Skip to content

Preserve the Gaussian copy floor with private routing headers (no κ) - #7516

Open
petalas wants to merge 1 commit into
CrocSwap:mainfrom
petalas:research/w4-1113b-interface-obstruction
Open

petalas wants to merge 1 commit into
CrocSwap:mainfrom
petalas:research/w4-1113b-interface-obstruction

Conversation

@petalas

@petalas petalas commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Scoped written proof with exact finite corroboration, produced by Nick Petalas. A localized private-reader Gaussian routing header preserves the nonlinear copy-cost floor when its adaptive payloads stay inside its paid reader set and each encoder can compute the header from its declared original source inputs.

M=h(X_A), visible only to R
eta = retained weighted-clique incidence meeting both A and R
E[C] >= (2/3)B + c(R,A)L - eta H(M) >= (2/3)B

For equal b-bit demands this retains expected and worst-case charge at least 40b. The new step is a private schedule simulation with a joint source/reader transcript cut. Source and reader supports, variable prefix codes, branch-dependent payload lists, nonlinear encoders and uncharged schedule leakage are explicit in PROOF.md. The cited rank-one support pair has c=eta=1, so this argument establishes no positive private-header surcharge.

This takes the localized/private-header direction in ywz65 #2641, head d4563ba. Credit thakrarsagar #738 for the geometry and joint-support inequality; ToikY #888, azide0x37 #785, shiftybitshiftr #814 and CubeJerry #912 for the retained transcript/header lineage, retaining the earlier source lineage. petalas #4897 concerns public readers and a separate strict surcharge; its theorem and construction are background, not steps in this private-schedule proof. No predecessor scientific replay or geometric census runs here.

Advance claim. Canonical main remained e545235 on the submission refresh. Source and author overlap, active claims, open PRs and merged/main object searches were checked. These checks exclude unpublished work and arbitrary historical diffs. Only research/private-gaussian-header-cut is added.

Evidence

Before, #2641's theorem assumes a public full-space header and explicitly leaves localized/private headers outside its scope. After, the joint-cut proof covers one private header under the stated encoder-knowledge and schedule-visibility contract.

  • PASS_PRIVATE_GAUSSIAN_HEADER_CUT: all 316 declared-source/header-table choices on three bits crossed with seven nonempty private-reader sets give 2,212 protocols, 53,088 receiver checks and 15,484 exact integer entropy-cut checks. These are generic private protocol checks with no side information, not a Gaussian geometry census.
  • Three six-active-port protocols, using parity, nonlinear and variable-prefix headers, pass all 64 inputs each, 384 receiver checks and 63 cut checks each. Their exact charges are 7, 7 and 29/4 under the cited support-pair price.
  • Seven rejecting controls cover three invalid prefix codes, an escaped private recipient, missing encoder knowledge, a changing fixed schedule and a missing demand payload. A two-input zero-payload-bit presence witness exposes uncharged schedule leakage.
  • Local CPython 3.12.15 normal and optimized warning-strict runs pass in about 0.12 seconds each, below their 120-second registered timeouts. Receipts match exactly and bind all six package files. Registered metadata/replay passes. Committed bytes and source-history preflight pass at 34b7d83. A Git-free six-file export reproduces the same optimized receipt. In-memory compilation, AST/function-length inspection and whitespace checks pass.

Exact reproduction from the repository root, with any CPython 3.12 executable substitutable for the local path:

/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -W error -B research/private-gaussian-header-cut/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -O -W error -B research/private-gaussian-header-cut/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/research_package.py --package research/private-gaussian-header-cut
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/check_research_commit.py --package research/private-gaussian-header-cut --check-source-history

Limitations and unrun obligations

The geometric incidence/support inequality and sharp support pair are cited premises, authenticated as historical provenance. They are not independently recounted or accepted by this replay. The quantified theorem rests on the written simulation and entropy argument; finite checks corroborate it.

The private list's readers must lie in R, and every adaptive encoder's declared source set must contain A. Receivers outside R observe no private value, timing, presence or routing information. Arbitrary private encoder knowledge, relays, correlated demands and changed physical prices are excluded. No strict surcharge, complete optimal Gaussian code, physical supplier, restored-helper construction, native/tape/precision/stock bill, selected-frontier change or multiplication theorem is claimed. All five selected framework assumptions remain. No selected cold replay, Lean or Docker run is required or performed. Independent human and formal proof-kernel review are unclaimed.

Open next questions

  • Replace the sufficient A-subset-S encoder requirement with an explicit factorization h(X_A)=g(X_S), retaining a complete knowledge and routing contract.
  • Determine whether a complete private-routing code attains 40b, or whether a stronger lower bound forces a surcharge. The parity witness here attains only the header entropy comparison.

Related open work

No open PR targeting Gaussian private routing or its nonlinear copy-cost floor was found in the refreshed open-PR corpus. The completion check listed all 113 open PRs on this account and examined the eight created since the CLAIM, including this PR. The seven others, #7511–#7515 and #7517–#7518, study different objects and supply no competing value for this bound. The merged #2641 and #4897 use different public-reader contracts. Broader Gaussian phase, transport and conditioning proposals do not compute this quantity.

CI notes

CI did not certify this package at unchanged head 34b7d83. In run 43442770498, the plan job 131944347029 failed in "Onboarding helper tests" on exactly one of 760 tests, with 24 skipped:

test_research_current_status.CurrentCiNavigationTests.test_current_ci_section_matches_selected_record

The assertion was that 'make frontier7000-verify' was not found in the CI verification doc. I read the job log and confirmed the underlying files on canonical main e545235: certificates/selected-result.json names make frontier7000-verify, while docs/ci-verification.md still describes #6030 and make frontier6030-verify. This PR changes neither file. The package replay job was skipped after plan failed, so it supplies no hosted package evidence.

I also read the failed-job logs for rohanarun run 42905094314 and dillona run 42903313790. Each reports the same exact failing test and missing command, with one failure among 760 tests and 24 skipped. These are the other runs verified here; no broader list is asserted.

Both registered package commands and the publication checks passed locally as recorded above. No code change, retrigger commit or CI rerun was made. The repository-level documentation mismatch remains for maintainers. Local replay is not independent human or formal proof-kernel review.

Merge Danger

Door: two-way. Blast radius: isolated. Six new package files; no shared runtime, selected certificate or predecessor bytes change.

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