Skip to content

Conditional κ = 0.001283479303751122: E8 bit word R715 (carrier lower bound + F2 overlap cancellation) with #458's complex supplier (+0.711% over #460) - #463

Closed
DreamingOfClouds wants to merge 9 commits into
CrocSwap:mainfrom
DreamingOfClouds:e8-carrier-rekid-r715
Closed

DreamingOfClouds wants to merge 9 commits into
CrocSwap:mainfrom
DreamingOfClouds:e8-carrier-rekid-r715

Conversation

@DreamingOfClouds

@DreamingOfClouds DreamingOfClouds commented Oct 11, 2026 •

Copy link
Copy Markdown
Contributor

Conditional κ = 641739651875561/(5·10¹⁷) ≈ 0.001283479303751122, bit-bound: +0.711% over #460 (0.001274415506145115) and
+0.741% over #458. Whole-residual hosting on both sides: no cross-family residual split, no cross-copy lemma, no new complex lemma.

side supplier endpoint
bit (binds) R715 E8 bit word, eleven stages, #442 whole-residual hosting, target mixing c ≈ 1.285129·10⁻³ (+0.712% over #459's 1.276042·10⁻³, +0.742% over #454's)
complex #458's paid-adapter kernel, unchanged b ≈ 1.297204·10⁻³

#465 (Srikrishnabalaji) has independently certified this outer assembly (the step #463's draft left open) at 1.283479306671789·10⁻³
with a sharper bit charge, crediting #463; this PR supplies the full bit supplier and its verifier.

What is new: the bit word R = 715 (#436's R740 → 715).

  1. A carrier lower bound, and a word that meets it. In this format a copy of source x_s that first acts above its line frame
    serves only targets in a hyperplane section of s^⊥ (frames contain everything they touched). The 63 lines of E7 = s^⊥ are a
    spherical 4-design, Σ_r ⟨r,u⟩²⟨r,v⟩² = 16(|u|²|v|² + 2⟨u,v⟩²) > 0, so no two hyperplanes contain them: every source needs ≥ 3
    registers, i.e. ≥ 240 carriers; likewise every target needs ≥ 2 helpers reaching its hyperplane (240, met since κ = 8.75481145306751e-4: first E8 bit word (complex-bound; +9.29% over #352) #370).
    κ = 8.75481145306751e-4: first E8 bit word (complex-bound; +9.29% over #352) #370-lineage words spend 272 carriers because each frame line 2e_m enters the D4 sums of seven blocks. Frame-pair D4 sums
    (D4(c) = Bcw[c] + FP + FP', FP = 2e_a + 2e_b over the 12 pairs of one AG(3,2) direction plane) meet the bound: R 740 → 724.
  2. Eleven cancelling re-expressions (a rule rewritten as x ⊕ y of existing symbols; overlap cancels inside its frame): R 724 → 715.
  3. Conditional κ = 0.001138823042306561: retain rank138 with complete 32-replica banks #443's replica kernel rebuilt for the new word (deterministic search, coordinate entrances), and Conditional κ = 0.001347744547714835: eleven-stage cross-copy hosting with target mixing (+1.278% over #448) #450's target mixing
    generalised to chained receivers (a donor that is itself a receiver); it reproduces Conditional κ = 0.001347744547714835: eleven-stage cross-copy hosting with target mixing (+1.278% over #448) #450's mixed word byte for byte.

Verification (python -B research/e8-r715-paid-complex/verify.py /abs/out; CI workflow e8-r715-paid-complex): #458's full
wrapper and the R715 package's full wrapper from scratch, then the exact composition (two moment engines per side, six bootstrap
levels, all 47 strict outer inequalities, next-grid and parameter controls). The bit package regenerates the word from its frozen
hierarchy (legality of every rule and read checked), runs #370's independent all-column checker and mutants, a relaxed copy of that
checker on the kernel and mixed words, #443's C++ cohort legality and formal columns at five and eleven stages, mixing mutants, Gram,
eleven-stage template and hosting audits, and its own assembly. Two fresh-copy runs agree apart from timing fields (results, both
supplier RESULTs and the complex AUDIT byte-equal). The R715 word was also checked by an independent validator.

Review points

Credits: chafreaky (#458 paid adapters, #451/#454 whole-residual hosting and the cross-family obstruction, #443–445 kernel and
staging), rohanarun (#460 composition and verifier structure, #453, #459), Srikrishnabalaji (#465), DaysSky (#442 idle-axis hosting),
Dylan Crow (#370 bit word, producer and checkers), eumemic (#436 R740 word and reuse, #440/#441 retiming), higherbeing (#448),
icekylinx (#433), jacklightChen (#425), Jacob Sussman, Zhihao Chen and all source-pinned authors. Carrier bound, frame-pair
construction, cancelling re-expressions, kernel rebuild, generalised mixing and composition by DreamingOfClouds with
Claude Opus 5.5 (Anthropic), via Claude Code.

🤖 Generated with Claude Code

chafreaky and others added 9 commits October 11, 2026 17:33
…ation

Validated word and exact bit supplier; full package pending.
Prepared by DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…39651875561/5e17

Adds research/e8-r715-eleven-stage-whole (R715 bit word: carrier lower bound
plus F2 overlap cancellation, target mixing, eleven stages, whole-residual
hosting), research/e8-r715-paid-complex (exact composition with PR458's
unchanged complex supplier) and its CI workflow. Two fresh full replays pass;
independent audit passed with its fixes applied.
Prepared by DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@DreamingOfClouds DreamingOfClouds changed the title [Draft] E8 bit word R715 (carrier lower bound + F2 overlap cancellation): bit c ≈ 1.28513e-3, projected κ ≈ 1.2835e-3 with #458's complex Conditional κ = 0.001283479303751122: E8 bit word R715 (carrier lower bound + F2 overlap cancellation) with #458's complex supplier (+0.711% over #460) Oct 11, 2026
@DreamingOfClouds
DreamingOfClouds marked this pull request as ready for review October 11, 2026 17:34
Mister-Meeseeks pushed a commit that referenced this pull request Oct 12, 2026
…over #493)

The R480 word of #503 (PR #493's word, totals on dead registers), priced in
the nine-stage whole-residual hosted pipeline (32 replicas, m = 64; #443/#444
template and emitter with #442 hosting, vendored byte-identical from #463).
All 138,240 helper residuals hosted whole, no banks. Nine-stage formal columns,
template, parent and hosted audits, invoice, both moment engines and #493's
47-inequality outer assembly pass; kappa unchanged (complex-bound).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Mister-Meeseeks

Copy link
Copy Markdown
Contributor

Thank you @DreamingOfClouds. This contribution is now retained on main in the scoped 100-PR review and contributor credits.

Contribution: Carrier-bound-saturating R715 word and overlap cancellation.

Review and limits: Retains the carrier lower bound's compiler restrictions and full paid-complex composition. Earlier review of fourteen vendored files did not cover this entire contribution.

The complete original source, notices and relevant execution evidence are retained through the pinned source archive. The review distinguishes fresh local runs, source-bound historical Linux CI and prior exact-head review; it does not claim 100 fresh full mathematical audits. Closing this historical PR as reconciled. This preserves parallel and cumulative credit without replacing the selected frontier or promoting priced research targets to completed constructions.

Reviewed head: cdf0c1c6540df987467988fd771340670f4136f3. Publication: 64039f1bd3b6a523b96083fd858ad040a75e3030.

DaysSky added a commit to DaysSky/integer-mult-bounds that referenced this pull request Oct 18, 2026
…sses a trivial word at R = v — the checker-scope finding, the completeness retraction, the sparse freedom restored

- ii19_classdecoder.py: the class-blocked decoder verified VALID under the TRUE relaxed
  definition (B = U W^T, diag 1, zeros only on non-orthogonal pairs, FREE on orthogonal):
  rank 28 = 3x9+1, junk = the within-class perp (mean 62.8), cst 235. The earlier 'decoder
  rank obstruction' (rank(M) = 241) tested only the naive symmetric B0 = J+G convention.
  The class word's honest mass-exact prices: -30 to -45% vs the frontier at every mask
  depth (the h=10 window penalty: 2v(h-1) mass, m=82 root parameter, width-10 hosting).
- e8_sparse_decoder.py: the masked-B0 hunt on the FRONTIER SYSTEM (E8, h=8): random masks
  rank 117 at every f (total B0 rigidity), prefix masks 75-107 (structure matters but
  heavy). Frozen-shape calibration: the R434 parameters price 0.0016257 vs the true
  0.0017444 (the projection family under-prices ~7.3%). The low-rank sparse decoder
  (c <= 25, rows 12-25; win zone mass/v ~37-40 vs 43.5) NOT FOUND in the masked families;
  asymmetric/row-space constructions remain.
- e8_identity_word.py: THE DECISIVE EXPERIMENT. The trivial word (B = I, one retained
  register per line, NO phase-B reads, R = v = 120, D-phase pre-reads cancel the scattered
  dirt by doubling) PASSES GCHECK: status PASS, hist {1: 240, 7: 360}, rank_mass 2760 =
  8x120 + 120 + 2x120x7. The price model REFUSES it ('no root' — implied root ~0.38,
  far outside the modeled bracket [0, 0.05]).
- THE CHECKER-SCOPE FINDING: gcheck (and the E1-style conclusion decomposition
  U_block_form / J_identity / K_equals_D / C_is_U_inverse, read in xcheck_e1.py) verifies
  the FLOW — the delivery (x_t), the dirt cancellation, the helper restoration — NOT any
  multiplication completeness. The identity word passes all of it. A gcheck PASS is
  necessary, not sufficient: the pipeline layer (the hub's CrocSwap#463/CrocSwap#510 template: hosting
  and all-size realization) is where the words' real requirements live.
- RETRACTION: the 'completeness forces B = B0' derivation was wrong — the products are
  extracted by the outer pipeline via the norm-polarization (Q(x+y) - Q(x) - Q(y) = 2<x,y>,
  the full double sum; no per-pair coverage needed inside the word). The sparse-decoder
  freedom (HANDOFF 1444) is RESTORED at the flow level.
- Open: the low-rank sparse-decoder search (c <= 25); the pipeline-requirements read.
  The frontier (E8 R434, c = 0.0017444) stands unbeaten.
DaysSky added a commit to DaysSky/integer-mult-bounds that referenced this pull request Oct 18, 2026
…ized) — and three measured walls; the price regime discovery (R/v >= 3.5)

- PHASE 1 (e8_rowspace.py): the rank-9 row space PROVEN rigid by exhaustive enumeration:
  all 512 codewords have weights {0: 1, 56: 255, 64: 255, 120: 1}; every target's ONLY
  legal codeword is its own full perp (weight 64). No sparse decoder at rank 9 — proven.
  (RAM incident: the first run's basis reduction was broken — 48 pseudo-basis vectors ->
  a 2^48 enumeration -> 22 GB, killed; fixed with the dict-pivot elimination + an abort
  guard. The macOS ru_maxrss is bytes: the old 'rss MB' labels were KB.)
- PHASE 2 (e8_enrich.py): THE SPARSE DECODERS EXIST — the cancellation construction
  r'_t = r_t XOR ind(perp(t) & perp(x_t)): rows legal BY CONSTRUCTION (the XOR only
  removes bits from the full perp). Every orthogonal pair's common perp is EXACTLY 30
  lines (7,560 pairs, one size). Greedy partner curve: k=1: rows 34, rank 113;
  k=2: rows 30, rank 33; k=3+: rows 4, rank 57 (saturated).
- PHASE 3 (e8_sparse_word.py): the economics — three walls, all measured:
  (a) SHARING LOSS: the uncoordinated greedy randomizes the junks — 120 distinct sets,
  pairwise overlaps mean 9.2 (the I13 pair-specific signature): the E8's shared middle
  calculus is destroyed; (b) CST PENALTY: retained 33 (cst ~200) vs B0's 9 (cst 72):
  the decoder-side cost exceeds the load savings at R-parity; (c) THE PRICE REGIME: the
  frozen-shape family reproduces the R420's published c EXACTLY (0.001653826988146798944268,
  ratio 1.0000 — the perfect control) and prices the R434 parameters at 0.0016257 (the
  7.3% calibration), but the ENTIRE k=2 grid (R 323-403, mass/v 36.8-43.4) returns
  NO ROOT: the root's existence is R-driven — the model roots only at R/v >= ~3.5
  (published words: R420 = 3.50v, R434 = 3.62v, P111 = 3.95v). The lighter sparse words
  (R/v 2.7-3.4) fall in the IDENTITY-WORD CLASS: the model's refusal band.
- VERDICT: within the regime the k=2 word prices below the frontier (the cst + sharing
  penalties); outside the regime, unpriceable (the hub's template question — the same
  open question as the identity word: whether the pipeline could ever use lighter words).
  The frontier (E8 R434, c = 0.0017444) stands unbeaten.
- LIVE REFINEMENTS: the frame-aligned cancellation (coordinated partner choice
  preserving the E8 sharing — the R420's own x(t) 4-frame partners); the exact
  minimal-cst factorization; the regime question itself.
- SESSION-LEVEL DISCOVERY: the format's validity lives in the PIPELINE/MODEL layer
  (the R/v ~ 3.5 regime), NOT the CHECKER layer (the identity word passes gcheck).
  Checker hierarchy: gcheck (flow) < xcheck (flow + delivery decomposition) < the price
  model's regime (the pipeline's asymptotic requirements — the hub's CrocSwap#463/CrocSwap#510 template).
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.

3 participants