Skip to content

docs(now): eight decisions that are not mine (#2474) - #2487

Merged
gHashTag merged 4 commits into
masterfrom
docs/decision-brief
Aug 23, 2026
Merged

gHashTag merged 4 commits into
masterfrom
docs/decision-brief

Conversation

@gHashTag

Copy link
Copy Markdown
Collaborator

Every question this campaign raised that is an owner's to answer, collected with options and the cost of doing nothing:

https://claude.ai/code/artifact/61cc6ccb-4173-4ddf-9077-e86e000adb29

Repository settings (#2455, #2474), reproducibility records (#2477), language and corpus (#2479), and four small measured defects (#2476, #2483, #2466, #2472). Two of those need the stage0 freeze lifted, which is why they are filed rather than fixed.

Regression checked first

After 70 merges in 24 hours, everything built in this campaign holds on fresh master:

gates killing every mutant 12 of 13
surviving mutants 2 — exactly the pair declared UNCOVERED
verify_exhaustive (was permanently red) exit 0, control green

The UNCOVERED declaration still matched after other people's edits moved those lines from :359/:403 to :451/:495, because it names branches by message rather than by number.

No duplication this time

Before proposing a "regression command" I checked t27c gates and t27c battery. gates runs every check_*.py with each child's own exit code and separates scripts that cannot fail; battery is for document reconstruction oracles. Neither runs the negative controls, which is what tri gates sweep and mutate add.

First clean §21 check in four iterations.

The connection worth reading twice

coverage cannot be moved by bookkeeping or by one compiler fix. Re-sealing the 18 stale twins moved the count by zero — 14 do not parse — and t27c backlog reports zero specs at depth 1 with 94% of the backlog four or more classes deep. Multi-month work, or a decision to call it accepted debt. There is no third option and the brief does not invent one.

Refs #2455, #2466, #2472, #2474, #2476, #2477, #2479, #2483

The gate audit produced eight questions I deliberately did not answer:
repository security settings, deleting reproducibility records, and
corpus-wide language decisions. Collected with options and a stated cost
of doing nothing.

Regression checked first: after 70 merges in 24 hours everything built in
this campaign holds on fresh master. 12 of 13 gates kill every mutant,
the two survivors are exactly the pair declared UNCOVERED, and the note
still matched after other people's edits moved those lines because it
names branches by message rather than by number.

Also checked, before proposing a new command, that t27c gates and
t27c battery do not already do it. They do not -- neither runs the
negative controls. First clean section-21 check in four iterations.

Refs #2455, #2466, #2472, #2474, #2476, #2477, #2479, #2483
…next thing (#2474)

70 merges landed in the 24 hours after this campaign's work and none was
reviewed against its numbers. Re-running the measurements is the first
item of the next iteration, and the only way 'the new cycle does not
break the old work' is a fact rather than a hope.

Everything held; the fragile piece held because it names its uncovered
branches by MESSAGE and other people's edits moved those lines.

Plus what makes a decision brief usable: every number from a command
named so it can be re-run, and a statement of how often its author has
been wrong -- seven times in this campaign, each correction appended to
the issue that carried the claim.

Refs #2474
@github-actions

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

This notebook contains session context, decisions, and artifacts for this work.

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-08-23 01:56:04 UTC

Summary

Status Count
Total Open PRs 5
PRs with Failing Checks 1
PRs with All Checks Green 4
READY 3
FAILING 1
PENDING 0

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e905ba8eeb53 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

This was referenced Aug 23, 2026
This was referenced Aug 23, 2026
The skill file's numbering has read 23, 25, 26 on master for days. I
checked it twice with `git log -S '## 24. ' -- <file>` and both times
read the empty result as "the number was never used".

It was used. The commit is on THIS branch, unmerged since 2026-08-23.
`git log -S` searches the history of the current branch, so a section
that exists only on an unmerged branch is indistinguishable from one
that never existed -- unless the search says `--all`.

Resolved by taking master's file and inserting section 24 where it
belongs. Numbering is now contiguous with no gaps for the first time
in this campaign.

Refs #2474
@github-actions

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

This notebook contains session context, decisions, and artifacts for this work.

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-08-23 15:44:43 UTC

Summary

Status Count
Total Open PRs 6
PRs with Failing Checks 2
PRs with All Checks Green 4
READY 3
FAILING 2
PENDING 0

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e905ba8eeb53 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-08-23 15:47:43 UTC

Summary

Status Count
Total Open PRs 6
PRs with Failing Checks 2
PRs with All Checks Green 4
READY 3
FAILING 2
PENDING 0

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e905ba8eeb53 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@github-actions

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

This notebook contains session context, decisions, and artifacts for this work.

@gHashTag
gHashTag merged commit 809d3d4 into master Aug 23, 2026
20 of 21 checks passed
@gHashTag
gHashTag deleted the docs/decision-brief branch August 23, 2026 15:50
gHashTag added a commit that referenced this pull request Aug 23, 2026
Section 74 told you to resolve a gap in the numbering with

    git log --oneline -S'## 63. ' -- path/to/file.md

and to read an empty result as "the number was never used". It cited
the gap at section 24 as an example of exactly that.

Section 24 existed. It was in PR #2487, authored days earlier and
unmerged, and it merged today into the gap it was written for. The
numbering has no gaps for the first time in this campaign.

`git log -S` searches the history of the CURRENT BRANCH. A section that
lives only on an unmerged branch produces the same empty output as one
that never existed, and I read that output twice as proof of the second
without noticing it could not distinguish the two.

The command now says --all, the third outcome is listed, and the
correction is stated where the wrong rule was, rather than appended as
a new section that leaves the wrong one readable.

Refs #2566
gHashTag added a commit that referenced this pull request Aug 23, 2026
#2568)

* gate: no emitted Verilog signal is four billion bits wide

`range_decl(width: u32)` formats `[width - 1 : 0]` and guards only
width 1, so a packed width of 0 underflows. Which bug that is depends
on a build flag, and CI picks the quiet one:

  debug   (overflow-checks on)   panic, "attempt to subtract with overflow"
  release (overflow-checks off)  `function [4294967295:0] cover_point;`
                                 emitted, exit 0, stderr empty

CI builds --release, so Corpus ratchet has been green over Verilog no
synthesiser accepts, and the loud version of the bug is the one CI
structurally cannot see.

Ten specs, three mechanisms: u32 underflow (8 specs), the same
underflow on a u64 path (notebooklm), and a Map flattened to 4198431
bits (stdlib, apparently not an underflow). Searching for the literal
4294967295 finds eight of the ten; keying on the property -- no real
bus is a million bits wide -- finds all three classes.

The checker treats a debug panic as the same finding as a release
emission. A checker that scanned only emitted text would report CLEAN
against a debug compiler, because a compiler that panicked emitted
nothing to scan.

The compiler-side fix is not made here. compiler.rs is sealed by
stage0/FROZEN_HASH and FROZEN.md reserves re-sealing for a deliberate
maintainer ceremony (M5). Filed as #2566 with the reproduction and the
design question it raises. The ledger refuses the eleventh spec; it
does not bless the ten, and says so in its own header.

Registered in check_pr_branch_filters.py as merge-critical, so the
meta-gate covers it -- verified by adding a branches filter and
watching that gate go red.

Refs #2566

* skill: `git log -S` without --all answered the wrong question

Section 74 told you to resolve a gap in the numbering with

    git log --oneline -S'## 63. ' -- path/to/file.md

and to read an empty result as "the number was never used". It cited
the gap at section 24 as an example of exactly that.

Section 24 existed. It was in PR #2487, authored days earlier and
unmerged, and it merged today into the gap it was written for. The
numbering has no gaps for the first time in this campaign.

`git log -S` searches the history of the CURRENT BRANCH. A section that
lives only on an unmerged branch produces the same empty output as one
that never existed, and I read that output twice as proof of the second
without noticing it could not distinguish the two.

The command now says --all, the third outcome is listed, and the
correction is stated where the wrong rule was, rather than appended as
a new section that leaves the wrong one readable.

Refs #2566
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant