Skip to content

Deferred readouts on the cyclic-strip complex network: conditional κ = 9.39074e-5 - #113

Closed
rohanarun wants to merge 7 commits into
CrocSwap:mainfrom
rohanarun:research/deferred-cyclic
Closed

rohanarun wants to merge 7 commits into
CrocSwap:mainfrom
rohanarun:research/deferred-cyclic

Conversation

@rohanarun

@rohanarun rohanarun commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

Validation complete: full repository coverage passed (78 isolated test modules, 20 historical patch checks), both fresh producer replays passed, and all 36 corrected research-head CI checks passed. The original full invocation tested 4dcb515; a separate replay checked the expanded-charge certificate at 070b4d6 after verifying unchanged producer/profile bytes. Both logs and exact commit coverage are recorded in research/deferred-cyclic/validation.json.

Conditional κ = 469537/5000000000 = 9.39074×10⁻⁵, +9.39% over #110 (429239/5·10⁹), +9.47% over #111.

#110 showed that deferred readouts pay on both stopped networks under #104's one-child-per-residual rule; its binding side was the complex network on #104's unchanged h=24 DAG. #111 found a much smaller complex DAG (cyclic interval strips in the PairedTriple recursion + #108's dual-suffix pair stars). This PR runs #110's deferral pipeline verbatim on that DAG.

h=24 complex #110 this
matched carriers 8,966 37,802
roles R 44,918 38,506
deferred roles 8,893 5,887
complex saving (binding) 8.58628e-5 9.39253e-5
κ 8.58478e-5 9.39074e-5

Bit side: #110's one-child profile of Swapnil Jain's round-seven deferred word, unchanged (stopped saving 1.24019e-4).

Checks: complex_deferred.py (maximum carrier matching under the inherited adjacency, explicit role compile, lifted binary frames, deferral, replay with arbitrary scratch, exact F2 frame checks; pins 37,802 links / 38,506 roles) and certificate.py (exact moments, next-grid rejection of coarse bit saving, complex saving and κ, 47 strict constraints, 7 margins; requires κ above #110 and #111). Scope and stated-but-unchecked steps are exactly #110's (PROOF.md, with an addendum).

Credit: Avi Eisenberg (#110, #62), Swapnil Jain (deferred readouts), icekylinx (#104), Rohan Gupta / gupt1156 (#55 dual-suffix layout), #107–#109/#111, and all predecessors in #104's NOTICE/SOURCES. Composition by Rohan Arun with Anthropic Claude assistance.

Independent local verification reproduced #110’s baseline profile and this branch’s complete complex profile exactly, and reran the exact certificate. Complete repository and corrected-certificate coverage has now passed, as documented above. The inherited complex stage-two complement/time-reversal argument and all-size interfaces remain written proof dependencies; these finite checks do not establish those general arguments.

CI status: the initial joint jobs failed at an inherited Makefile fingerprint, before the new construction checks. The three stale inherited root-file fingerprints (Makefile, README.md, NOTICE) were repaired. The original independent h23/h25 physical verifier passed, including full dirty-basis replay, all frame paths/profiles and exact recurrence checks. The derived certificates differ only in checksum metadata; mathematical values are unchanged and no check is disabled. A full corrected-head rerun is starting.

Scalar-accounting correction (credited to eumemic #114): the inherited original-DAG charge did not explicitly cover expanded deferred readouts. The certificate now charges their complete literal operations, raising G from 2,019,773,888 to 76,407,506,746,560 and recomputing precision constants/cutoffs. κ and both physical profiles are unchanged. Independent exact recertification passes all 47 constraints, seven margins and next-grid exclusions. The full run continues on the preceding fingerprint-repaired commit; its final coverage will be combined with an explicitly recorded replay of this corrected certificate before readiness. #114 now reports a larger κ; this PR is no longer the current record.

icekylinx and others added 5 commits October 9, 2026 07:11
…78e-5

Under PR104's opposite-bank rule every residual is one child, so deferred
readouts pay on both networks. Bit: Swapnil Jain's round-seven deferred word,
unchanged (stopped saving 1.2402e-4). Complex: PR104's h=24 network with lifted
binary frames and deferred readouts on 8,893 of 44,918 roles (8.58628e-5).
Replay with arbitrary scratch, exact F2 frame checks, exact moments and the
47-constraint assembly give kappa = 429239/5000000000.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ppa = 9.39074e-5

Run PR110's deferral pipeline unchanged on the complex DAG with cyclic interval
strips and dual-suffix pair stars (38,506 roles instead of 44,918). Complex saving
939253/10^10 still binds below PR110's bit side; kappa = 469537/5000000000,
9.39% above PR110.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011S35E2acyQDaVRjPRw4TDy
@Mister-Meeseeks

Copy link
Copy Markdown
Contributor

Thank you @rohanarun. This contribution is now recorded on main in the scoped reconciliation review and contributor credits.

Contribution: Refinement of deferred/stopped accounting.

Review and limits: Exact-head CI evidence. Early copied PROOF sections describe an older checkpoint; current certificate and README control its historical claim.

The complete original source, proof/research notes, notices and relevant replay/CI evidence are retained through the pinned source archive. The review distinguishes fresh runs, exact-head historical CI and prior integration evidence. Closing this historical PR as reconciled; its obsolete headline does not replace the current selected bound, and draft/retracted claims retain their explicit qualifications.

Reviewed head: 502761e7e02e25c737f61be718e29bca3f391d19. Publication: b84ba67156a01f9c2cb34fe1e3cb1dcdeab00d9c.

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.

4 participants