Skip to content

Conditional κ ≥ 0.00374039009039 (+0.27743% over #7704) by recoding the role-1256 relay cuts - #7708

Open
aliromman wants to merge 2 commits into
CrocSwap:mainfrom
aliromman:codex/p104-banked-cut-composition
Open

aliromman wants to merge 2 commits into
CrocSwap:mainfrom
aliromman:codex/p104-banked-cut-composition

Conversation

@aliromman

@aliromman aliromman commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Result

This construction gives conditional κ ≥ 0.00374039009039 on the signed 16-stage P104 caller. It exceeds the upper endpoint of Clear1nsight’s #7704 graph by 0.00001034832006 (0.27743%). That increment is 2.48 times the 0.0000041662072 gain in #7674. The selected maintainer-reviewed reference at the base of this PR is 0.00363005817. Parent and its tight bracket · Application receipt

Full application #7704 New construction
Paid recursive jobs 101,840 101,488
Productive jobs 21,152 21,152
Paid rank mass 271,568 271,568
Local scalar terms per native call 172,414 177,531
Numeric bank homes per native call 9,150 9,128
Charged scalar services across the caller 122,001,408 124,542,464
Directed raw κ crossing (0.00373004177032, 0.00373004177033) (0.00374039009047, 0.00374039009048)

Construction and review

The input is the exact #7704 role-1256 graph, reconstructed from SYLforge’s #7700 source word. The ordered certificate has 22 eliminations, 21 grouped hub reroutes and three that recover a signal jointly from banks at two charts. At cut 38 the required signal has recovery denominator five. At cut 40, a sender routes a sum to chart 778, which previously received no edge from the old middle chart. Two paid rank-two jobs become ranks three and one. Cut 42 then reduces them to one rank-four job. This illustrates how one cut makes another possible. Proof and worked examples · Exact witnesses

  • Every intermediate graph has an acyclic schedule and strictly nested replacement routes. Each saved identity is checked on independent inputs. Forward replay checks all 1,501 public rows with 9,280,471 exact products. A separate reverse accumulation compares all 1,501 parent and candidate columns.
  • The native checker verifies rank, signed Pauli action and absolute phase on 3,023 chart pairs. For all 54 distinct changed pairs on surviving slots it also checks 512 forward and 512 adjoint columns. All 104 main adapters and 1,664 port-stage contact bindings are replayed.
  • The complete paid histogram changes at ranks 1–7 by $(−288,+48,−368,+96,+80,+48,+32)$. The target histogram is identical. The exact concave-order comparison gives a strict decrease for every $r^t$, $0<t<1$. Two directed interval evaluators bracket the crossing. The compiler tests 0.00374039009046, and all 54 strict and four functional outer predicates pass.
  • Physical replay includes 4,863,931 nonintegral-coefficient basis divisions at exact ticks, complete charged retirement, four guarded streams per stage and the higher scalar cost. The no-cancellation common-root guard passes with $c=1156$ at the candidate’s own $H=14617400000000000000$. Physical and compiler receipts

The eight registered checks passed locally on arm64 macOS with Python 3.12.16 and the standard library. The standalone constructor emits the candidate graph with canonical SHA-256 db10e4ca7154b454f86bf404bc34b3a16d02de6cc008d5227e4eef2654cb0778. Replay manifest

python3.12 -B scripts/research_package.py --package research/p104-relay-cut-recoding

Attribution and scope

Base e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8. Clear1nsight supplied the role-1256 contraction, the prior role-230 reroute and the open combination question. SYLforge supplied the nine-swap word. DreamingOfClouds supplied the grouped hub method on the separate chart-595 graph in #7705. Ali Romman supplied the #7674 banked pivots. The signed caller follows Rohan Arun’s lineage. Jerry Chen (CubeJerry) supplied the selected application and BIT input. Douglas Colkitt and the recorded contributors supplied the framework and verification infrastructure. Ali Romman received substantial OpenAI Codex assistance with the present research, code, verification and writing. Source inventory · Preserved credits · Original claim

The new bound inherits the five interfaces for uniform native realization, owned erasure, compiler closure, fixed-tape precision and analytic recovery. Independent scientific review and a materialized full multiplier remain open. A reviewer can start with the two worked cuts, then check their physical bank charges and the directed signed moment at the parent’s upper endpoint. Selected proof review

The standalone scientific commit is f0e2db5b5b46b139da17d2b32fab6633164443a9. The separate CI prerequisite carries ToikY’s reuse of Clear1nsight’s #7257 repair. Main’s onboarding test failed locally because its CI guide specifies #6030 while the selected record specifies #7000. The prerequisite matches all five original file blobs, and all six navigation/status tests pass with it applied. Scientific diff · Shared CI repair

Verification at the published head

GitHub passed planning, the historical prerequisite replay and all eight registered scientific checks on 7017edbe4bc5ed1b6d5b6defedbc7e472d83c65c. A Git-free export of the 48 package files and four declared inputs passed all eight checks and reproduced the candidate SHA-256. A fresh full-history clone also verified the committed bytes, file modes and source-commit ancestry. All 27 files in the external source inventory matched the Git blob at their stated original commit. At the parent’s upper endpoint 0.00373004177033, direct 256-bit interval evaluation places the new target-minus-paid margin between 3.64608447709 and 3.64608447710. Hosted scientific replay · Complete run

A separate local dense replay executed all 6,121 paid routes on the full 512-coordinate arrays. For each of the four input fixtures in the parent’s published DENSE.json, the SHA-256 of the 1,501 physical output arrays matched its pre-existing oracle. The fixtures cover two primes. They supply an end-to-end phase and scalar check at width 512. Upstream public oracles

Reproduce the supplemental finite-field check

Use the candidate checkout as the first argument, the scientific package directory of #7704 at e29a5ae8591243c1178067727f12ce4abefa0b93 as the second, and the canonical JSON emitted by recoding.py as the third. The script runs with isolated Python 3.12 and the standard library.

"""Check the exported candidate against the dense public oracles published in PR 7704."""
from functools import cache
from pathlib import Path
from random import Random
from sys import argv, path
import json

root, old = (Path(s).resolve() for s in argv[1:3])
path.insert(0, str(old / "dense_support"))
path.insert(0, str(root / "research/shared-wire-split-composition/frozen"))
import clifford_exact as cliff
import native_verify
from field_interpreter import GateInterpreter, digest

def require(test, message):
    if not test:
        raise ValueError(message)

g = json.loads(Path(argv[3]).read_bytes())
receipts = json.loads((old / "evidence/DENSE.json").read_bytes())["fields"]
moves = {k: (a, b) for k, a, b in g["moves"]}
forms = dict(g["forms"])

@cache
def chart(i):
    return cliff.canonical_frame(9, tuple(g["frames"][i]))

@cache
def body(a, b):
    recipe, _ = cliff.lower_child(chart(a).adjoint().then(chart(b)))
    return native_verify.recipe(recipe)

for field in receipts:
    p = field["prime"]
    engine = GateInterpreter(p, require)
    coeff = cache(engine.coefficient)

    @cache
    def prepared(a, b):
        return tuple(engine.gate(gate) for gate in body(a, b))

    for case in field["cases"]:
        rng = Random(case["seed"])
        banks = {~r: [rng.randrange(p) for _ in range(512)] for r in range(len(g["initial"]))}
        require(digest(banks.values()) == case["inputs_sha256"], "input fixture")
        for key in g["schedule"]:
            row = [0] * 512
            for donor, literal in forms[key]:
                q = coeff(literal)
                row = [(x + q * y) % p for x,y in zip(row, banks[donor], strict=True)]
            banks[key] = engine.apply(prepared(*moves[key]), row) if key in moves else row
        actual = digest(banks[g["first_output"] + r] for r in range(len(g["final"])))
        require(actual == case["oracle_sha256"], "typed dense output")
        print("PASS_DENSE_CUT_GRAPH", p, case["seed"], actual, flush=True)

aliromman and others added 2 commits November 18, 2026 09:13
… relay cuts

Compose 22 paid-edge eliminations and 24 hub reroutes on the exact CrocSwap#7704 DAG. Include the ordered rational witnesses, signed caller, charged physical replay, independent column check and directed outer price. Prepared with substantial OpenAI Codex assistance.
Reuse Clear1nsight's exact CrocSwap#7257 guide and historical-source snapshot repair, already validated by the hosted checks on CrocSwap#7644. The research package bytes are unchanged.

(cherry picked from commit 694667c)
(cherry picked from commit c58206c)
(cherry picked from commit 75cdd54)
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.

2 participants