Skip to content

Independent recount of the #7630 word, paid and target invoices (no κ) - #7635

Open
fluorine wants to merge 2 commits into
CrocSwap:mainfrom
fluorine:audit/pr7630-independent-recount
Open

fluorine wants to merge 2 commits into
CrocSwap:mainfrom
fluorine:audit/pr7630-independent-recount

Conversation

@fluorine

Copy link
Copy Markdown
Contributor

Change and mathematical scope

Independent finite audit. No new κ, no frontier change, no supplier admission.

This PR adds one isolated package, research/pr7630-independent-recount. It audits the open candidate #7630 (conditional κ = 0.0036667343458, DreamingOfClouds) at exact head 6026e87d1cac03eb5bd2aa8524a8f5891dd9934e. It applies the method of #7628 to a word that also changes the productive target histogram.

The generic replay of #7630 builds the candidate invoice as the #7623 invoice plus a declared per-call delta, for the paid side and for the helper targets. This package binds both sides of that delta to the word bytes with separate code that runs in about two seconds. verify.py imports no constructor, application or pricing engine from #7630.

Result, for the four pinned files (bytes equal to the PR head blobs; the #7623 word equals the word published in #7623):

  1. Header. The words are equal except initial. Exactly two fields change: initial[457] 3 to 281 and initial[461] 3 to 285.
  2. Chart chronology. Both complete words (77,277 and 208,738 events) are legal: each MOVE starts at the current chart of its field, each ADD/COPY/SCALE acts at the chart of its operands, INIT/CLEAR ownership is consistent, and the public fields end at the final charts.
  3. Paid recount. With rank(a->b) = dim(A+B) - dim(A∩B) over GF(2), each MOVE joins two nested frames. Exactly: paid = 17 · H(word) + K, with the same K = {1:384, 8:656, 9:32, 16:416, 24:416, 32:416, 40:416, 48:416, 56:416, 64:208} for P104: loan + stationary-snapshot engine (14 instances on #7621), conditional kappa 0.0036624027915 #7623 and P104: seven stop edits + joint hub re-settle on #7623, conditional kappa 0.0036667343458 #7630. This K also equals the one of the Conditional κ = 0.003652740309 (+>3.16496849e-7 over #7611) #7612 recount. The recounted native delta per call is {1:-2, 2:-3, 3:-5, 5:+1, 6:+2} (6,257 to 6,250 MOVEs). Paid positions are 110,145 and 110,026.
  4. Target recount (new compared with Independent recount of the #7612 word, paid invoice and moment signs (no κ) #7628). For each helper field, the target rank is rank(final frame) - rank(initial frame). Exactly: targets = 17 · T(word) + K_T, with the same K_T for both words. The recounted helper delta per call is {6:+2, 9:-2}. Target positions stay 22,461 and the rank surplus stays 1,424.
  5. Moment signs. A separate 256-bit integer interval engine agrees with the PR at seven points: P104: seven stop edits + joint hub re-settle on #7623, conditional kappa 0.0036667343458 #7630 is positive at the claim, at the Conditional κ ≥ 0.003665731276 (+>3.607758e-7 over #7626, common U) #7627 value and at the P104: loan + stationary-snapshot engine (14 instances on #7621), conditional kappa 0.0036624027915 #7623 value, and negative at 0.0036667343459. P104: loan + stationary-snapshot engine (14 instances on #7621), conditional kappa 0.0036624027915 #7623 is positive at its own value and negative just above it and at the P104: seven stop edits + joint hub re-settle on #7623, conditional kappa 0.0036667343458 #7630 claim. An upper bound of dM/dκ on [0, 0.00376669] is negative for both invoices, so each crossing is unique there.
  6. Tent test. The change in the moment is zero or positive for every tariff min(r, a). Every increasing concave tariff with f(0) = 0 is a non-negative combination of these on integer ranks, so P104: seven stop edits + joint hub re-settle on #7623, conditional kappa 0.0036667343458 #7630 is no worse for any such tariff. This is an independent check of the "uniformly better" statement.
  7. Controls. Ten semantic mutations reject.

Remaining obligations (not established here):

Task and overlap

Base: canonical main e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8, fetched at session start and again before submission on 2026-11-17 UTC (unchanged). Task card independent-moment, scope refined: the card's first experiment is applied to the newer exact head #7630. Dependency head: #7630 6026e87d1cac03eb5bd2aa8524a8f5891dd9934e (unchanged at submission).

Overlap: searches of the open PR titles and bodies by number, and a diff-level read of #7630, found no audit of #7630 or of #7623. #7631, #7633 and #7632 mention #7630 but do not audit it. #7276 (CI navigation repair) is separate and not touched. scripts/find_overlap.py was not run. Discussion bodies were not searched.

Validation

python3 scripts/research_package.py --package research/pr7630-independent-recount --check-only
python3 scripts/research_package.py --package research/pr7630-independent-recount
python3 -I -B research/pr7630-independent-recount/verify.py

Evidence at head 9d6d3d8ba3060ed2788cbb777d0b7c1d35f5bd12, CPython 3.13.9, Windows 11:

  • --check-only: PASS_PACKAGE_METADATA_AND_PINS.
  • Registered argv run directly (-I -B -W error), normal and -O: PASS_INDEPENDENT_RECOUNT_PR7630_DECLARED_SCOPE, about 2 s.

Not run: the Python 3.12 runner replay (this machine has no Python 3.12, and the runner requires it), the repository test suites, hosted CI, and every #7630 application stage. The hosted plan job may show the inherited navigation failure that #7276 repairs. All checks are finite checks on pinned bytes. None is a proof of the multiplication theorem.

Credit

The audited construction, word, invoices and claim are by DreamingOfClouds (#7630, #7623), on the P104 lineage of SYLforge (#7620), Clear1nsight, Ali Romman, Hao Yan, ToikY, Rohan Arun and their predecessors. The vendored files keep their Apache-2.0 license. Task card and package tooling: Douglas Colkitt.

Prepared by fluorine with substantial Anthropic Claude (Claude Code) assistance. No independent human review of this audit exists yet.

Open next questions

See the package README and TASK.json. Smallest useful follow-up: recount the ordinary bill of the #7630 word with the formula that #7630 names. Blocker: that formula lives in a repository script that this audit did not read.

🤖 Generated with Claude Code

fluorine and others added 2 commits November 17, 2026 12:21
Add the CrocSwap#7623 and CrocSwap#7630 words and invoices, unchanged, from the exact
CrocSwap#7630 head 6026e87. The CrocSwap#7623 word bytes
equal the published CrocSwap#7623 word.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…es (no κ)

Recount the CrocSwap#7630 word at head 6026e87
without any constructor or pricing code from the PR. The verifier binds the
declared paid and helper-target deltas to the word bytes with a GF(2) rank
law, follows every chart through both words, and checks the moment signs with
a separate 256-bit interval engine. Ten controls reject.

No new κ, no frontier change and no supplier admission.

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