Skip to content

Conditional κ ≈ 0.001717794984118634663827: two hostee moves and seven retired-lender aliases on #2751's joint twelve-source words - #2802

Merged
Mister-Meeseeks merged 3 commits into
CrocSwap:mainfrom
mdumitrean:research/t74-joint-twelve-retired7
Oct 16, 2026
Merged

Mister-Meeseeks merged 3 commits into
CrocSwap:mainfrom
mdumitrean:research/t74-joint-twelve-retired7

Conversation

@mdumitrean

@mdumitrean mdumitrean commented Oct 15, 2026 •

Copy link
Copy Markdown
Contributor

This extends the words of #2751 (@tekuuu, head 4cfbb0cc) with seven retired-lender aliases. Five need only exact refits. The other
two are opened by moving two existing borrows onto free retired lenders.

The new conditional κ is 0.001717794984118634663826999…. The exact rational is 0.001717794984118634663827 - 10^-46 - 10^-70;
the selected frontier is unchanged.

over margin
#2751 (tekuuu) +6.53397770208325718e-7 (exact 326698885104162859/(5·10^23))
#2766 (DavidAriasC) +5.38010651632355557e-7 (exact 538010651632355557/10^24)
#2767 (jeremykothe2) +5.38010651632355556055e-7

What is new, and what is credited.

Construction.

Result.

  • 408 helpers {4:26, 5:2, 7:62, 8:78, 9:240}, 72 buffers, 666 calls, 271728 roles; physical 66794.
  • Local profile {1:2382,2:648,3:268,4:128,5:26,8:8}.
  • The owner fill is Conditional kappa ≈ 0.00171714158635 via joint twelve-source recovery #2751's fill with the seven helpers' 4662 roles removed (renumbered, every bin re-tiled as a prefix). Its
    pattern relaxation has the same optimum.
  • Complete profile {1:1603728,2:431568,3:185082,4:85248,5:27194,8:5328,32:1,48:26558}; rank mass 4816512.

Independent checks.

Packaging.

Checker adaptations (diffs in RETIRED5-CODE.patch and RETIRED7-CODE.patch, reasons in the README).

Validation.

  • Full cold release from the committed package through full_replay.py: PASS (PASS_T74_JOINT_TWELVE_COLD_COMPLETE_RELEASE,
    411 s, run cold from the committed package). It covered:
    • all 491 pinned PR2679 files authenticated;
    • the 36-file overlay;
    • four fresh native providers (five stages each);
    • the retained-parent admission;
    • owners, lift, rebind, and discovery/unpin/freeze at Conditional kappa ≈ 0.00171714158635 via joint twelve-source recovery #2751's 24-digit precision;
    • the outer ledger and independent derivation;
    • the original verify_release.py: PASS_RELEASE, including rebuild-borrow-words with all stage checks and controls, the strict
      T74 replay with its admission, bit and independent controls, and the geometry controls.
  • Registered hook, about 100 s, budget 1020 s: python3.12 -B scripts/research_package.py --package research/t74-joint-twelve-retired7.
    • verify.py literal review;
    • verify_retired7.py: seven-stage rebuild, the vendored byte-identical PR2679 auditor on eight states, 142 tamper controls, the
      literal owner fill, the 80-digit root and κ margins;
    • packing_cap.py: cap ≈ 0.00175808593463, realized 0.9771;
    • test_integrity.py.
  • python3.12 -B -m unittest discover -s tests -p 'test_research_*.py': 69 tests OK (3 skipped).
  • Evidence is in evidence/retired5/ and evidence/retired7/, sanitised of local paths.
python3.12 -B scripts/research_package.py --package research/t74-joint-twelve-retired7
python3 -B research/t74-joint-twelve-retired7/full_replay.py --work /tmp/FRESH-retired7   # fresh external dir; requirements.txt

Scope. This is conditional finite construction. All all-size, interface, hosting, precision, recurrence, bit-supplier and outer
obligations of #2751 and its lineage are inherited unchanged. No optimality is claimed beyond the exact census of retired-lender
aliases with movable hostings.

Credits.

and the lineage these retain (SOURCE-NOTICE).

Prepared by Marius-Paul Dumitrean with substantial AI assistance (Anthropic Claude, via Claude Code). Apache-2.0.

tekuuu and others added 3 commits October 15, 2026 23:04
…s words

Conditional kappa 0.001717794984118634663827 - 1e-46 - 1e-70
(+6.534e-7 over CrocSwap#2751, +5.380e-7 over CrocSwap#2766 and CrocSwap#2767).

New package research/t74-joint-twelve-retired7; CrocSwap#2751's directory is
unchanged. It contains the five retired-lender aliases of the local
retired5 construction (606<-550, 635<-551, 637<-552, 638<-553,
639<-562) and adds two hostee moves: the CrocSwap#2226 borrows of B-warps 30/31
move from lenders 401/403 (own lives ended in A) onto the free retired
lenders 564/565 (last gates refit to their exact floor, dim 6 -> 5),
which frees 401/403 for two more CrocSwap#2679-type aliases 667<-401 and
669<-403 (+3.857e-7 over retired5). Stage inputs refit7/hostmove2/
alias7/refit7c with exact structure checks and 25 tamper controls per
word (retired7_controls.py), certified optimal frames, the compacted
owner fill (271728 roles, R 408), checker adaptations
(RETIRED7-CODE.patch), a registered bounded hook and evidence.
Cold full replay of 043467a from the committed package (fresh external
work, PR2679 source authenticated, four fresh native providers, retained
parent, complete admission, verify_release PASS_RELEASE with the retired7
stages and 25 controls per word inside rebuild-borrow-words):
PASS_T74_JOINT_TWELVE_COLD_COMPLETE_RELEASE in 411 s. Evidence sanitised
of local paths; README states the exact expected result.
@philippeVV

Copy link
Copy Markdown
Contributor

On your open question about followers 670/671 (no κ): under every refit of the fixed gate list, they have no lender. I ran an augmenting-path census that treats all 72 current hostees plus 670/671 as movable, using exact lattice frame bounds. The ceiling of every life's first gate is ≤ 5. Exactly 72 registers have floor ≤ 5, and every one of them already hosts. All free registers have floor ≥ 6. The 74 lives overlap pairwise in time, so no register can take two in sequence. That makes it a pigeonhole: 74 lives on 72 registers, with 0 augmenting paths. Opening a slot would need a gate-level edit. Your fill is also at the class-level ceiling to about 1e-14, and pre-life/refit censuses find no new legal hostings or aliases on these words. Checker: https://github.com/philippeVV/integer-mult-bounds/tree/0610f3de/research/t74-lender-pigeonhole (Claude Code assisted).

@mdumitrean

Copy link
Copy Markdown
Contributor Author

Thank you both.

@philippeVV, followers 670/671. Thank you, that settles the question. It agrees with our census: the only free floor-5 sinks were 564/565, and the two moves use them. Your pigeonhole extends that to every refit with all 72 hostees movable. So a further width-5 alias needs a gate-level edit, which is open question 1. The PR body now records the answer and credits your checker (0610f3de).

@thakrarsagar, #2847. Thank you for the independent check. The provenance correction is right. On both words, #2766's own words differ from #2751's at one frame only, B1151 (rank 6 → 5). The other 19 frame changes we credited to #2766 (A0–A11, B771/773/776/778/783/788, restore60) are our refit5c choices.

  • The PR body is corrected.
  • I have not pushed to the branch, because your check pins 17b2ee6e and several downstream claims vendor its words. The package README and CLAIM.json keep the old wording until the next package revision.

On your open question 2: the final frames are the #1186 toolkit's certified optimum for the borrow-word model, with gates, warps, start/final frames and widths fixed (inputs/refit7c_43*.json; only B1170/B1171 move, dim 6 → 5). That certificate does not cover frame changes that alter start or final frames, and with them the widths and the fill.

Your questions 1 and 3 are open on our side too. Our cold release ran fresh native providers on the shipped words only, with no occupancy mutants.

Mister-Meeseeks pushed a commit that referenced this pull request Oct 16, 2026
#2818

Fresh standard-library checks of the seven declared stages on both T74 words:
event diff and own stage-by-stage rebuild, per-register occupancy timeline,
forward/dual free-algebra replay plus a matrix realisation, GF(2) warp frames
and refit exactness, profile and inventory recount by two subspace
representations, the 271728-role fill with R derived from the words, the
complete profile, mass and root bracket, and the PR2766/retired5/retired7
gain attribution. 26/26 rules pass and 15/15 negative controls are rejected
by their named rules. #2818's words, fill, order and providers are
byte-identical. No kappa claim; inherited obligations remain.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mister-Meeseeks pushed a commit that referenced this pull request Oct 16, 2026
… PR2766 provenance, root-bracket precision

- Occupancy: add controls C4c-C4f (extra-word lender 403, lender writes moved
  to the end of restore at frame FULL, a moved lender read) with machine-checked
  must-accept / exact-failing expectations. A lender write moved past its alias
  warp at FULL is accepted by both replays, the frame chain, recount, refits,
  fill and price; only OCC-TIMELINE (plus undeclared-edit checks) rejects it.
  Occupancy is therefore a conformance property, not a kappa-soundness
  condition, in the finite model; #2802 enforces it structurally in
  verify_retired7.py, #2818's patch check has no occupancy predicate (read,
  not run). Corrects the earlier "only frame monotonicity" wording.
- Attribution: compare against PR2766's own pinned word (only B1151 differs);
  the other 19 frames #2802 lists with it are checked cost-neutral; the
  reconstructed retired5 state now matches #2802's recorded endpoint, local
  profile and fill sha256.
- New rule S3-ABOVE-FLOOR-NEUTRAL: the seven above-floor refit5c frames
  carry no cost.
- Root bracket: the 1e-70 slack was applied in the default 28-digit Decimal
  context, where it rounds away; widening now uses a 100-digit directed
  context, with a written error budget. Conclusions unchanged (margins
  >= 1.0e-20 in all four brackets).
- Verifier emits prelude warp count, 35/14/14 split and refit7c dims; a pin
  mismatch prints a FAIL line. README adds "What this decides", terms,
  downstream claims #2824/#2827/#2833, credits (#2315, #2656, #2767, #2322),
  and Python-version scope.

PASS_T74_RETIRED7_INCREMENT_AUDIT rules=28/28 controls=19/19

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mister-Meeseeks pushed a commit that referenced this pull request Oct 16, 2026
… scope, C4g control, disclosures

- Occupancy: the conformance-not-soundness conclusion now rests on the
  structural fact that no pricing check reads occupancy; C4d/C4e are
  described as a lender's dead last write (after its last read), with
  the dirty-borrow/cleanup mechanism; examples are not a classification.
- New control C4g (401's write A3158, still read at A3159, moved to the
  end of restore at FULL): rejected by both replays, accepted by frames,
  recount, refits, stage records, fill and price. C4f gains the same
  must_accept/exact_failing expectations. Controls 19 -> 20.
- verify.py prints a NOTE line if the zlib fallback of the retired5
  fill-hash comparison runs; README discloses the fallback.
- README: PR2226 identity checked only through its binary-symplectic
  conditions; strict-t74 record carries the supplier-only kappa;
  #2802's structural checks stated exactly; retired5 words rebuilt from
  published stage files; Open Q1/Q3 access via full_replay.py; warp
  vs gate numbering; three root brackets; Reproduce moved up; What this
  decides refreshed at 02:10 UTC (#2839, #2843; seven pending claims).

Result unchanged: PASS rules=28/28 controls=20/20; receipt byte-identical
on rerun.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Mister-Meeseeks
Mister-Meeseeks merged commit a36ebe3 into CrocSwap:main Oct 16, 2026
Mister-Meeseeks pushed a commit that referenced this pull request Oct 16, 2026
… floor

Adds research/t74-2751-joint-twelve-audit, a standard-library, source-bound
independent verification of open PR #2751 (tekuuu, head 4cfbb0c; T74 joint
twelve-lender recovery, conditional kappa ~0.0017171416). It also adds a
computed reconciliation with open PR #2720 (petalas, head 813fae6;
fixed-contract inherited-correction floor). PR files are read as 49 pinned
git objects and never executed. No defect found.

- Both words are exact in the free-operator model through cleanup and the
  target endpoint, forward and dual. All 12 lenders and all 4 pair
  recoveries are checked.
- The change set and per-word delta 4(H8-H7-H1) are checked against
  PR2733@2761b52b.
- PR2720's Q (det -1/16) is reproduced on its own input words. On PR2751's
  words the projection is zero on its rows; the correction moves to
  21/37/47/110 (det 1/2), which are never cut-written.
- PR2720's hypotheses are evaluated from verbatim quotes. Changed: h1, h2a,
  h3, line-30. Kept: h2b, h2c, cut incidences, zero readers, rank-8.
- The complete profile is rebuilt from 276,390 roles in two
  implementations; also checked: stock 66801, the endpoint enclosure, and
  the comparison with PR2733's current head.
- The #2766/#2767 profile identity and the downstream word structure
  (#2802, #2818, #2824, #2827, #2833) are checked.
- 356/356 negative controls are rejected, each by its named rule, with 9
  positive controls.

No new kappa; the selected frontier is unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mister-Meeseeks pushed a commit that referenced this pull request Oct 16, 2026
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.

4 participants