Skip to content

Record that the E7 weight-12 words equal the two receipt orbits - #133

Draft
unpaidmoon wants to merge 3974 commits into
mainfrom
cursor/e7-orbit-hash-partition-3402
Draft

unpaidmoon wants to merge 3974 commits into
mainfrom
cursor/e7-orbit-hash-partition-3402

Conversation

@unpaidmoon

Copy link
Copy Markdown
Owner

Change and mathematical scope

Scoped coverage check, no new κ. The selected conditional exponent stays 781451/250000000.

An independent half-split of the pinned 27-dimensional E7 kernel returns zero plus 1638 words of weight 12, and no lighter nonzero word, across all 2^27 codewords. The same set is returned by low_weight_words at batch sizes 16 and 8. The frozen 63 root reflections in research/e7-bitsliced-exhaustion/source/verify.py send the two receipt representatives to orbits of sizes 630 and 1008. Their SHA256 values are the orbit_sha256 fields of source/RESULT.json:

  • diagonal: d08f659fab919baab7b2c87c0b7e55edb5a3e23e05a313522f0b17e684d8ee39
  • product: a9bb36def7396f184d343e38f410d705a50be3188b40725235f866ad0753309b

The orbits are disjoint and their union is that weight-12 set. This does not recompute quadratic ranks, single-omission kernels, or the 120 ambient root slices, and it does not time the bit-lane engine.

The branch is one commit on canonical main 7c49fe32c55de02c2c873655c804b16e4c3fef0a. A review against this fork's main includes unrelated history. The scientific diff is that single commit:

7c49fe3...cursor/e7-orbit-hash-partition-3402

Task and overlap

Base commit: 7c49fe32c55de02c2c873655c804b16e4c3fef0a on canonical CrocSwap/integer-mult-bounds main. Priority revision 8. Catalogue card replay-runtime. Scope status: refined.

The card asks for a new measured speedup. PR CrocSwap#5729 already published the bit-lane speedup, and research/e7-bitslice-coverage-check compared the iterator with this census while leaving the reflection orbits unreplayed (orbits_replayed=false, open question 1). This package is that comparison.

Session r93, item 6 of the ordered eligible list. Structured handoffs are dropped when the post is completed, a merged result answers the question, or the canonical artifact is an open pull request. That removes the 19 October index, the H25 requests, the 2 November PR6353 review chain (CrocSwap#6355, CrocSwap#6357, CrocSwap#6359, CrocSwap#6360, CrocSwap#6361), and the later CrocSwap#6376, CrocSwap#6378 and CrocSwap#6396 artifacts. The 28 October null-artifact handoffs that were later answered are dropped with them. Prose handoffs already marked DONE or RELEASED, and the PR6551 kappa handoff whose artifact is an open pull request, are dropped. No structured claim was unexpired at 2026-11-03T13:22:29Z. Work already delivered by this account is dropped. The revision-8 "Next bounded decisions" table then supplies six rows, and unnamed catalogue cards follow:

  1. Complex helpers (p104-side-correction, coverage-law-change).
  2. BIT (bit-helper-contract).
  3. New decoders and interfaces.
  4. Supplier audit (supplier-boundary-audit).
  5. All-size audit (uniform-transfer-audit).
  6. This package. Verification tooling (replay-runtime): the orbit-hash partition left open by the Accelerate complete E7 classification replay 17.5x with exact bit lanes CrocSwap/integer-mult-bounds#5729 coverage check.
  7. bit-new-register-model.
  8. r437-carrier-integration.
  9. complex-rank-shape-decoder.
  10. complex-new-helper-inventory.

Nearest work:

The overlap snapshot at 2026-11-03T13:22:29Z covered 572 open PR bodies and 6625 discussion records. No open pull request states this orbit-hash comparison. A title search immediately before submission did not add one.

Validation

CPython 3.12.3, standard library only, Linux. From the repository root:

python3 -B scripts/research_package.py --package research/e7-orbit-hash-partition --check-only
python3 -I -B research/e7-orbit-hash-partition/verify.py
python3 -I -O -B research/e7-orbit-hash-partition/verify.py
python3 -B scripts/research_package.py --package research/e7-orbit-hash-partition
python3 -B scripts/research_selected_baseline.py

Both direct modes printed PASS_E7_ORBIT_HASH_PARTITION supports=1638 diagonal=630 product=1008 reflections=63 quadratic_ranks_replayed=false no_new_kappa=true. Metadata printed PASS_PACKAGE_METADATA_AND_PINS. The registered replay printed PASS_REGISTERED_RESEARCH_PACKAGE with outcome completed. The baseline printed PASS_RESEARCH_BASELINE_BINDINGS_ONLY with selected κ 781451/250000000. Checks raise ValueError, so -O does not disable them.

Controls: the two half-splits must agree; a deleted census word and an overlapping second orbit fail the partition check; deleting one diagonal-orbit word changes its hash; clearing one bit of the diagonal representative changes the hash and leaves the weight-12 set.

Not run: the CrocSwap#5729 timing harness, quadratic ranks, single-omission kernels, the 120 ambient root slices, another decoder code, Lean, Docker, and make frontier6030-verify.

Credit

The geometry and minimum-support statement are Nick Petalas's PR CrocSwap#5238. The classification, frozen reflection code and receipt are Ali Romman's PR CrocSwap#5669. The bit-lane engine is Ali Romman's PR CrocSwap#5729. The word-set comparison that left this question open is research/e7-bitslice-coverage-check.

Prepared by unpaidmoon with substantial Cursor Grok assistance. Apache-2.0.

Open next questions

  1. Do the two receipt representatives still have quadratic ranks 6 and 2, with every single omission of rank 11? The orbit hashes identify the sets and do not rerun that half of the classification. First evaluate form_a and form_b in source/verify.py and compare supported_rank and quadratic_matrix with the receipt.
  2. Does the same set equality hold for one other deletion or decoder code? The bit-lane proof leaves that consumer unbound. First name its generators, width and weight bound, then compare one half-split with low_weight_words on that basis alone.
Open in Web Open in Cursor 

petalas and others added 30 commits October 28, 2026 16:25
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ew scope [skip ci]

Scientific validation and selected-frontier admission remain open.
…ew scope [skip ci]

Scientific validation and selected-frontier admission remain open.
…updated on main)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ult.json to its blob at base_commit

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…reduction and the carrier cost floor

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011VhYmAARGrLk38Gj9eUPzP
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011VhYmAARGrLk38Gj9eUPzP
… the code

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011VhYmAARGrLk38Gj9eUPzP
Mister-Meeseeks and others added 30 commits October 31, 2026 00:04
…ew scope [skip ci]

Scientific validation and selected-frontier admission remain open.
Prepared for @CubeJerry with substantial OpenAI Codex assistance.
No new kappa; all selected framework assumptions remain.
Prepared for @CubeJerry with substantial OpenAI Codex assistance.
No new kappa; no physical supplier admission.
Prepared for @CubeJerry with substantial OpenAI Codex assistance.
Includes independently challenged root-oracle boundary handling. No new kappa.
…rence

Documentation only; mathematical source and executable bytes unchanged.
Prepared for @CubeJerry with substantial OpenAI Codex assistance.
Includes an explicit sparse construction, independent all-output verification,
and exact caller-compatibility and cost limitations.
Prepared for @CubeJerry with substantial OpenAI Codex assistance. No new kappa.
Theorem D: snapping a vertex of the centred fiber polytope to +-1/2 reduces
exact mode-preserving rounding to a signed balancing problem that the
block pair structure of the PR4933 mode matrix solves outright. With the
CrocSwap#6312 lower bound this gives 1 <= R <= 3/2, superseding the R <= 2 of CrocSwap#6320.
Standard-library exact verifier and constructive rounder included.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ew scope [skip ci]

Scientific validation and selected-frontier admission remain open.
The published selection matches its two draft hashes and AST import lists, leaves source adoption false, and names no receipt producer. It does not select the F05 trace control or record the W4 decimal extension and trap names that dependency N1 still requires.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
…tted BIT.

The selected PR6030 stock gate requires ordinary above admitted BIT. PR5708's ordinary component is strictly larger, so it cannot occupy bit_saving. The outer ledger would accept that value, and the complex uniform saving still binds.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
One fixed seven-tape inventory carries two presentations of the short-axis
witness and both heights of the long-axis ell=2 layout. Previous-output
reset keeps the live count at one footprint. No kappa change.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
Record a successor for the two PR6321 custody defects that block the
H25 N22/P07 reference comparison: convert the ordered identity tuple
at the consumer, and close every preflight descriptor without
replacing the first failure.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
When the body succeeds, the first close error propagates and is not
also stored as a secondary. Later close errors remain ordered
secondaries, and a pending body error is still left intact.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
Point the live parent input at the commit that contains the
primary-preserving close helper.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
The displayed gap is Q_{35/16}+Q_{85/8}+Q_{-205/16}+2 L_{-8}-16 L_{-1}.
Coefficient-blind prices win only below 14/3; binary synthesis with u≥ℓ consumes the eleven-event gap. No numerical backend is supplied.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
…leanup order.

Independent replay of the pinned PR5713 source and PR5718 word shows the output nuisance is the input nuisance minus a nonzero partner covector, and that inverse-first cleanup fails exactly rows 126 and 188. No new kappa.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
Extra clean or dirty monomial registers leave the direct m*t call count
optimal whenever the rational frame span has dimension at most m+1 and the
normalized extra frame is not in the |K|=2 block. No new kappa.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
…ment.

On the emitted CrocSwap#5708 word, every one-to-three-contact A2 conjugation keeps a nonnegative paid-query change on (0,1) when exactly one correction moves to another published route frame.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
…ght.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
… trees.

Every induced four-contact tree on the retained 640-input centre graph raises the complete paid moment by at least 3 - 2^theta + 7^theta - 8^theta for 0<theta<1, with helper targets fixed. One path attains the bound. The 62 induced four-cycles stay unpriced. No kappa change.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
…67's word (no new kappa)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ew scope [skip ci]

Scientific validation and selected-frontier admission remain open.
An independent half-split of the pinned kernel matches the two frozen reflection orbits and their receipt hashes. No new kappa.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
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.