Repository navigation
Tensor block norm accounting: T114 raises conditional kappa by 5.160482894338e-11 - #2839
Merged
Merged
Conversation
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>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The T114 allocation raises the conditional integer-multiplication κ to 0.0017177952742214166865779999999999999999999998999999999999999999999999, an exact gain of 2580241447169/50000000000000000000000 over the previously delivered T104 package.
The improvement uses the existing whole-family tensor-isomorphism normalization to replace total-call norm growth with eighteen active steps per physical replica. Exact all-column checks in both providers/orientations exclude cross-replica dirty-response leakage and verify cleared private work. Full group-indexed stock conversion is charged: the conservative prefix is1511<2048 and row19547<20161. All1026calls and their children remain paid.
The actual418608-role allocation passed independent chronology/family/capacity/mask checks and exact directed-rational moments. Every changed native-consumer/global/invoice/bootstrap/outer/full-release stage passed freshly; unchanged native chains were authenticated reuse. Fresh Mathlib-only Lean4.34.1 kernel checks cover67 declarations/60 theorems and four real axiom/ownership fixtures, with standard axioms only.
Inherited all-size/compiler/source-access/hosting/precision/tape/recovery/ordinary-recursion/bit-supplier/analytic-transfer hypotheses remain. CREDITS.md and FRONTIER.md identify relevant earlier PRs, including tensor normalization, Schur/gauge/retiming work and the native supplier chain. No general TI algorithm or global novelty claim is made.
Attribution and overlap
This derivative retains PR2802's actual native words and uses a new source-bound
application of elementary owner-block norm accounting to the precision invoice.
The T114 gain has a new checked literal owner allocation and complete
admission. It does not claim a new general tensor-isomorphism identity, a first
TI application to multiplication, or an improvement from covariance alone.
Direct native and allocation ancestry
Marius-Paul Dumitrean's PR2802,
head
17b2ee6ef3cf1468e78a6267f0ecbc9a0e6f60dd, supplies the retained retired7native words, frame refits and restoration; its NOTICE records substantial
Anthropic Claude/Claude Code assistance. Tekalign's PR2751 supplies joint
twelve-source recovery. PR2766 and PR2733 contribute frame-route/refit work;
Philippe Van Velzen's PR2679 contributes pre-life hosting and retired-lender
aliases; Mason Hale's PR2696 contributes the assignment; zxv's PR2530 supplies
source-partner loans; Jeremy Kothe's PR2588 supplies helper-label exchange;
Cinna Davis's PR2226 supplies anticipatory dirty borrowing. Their exact notices
and licenses remain authoritative.
PR2824, by Dugongue,
is the preceding checked T94 tail-dilution construction. The succeeding admitted
T104 allocation and its authenticated native receipts are direct predecessors.
The native words and paid local mechanism are inherited; replica dilution does
not earn credit for a new word construction. Existing Rohan Arun and selected
icekylinx/source-frame/compiler lineage remains credited by retained notices.
Earlier tensor, gauge and response mechanisms
jmbarrancoml)PR2742 records substantial OpenAI Codex assistance and credits ywz65/PR2491 for
the full dirty reader, signed data permutation and framed interpretation; Ross
Richardson/PR2500 for the dyadic chart and typed cost audit; Cinna Davis/PR1391
and PR1410 for order-four echo; Aaron Wang (
navinor)/PR2595 for phase windows;and Ali Romman/PR1719 for the Gaussian-dyadic compiler/interpreter. An unsigned
ERASE transport rule does not apply automatically to its mixed helper chart.
PR2341 (bitplane), PR2378 (AAAmirinu) and PR2422 (bitplane) provide earlier
conditional whole-array tensor/monomial semantics. Their retained credits include
Jose Manuel, Gareth Davidson, AAAmirinu, DreamingOfClouds and Lemma G contributors,
plus Mathlib. These interfaces do not by themselves discharge the full physical
compiler. Relevant older R715 lineage also includes DreamingOfClouds/PR463,
Rohan Arun/PR467/470 and chafreaky/PR458/472/502.
This review's scope
The current-source endpoint/block audit, finite allocations, directed arithmetic
and new Mathlib energy lemmas are distinct from inherited native evidence.
Authenticated reuse must be described explicitly. Current reader symmetries and
failed bounded gauge searches are research evidence, not additional kappa gains.
The finite-group and direct-sum mathematics is classical; no global novelty
claim is made. FRONTIER.md records the bounded source coverage and
missing historical records. Exact source hashes and original notices take
precedence over summaries or mutable PR titles.