding rather than
testing. And delegating the check feels like having made it -- a dispatched
task reads as settled, when it is settled only once its answer comes back.
Cross-referenced to
address-every-comment.md, which already carries theper-component versions (verify a suggestion's literals, read a cited source,
test a negative result's scope). Those govern what you do to the PR; this
governs what you assert, which is a separate surface with no reviewer on
it. Stating the boundary is what keeps this from reading as a duplicate.
2.
shared/workflow/fully-clean.md(+28)A third state for the same external reviewer, appended to the fifth case
rather than numbered as a ninth. Numbering it would have falsified the very
next line, "A sixth case runs the other way from all five above" -- and it is
genuinely a third state of that one reviewer, not a new case of the file.
The refusal in the fifth case leaves a record. This one leaves none: the
check run completes
successhaving posted no review at all, so a readerscanning checks sees a reviewer that ran and was content.
It also defeats a remedy already in this file. The "no verdict is its own
state" bullet in criterion 2's four-surfaces list covers a job that posts
nothing, and prescribes reading the job's own outcome. That works there
because the job failed. Here it succeeded, so the outcome reads
successand points away from the gap.
3.
shared/coding/avoid-hardcoding-external-data.md(+63)The boundary the parameterize rule must not cross, which appears to be novel
-- searches below found nothing. A quoted command that was executed, a
verbatim error string, and the stated conditions of a measurement are claims
about the past. Substituting a parameter does not generalize them; it
falsifies them, and leaves no trace that anything changed.
The tell offered is tense and mood: prescriptive text names the parameter,
evidentiary text keeps the literal, and one file routinely carries both.
Distinguished from
ascii-punctuation-in-source.md's whole-file-replacewarning, where the failure is scope and the diff stays true; here the
rewritten text becomes false, which no diff size reveals.
The case record cites only already-merged content:
memories/preferences.mdkeeps
git worktree add /tmp/wt-ums mainplus the scores measured with it andsays those runs "used a repo whose default branch is literally
main, whichis why they are written that way here";
skills/gip/SKILL.mdkeepsfatal: invalid reference: origin/main. #1008 made that judgment correctly inboth files and never wrote it down, so the next sweep had nothing to consult.
Declined
That
git merge-base --is-ancestorreports non-ancestor for every branch ina squash-merge repo. Measured and true here -- 8 of 8 local branches from
the session report non-ancestor against
origin/main, and all 8 of their PRsmerged. Declined because
CLAUDE.mdalready states it flatly at two sites,and adding a third would be the redundancy
challenge-redundant-content.mdasks reviewers to flag:
commits are no longer ancestors of
main's new tip --git merge-base --is-ancestor <old-commit> origin/mainreturns false."is not itself the signal", plus the content check
(
git show origin/main:<path> | grep) that is precisely what stops a readerconcluding their work was lost.
The brief asked to extend only if the existing wording was not plain enough.
It is.
Searches run
Normalized on both sides (
re.sub(r"[\*_\s]+", " ", s).lower()`), since thiscorpus breaks lines mid-phrase and a raw grep returns false negatives across a
line break or a code span:
memories/preferences.md, unrelated sense)address-every-comment.md, cross-referenced above)A null result is a fact about the pattern, not about the corpus
(
grep-is-not-coverage.md), so entry 1 was placed by reading the nearestsection rather than on the strength of these zeros.
Verification
wrong -- which is entry 1 demonstrating itself mid-pass. The completion
time supplied was
04:08:13Z;get_check_runson ums: three findings from a five-PR session #1008 reportsstarted_at 04:47:10Z,completed_at 04:50:41Z. The conclusion -- greencheck, no review -- was exactly right. The entry carries the measured time.
get_reviewson ums: three findings from a five-PR session #1008: 8 reviews, 4 from the repo's own review bot and 4from the maintainer, none from Copilot; page 2 confirmed empty.
get_reviewson ums: four signals that answered a different question than the one asked #1005: the Copilot quota refusal as aCOMMENTEDreview at2026-07-31T23:59:46Z. Both halves of the contrast measured, not recalled.git diff -U0 origin/main...HEAD(three-dot range,after committing,
LC_ALL=C.UTF-8): 162 added lines, 0 hits, 0 non-ASCII.check-new-line-breaks.py: flagged one mid-line semicolon on first run;fixed and re-run clean. Its exit code is advisory, so the output was read
rather than the status.
memories/untouched (git diff --statis threeshared/files);check-memory-file-size.pyreports no file over 1200.ascii-punctuation-in-source.mdexists and its "104-line diff" figure is atL233; the "no verdict" bullet is at L256; the sixth-case anchor still reads
"all five above" and is unaffected by the insertion.
git ls-remotematchinggit rev-parse HEADat394e387.Disjoint from #1013, which touches only
skills/gip/SKILL.md. No merge-orderconstraint; either can land first. Entry 3's case record deliberately cites
#1008's merged state rather than #1013's change, so it stands alone.
Generated by Claude Code