Skip to content

Conditional κ = 0.003646777872835 (+1.25916395e-7 over #7334) - #7480

Open
rohanarun wants to merge 5 commits into
CrocSwap:mainfrom
rohanarun:research/kappa-fused-449-return
Open

rohanarun wants to merge 5 commits into
CrocSwap:mainfrom
rohanarun:research/kappa-fused-449-return

Conversation

@rohanarun

@rohanarun rohanarun commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

The complete conditional signed 17-stage price improves from exact #7334's κ = 0.00364665195644 to κ = 0.003646777872835, an exact gain of 0.000000125916395.

Helper 449 bypasses a four-event block at chart 724 using low peer 1206. Readout 1477 explicitly pays an extra visit to chart 904. After charging that split, native paid ranks 3+5 become 2+6. All productive endpoints remain fixed: the complete caller has 22,461 productive and 110,519 paid positions, surplus 1,424 and no empty-root credit.

All four direct source-aware stages and one uninterrupted complete registered replay passed, reproducing every frozen receipt against unchanged scientific bytes. Coverage includes whole native forward/adjoint and actual COPY, fresh grids, ownership, ordinary costs, physical maps and phases, the complete signed caller, all 3,037 catalogue maps, precision 2430, fixed tape, 54 strict and four functional outer checks, frontier rejection controls, and the fixed-profile packing ceiling. Complete caller ordinary upper bound: 782,784,552. Commit bytes, modes and source-history checks also passed.

Reproduce with Python 3.12:

python3 -B scripts/research_package.py --package research/p104-helper-449-fused-return

See PROOF.md, REPLAY.txt and REPLAY-INTEGRITY.json. Hosted CI status is separate from these local checks.

Exact source #7334 head: aaeec8c0acf01e99a5f331ef363a698d9c084ace. All nested sources, notices and credit are retained, including Clear1nsight's #7334/#7314 constructions and Rohan Arun's #7277/#7113 lineage. #7338 fixes other paid itineraries; this witness changes and charges the high readout itinerary.

The first direct full-stage invocation rejected a mistaken adapter census assertion: the predecessor has 73 lifetimes and the candidate 74. The assertion was corrected; the word, invoice and κ were unchanged. The successful full and registered runs recomputed them. Peak 50, 68 declared owned labels and owned ascent mass 56 remain fixed.

Five inherited assumptions remain: native_all_columns, owned_erasure, compiler, precision_tape and analytic_recovery. Local source-aware verification is not maintainer selection, independent human review or formal-kernel acceptance. Selected records are unchanged. Includes the two attributed #7257 navigation/replay prerequisite commits, as in #7277.

Clear1nsight and others added 4 commits November 12, 2026 23:07
This is the exact documentation prerequisite retained by CrocSwap#7156. It points the onboarding check to the selected CrocSwap#7000/CrocSwap#7001 review. Credit to Rohan Arun and the source authors.

(cherry picked from commit bb36fc6)
The PR CrocSwap#6292 rejection census and package plan are statements about 2485805. Load its exact sources and planner in a private sparse checkout, keeping all source hashes, 19 expected rejections, nine package selections and the empty-plan control. Subsequent edits to the live Makefile and CI guide no longer invalidate this historical check.

The original audit is by unpaidmoon, building on Dennis’s historical replay gate. The pinned repository view follows the precedent of Matarener’s PR CrocSwap#5434. Repair prepared with substantial OpenAI Codex assistance.

(cherry picked from commit d2e54af)
@rohanarun rohanarun changed the title Conditional κ = 0.003646777872835: fuse helper449 with a paid readout Conditional κ = 0.003646777872835 (+1.25916395e-7 over #7334) Nov 13, 2026
@rohanarun
rohanarun marked this pull request as ready for review November 13, 2026 04:23
@Clear1nsight

Copy link
Copy Markdown
Contributor

Independent source-pinned review of 58a9d052143f5d85295ed2d0163591c47f89a331 (scientific execution source 206585f6125a0aaa2501e808dc4db1228e6e3497), against main e54523517d0f03e58352e647c403a5af3e525a57.

PASS: all four registered stages completed in 891.169 seconds, exit 0; peak aggregate RSS 1,083,392,000 bytes (1,033.203 MiB). All 28 package files remained byte-for-byte unchanged and the exact-head checkout is clean.

Executed in a separate exact-head checkout with CPython 3.12, one CPU, a 2 GiB aggregate RSS ceiling and per-process address-space limit, the original 180/480/495/45-second stage caps and 1,200-second overall limit:

/usr/bin/python3.12 -I -B scripts/research_package.py --package research/p104-helper-449-fused-return

The review checks the unchanged emitted word 5044adee019f073f286789ccb688ffca76d42b60518242bb4daea818d33d337a, reconstruction from exact #7334, arbitrary-input forward/adjoint and actual COPY, native physical maps and phases, ownership, fresh grids, full signed #7113 caller, all 3,037 catalogue maps, precision/tape, ordinary charges, both directed prices, outer predicates and rejecting controls. The complete signed claim is κ = 0.003646777872835, the stated +0.000000125916395 over #7334. The 18-stage ingredient price is a separate intermediate result.

Focused accounting/return observations:

  • Across 17 calls the paid delta is {2:+17, 3:-17, 5:-17, 6:+17}; the extra 1477 visit is charged. Productive endpoints stay fixed; paid/productive positions remain 110,519/22,461, surplus 1,424, with no empty-root credit.
  • The additional owned lifetime is counted: 73 → 74, with peak 50, 68 declared labels and ascent mass 56 retained. The full signed ordinary increase is 69,088 = 17 × (4,000 + 4 × 16), and the complete ordinary upper bound is 782,784,552.
  • The new auxiliary's normalized pre-CLEAR value is a+c−d, which can be nonzero. This review accepts only the explicitly stated conditional owned-erasure contract; it does not establish numerical-zero cleanup or unconditional reversibility.

For reproducibility, the original interrupted attempt and subsequent 1 GiB attempt remain recorded as unsuccessful. The latter stopped at 441.648 s on the aggregate RSS cap (1,074,225,152 bytes); this was a resource rejection, not an operator counterexample. The user explicitly authorized one 2 GiB retry with unchanged source and time limits. The approved retry passed. Raw stdout SHA256: 510748fcd1feea0f1e3da3d1bd85812a254e148e6b20f2cfc3d7a1976653dc47; review receipt SHA256: 8c161fc165f45e6a104854f9dfc42d99b50808d78a67d50e68f4ac06efe9d5f8. These hashes identify local evidence, not a separately hosted artifact.

Hosted evidence is separate: run 43365180333 actually passed this package's replay job, including PASS_REGISTERED_RESEARCH_PACKAGE, on the test merge of this head into the pinned main. It was not merely a planning-only pass.

The five inherited assumptions remain native_all_columns, owned_erasure, compiler, precision_tape and analytic_recovery. This is an external source review and execution of the retained engines, with substantial OpenAI Codex assistance; no independent human review or formal-kernel acceptance is claimed. The gain and construction credit remain with Rohan Arun, Clear1nsight's #7334/#7314 constructions and the attributed #7277/#7113 lineage, not this replay. Main and selected records were not changed.

The remaining admission step is maintainer scientific/frontier review of the complete conditional proposal, with particular attention to the owned-erasure premise and charged high readout. A possible later search will use a disjoint component and will exclude roles 449/485/1206/1477; no new candidate or κ gain is claimed here.

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