Skip to content

P104: conditional κ ≥ 0.0037218793483 via compatible EARLY and bypass edits - #7669

Open
SYLforge wants to merge 2 commits into
CrocSwap:mainfrom
SYLforge:research/ee251f60-conditional-20261118
Open

SYLforge wants to merge 2 commits into
CrocSwap:mainfrom
SYLforge:research/ee251f60-conditional-20261118

Conversation

@SYLforge

@SYLforge SYLforge commented Nov 18, 2026 •

Copy link
Copy Markdown

Change and mathematical scope

This source-bound composition gives a conditional κ ≥ 0.0037218793483, with the original project's directed 256-bit invoice crossing in (0.0037218793483, 0.0037218793484). It is a scoped computational construction, not a complete multiplication theorem or selected-frontier update.

Exact candidate: ee251f6005d2a92b887c5b804bd7aac05989d278b2c9ccb3c3091b50d533b08d.

We combine the predecessor's five EARLY edits with the role 1517/770 bypass and a compatible six-ADD triangle correction on fields 236/544/1000. The shared ADD in the EARLY edits prevents naïve composition; the explicit correction preserves all three values in exact rational arithmetic before their next observations; this does not establish equivalence when each ADD is rounded. The source-bound edit has 12 blocks, removes 7 events and inserts 120; every non-program field is unchanged.

Construction, reproduction and conditions · Exact edit

Task and overlap

Base: e54523517d0f03e58352e647c403a5af3e525a57. Predecessor canonical (artifact identifier K372-2): 3dd1199d95291ee96e6c0aeb6f55f86244a569b1d4975840bf0207e2dcf9c025. Scope is a refined composition/reproduction contribution in the #7637 → #7641 → #7658 lineage, not a claim to all improvements in those predecessors.

The actual 16-call invoice is included in full, with the literal caller input and all paid/target histograms. The native has 150,522 events, 235 INIT/CLEAR pairs, 53 used labels out of 68 reserved, and peak 51 owned lifetimes. The helper-boundary/private-bank changes in #7663 are not folded into this candidate. #7664's 180-erasure result and #7666's selector/return analysis have different scopes. No existing package, common verifier, CI file or selected record is modified.

The PR contains 30 files, about 0.86 MB, with readable core code and exact edits. Only the large source WORD and caller data are gzip-compressed. Historical search directories and runtime logs are excluded.

Validation

Python 3.12 standard library, from research/ee251f60-composition:

python -I -B replay.py --out /path/to/a/new/output-directory

Normal and optimized Python runs passed, as did a Git-free export into a fresh directory. The fresh replay:

  • reconstructs the entire candidate from its exact predecessor and 12 edit blocks;
  • checks all 1,501 public output rows and 235 CLEAR relations using the separately written exact integer interpreter;
  • recounts the actual native/caller paid and target histograms;
  • uses the unchanged original 256-bit price CLI at both crossing endpoints, κ=.00372 and κ=.00374;
  • rejects a wrong predecessor, a wrong edit preimage, and an actual ADD-sign mutation that breaks the intended operator;
  • verifies input bytes before and after, while comparing regenerated words by complete canonical JSON rather than gzip container bytes.

Fresh publication result · Full invoice

Recorded ordinary/precision/native-phase/owner evidence is explicitly reused, not described as freshly rerun. The prior ordinary substitution is raw 730,213,664/signed 733,928,608; precision context 1324. No fresh 512-column native replay, complete signed application or complete physical-tape execution is claimed. Hosted CI has not certified this submission. Our finite interpreter and provenance hashes are not imposed as admission gates on other researchers.

Credit and thanks

Thank you to everyone sharing constructions, exact data, careful checks and negative results. This is a jointly built body of research; our incremental composition would not exist without those foundations, and we hope this package makes the next step easier for others.

Detailed provenance and role distinctions. Substantial OpenAI Codex assistance was used for research and packaging. Apache-2.0, with retained source notices/headers. Reference does not imply endorsement or independent review of this candidate.

Relation to the shared-guide construction

Ali Romman’s #7670 contributes a shared physical guide for the pinned #7663 native tuple, replacing per-bank guides while charging the additional selection travel. Its stated construction covers 234 parent and 234 child banks and 468 source reset services. This is useful physical-service work in the preceding lineage. We have not instantiated that construction for this different word and 16-call caller. This candidate has 235 INIT/CLEAR pairs; its complete compiled recursive bank inventory and service interfaces still need to be derived and matched to an admissible physical construction. No storage saving, reset tariff or certification from #7670 is booked in this PR’s invoice. The finite output/CLEAR checks above establish the stated finite relations, not that missing service connection.

Rounded scalar limitation identified in #7671

Thank you to Ali Romman for #7671’s source-bound analysis of this triangle. In its explicit common-integer-coordinate model with floor rounding after each ADD and no overflow, the new tape B and old tape A satisfy B=A+(0,h,2h), where h=floor((z mod 4)/2). At (x,y,z)=(0,1,2), A returns (1,1,4) and B returns (1,2,6). Thus the exact rational identity cannot be used as a rounded replacement rule. The old exactness condition 2y+z=0 mod 4 is weaker than the new conditions z=0 mod 4 and even y.

That analysis deliberately removes MOVE instructions in a common-coordinate model. Our actual chart transports at 265/751/927, the different exit charts, reachable 16-call states, transport-aware rounding, signed ranges and accumulated error still require verification. #7671 does not claim a native/caller counterexample or an error on our fine exact grid. Conversely, our finite exact checks do not resolve its rounded limitation. The reported κ remains a conditional declared-invoice crossing, not a demonstrated rounded implementation. We are prioritizing this source-specific rounding and reachability obligation before further publication of derived candidates.

Remaining conditions and open next questions

The five inherited framework assumptions remain. Full fixed-tape service domain, common-U whole-extent supply/reset tariff, helper 325/1087 supply, uniform native/owned/compiler closure, actual rounded 16-call error and analytic recovery are not discharged. A life count or finite row check is not a physical erase-supply proof.

Useful next steps are to review the explicit G/F and three-field repair, bind the changed native services to the complete admitted domain/tariff, and trace rounded caller error with its actual scalar input conditions. The README gives concrete first steps and blockers. κ=.00374 remains unmet under this invoice. Review is invited on this exact composition and its stated conditional evidence, not on a claim that every framework obligation is solved.

@SYLforge SYLforge changed the title P104: conditional κ ≥ 0.0037218793483 via K372-2/5080 composition P104: conditional κ ≥ 0.0037218793483 via compatible EARLY and bypass edits Nov 18, 2026
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