Skip to content

Lower PR189 endpoint frames plus finite leaf bootstrap: conditional κ = 0.0006679123 - #198

Closed
sennemmi wants to merge 22 commits into
CrocSwap:mainfrom
sennemmi:codex/bit-frame-descent-102
Closed

sennemmi wants to merge 22 commits into
CrocSwap:mainfrom
sennemmi:codex/bit-frame-descent-102

Conversation

@sennemmi

@sennemmi sennemmi commented Oct 9, 2026 •

Copy link
Copy Markdown

Status

Draft follow-up stacked on #194 head a8c87783785be6fe20c4a8bb2ac4b3eec295a179. The branch includes that base construction, so the PR diff against main contains both the base and this follow-up.

Under the retained conditional interfaces, the exact composed result is:

This is not claimed as the current public record. Open #197 claims the higher 135216063303877/200000000000000000 = 0.000676080316519385 on a different packed PR187 bit supplier. It currently has no reported checks; its description says the new workflow has not run and does not claim a fresh full make verify on current main. #199 is also open with no reported checks. I keep this PR in Draft and do not claim to beat #197.

Construction

A deterministic search selected 102 nonadjacent endpoint-frame lowerings from 2,534 structurally eligible candidates on PR189's physical bit word. Candidate costs use the complete physical chain order: donor operations, recipient gauge splice, then recipient operations. Each selected frame contains the node value span and actual predecessor frames, stays inside both physical successors, and preserves every terminal-sink write. The Boolean operation word is unchanged. The exact assignments, 70 added frame-basis records, histogram delta, and rational cost receipt are in research/paired-cube-twin-local-168/bit/frame-descent.json.

The descended bit supplier has exact coarse saving 167089706787281/250000000000000000. PR184's unchanged pricing gives selected coarse saving C = 1670897/2500000000 and initial effective saving a_0 = 6679380665463285737/10^22. The finite depth-2 ordinary-leaf wrapper uses the PR185 recurrence first applied to PR189 by PR199:

  • a_(j+1) = (1-C)C + C a_j.
  • a_1 = 16708962969977344607748096089/25000000000000000000000000000000.
  • a_2 = 41772424988253556235173052470510821833/62500000000000000000000000000000000000000.
  • The checks establish a_0 < a_1 < a_2 < C < 1-a_2; the assembly then rechecks all 47 strict constraints and seven margins.

The complex supplier remains 219037/312500000, so the bit supplier binds. The PR189 physical word is inherited from #189; the finite leaf wrapper is inherited from #185. This composes #199's already-public pricing recurrence with the new frame descent; the recurrence and wrapper are credited, not claimed as new work.

Reproduction and validation

Local lightweight checks passed: exact frame geometry and physical rows, 58 source pins, exact recurrence, source/certificate consistency, PR184 rational assembly, all 47 strict constraints and seven margins, and git diff --check. No full high-cost verifier was run locally.

Cloud verification is running on head 2d6aac706f274812f260792f456c1b9458ce3026:

Reproduce from the repository root:

python3 -m pip install -r research/source-assisted/requirements-round13.txt
python3 -B research/source-assisted-v4/verify.py
python3 -B research/paired-cube-twin-local-168/verify.py
make verify

Scope and attribution

The result remains conditional on PR184's analytic, recursion, precision, weighted-frame, fixed-tape routing, restored-row and all-size semantic interfaces, the PR189 physical word and contracts, and PR185's completed finite ordinary-leaf construction. A globally renumbered scalar transcript of the new complex word and a full Clifford/router replay are not included. This is a conditional finite witness, not an unconditional theorem or a global optimum.

PR184, PR189, PR193/194, PR185, PR199 and PR168-v4 notices and authorship are retained. The 102-frame search and certificate are credited to OpenAI GPT-6 (Codex). The finite wrapper follows PR185 (Rohan Arun, OpenAI Codex assistance); the recurrence/pricing pattern follows PR199 (Maxime Fleury, Codebuff assistance). Apache-2.0 applies; see the package NOTICE files for the full predecessor attribution.

eumemic and others added 20 commits October 9, 2026 00:16
…s: kappa = 5.878747e-4

Complex: PR144's paired cubes at p = 11 with a searched triple module and an
annealed pair module; the face-2 and edge-02 outputs of each port are merged
into one signed node per (b0, b2); 7,880 operation frames moved by physical
descent and 2,970 gauged roles placed in dead donors' slots with deadline
reads (AC 5885669/10^10). Bit: the paired-cube bit word with an annealed pair
module (COARSE 5936323/10^10). Phase stop 1e-9. The complex side binds and the
47-constraint assembly gives kappa = 5878747/10^10.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…the balanced prefix on paired cubes: kappa = 6.096379e-4

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
… on paired cubes: kappa = 6.457084e-4

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…es: kappa = 6.489120e-4

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…derived terminal sinks: kappa = 6.558894e-4

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…on PR168 v4: kappa = 6.602166e-4

Complex: the direction-0 shared-edge local circuit of PR181 (13,042 virtual
roles, 44 terminal sinks). Bit: 556 connected-component endpoint frame
changes on PR168 v4's physical bit layer and 34 PR166 terminal sinks on
face0 roots, rechecked on the overridden word with exact prime witnesses
for all 24,577 used frames. Both suppliers and the paid balanced
composition are independently certified above PR168 at
4a3c769: kappa = 330108276030861/(5*10^17).
The bit branch still binds. The exact four-file PR175 overlay is retained.

Prepared by Chafik Boukhalfa with substantial Anthropic Claude and OpenAI
Codex assistance.
PR184 edits README.md, Makefile, NOTICE, SOURCES.json, the verify
workflow, docs/reproducibility.md and the joint-dual source manifest.
PR182 and PR185 pin those files, so their manifests go stale after the
merge. This commit updates only the stale hash values:

- certificates/joint-dual-kappa.json: source_manifest_sha256
- certificates/joint-dual-validation.json: certificate_sha256
- research/paired-cube-local-bit-168/SOURCE.json: nine shared files
- research/bit-leaf-bootstrap182/SOURCE.json: regenerated with its own
  pin_sources.py

No certificate value, construction file or package file changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Apply PR184's source-parity local word and source-assisted frame flow
to PR168 v4's query modules and physical layer instead of the older
fd25adb modules. The complex supplier now has R = 10,824, W = 13,464,
1,412 source-controlled erasures and 707 kernel reuses. Its saving is
3315351/(5*10^9) = 6.630702e-4, about 0.91% above PR184's complex
saving and 0.113% above PR186's.

PR184's bit supplier is larger, so the complex side binds. PR184's
unchanged finite bridge and 47-constraint assembly give
kappa = 6626307/10^10 = 6.626307e-4, about 0.1126% above PR186
(6.61885549259598e-4).

verify.py regenerates the v4 cache, the new local word, the flow, the
exact lift, the contract checks and the assembly, and compares the
result with certificate.json. A workflow runs it in CI.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR184 pairs each live gauge with a donor by a plain maximum matching.
source_aligned_local_v4.py --avoid-parity-erasure instead takes a
minimum-weight full matching that avoids pairs PR184's parity
purification would erase. With the frozen result in
data/physical-pairs.json the flow has 8,116 dependent outgoing values
instead of 10,364 and needs 1,412 fewer dirty births: R = 9,412,
W = 12,052. PR184's exact lift and contract checks pass unchanged.

The complex saving rises to 219037/312500000 = 7.009184e-4. PR184's
effective bit saving (6.6523e-4) now binds, and the unchanged assembly
gives kappa = 103873/156250000 = 6.647872e-4: +0.438% over PR186 and
+0.325% over PR191.

verify.py now records the binding supplier and rounds binary-float
diagnostics in the certificate, so last-digit libm differences between
platforms cannot change the comparison.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… kappa = 6.617552e-4

Complex: the direction-0 shared-edge local circuit of PR181 (13,042 virtual
roles, 44 terminal sinks). Bit: assemble A[2,0] and A[2,1] from the two
face diagonals (19,348 virtual roles), with a fresh carrier matching and
gauge selection by the retained generator rules, our own physical layer
(6,138 descended frames, 1,760 compensated handoffs) and 34 PR166 terminal
sinks on face0 roots. Both suppliers are independently certified above
PR168 at 4a3c769 with the paid balanced
composition: kappa = 330877589979561/(5*10^17); the complex branch binds.
The exact four-file PR175 overlay is retained.

Prepared by Chafik Boukhalfa with substantial Anthropic Claude and OpenAI
Codex assistance.
…ppa = 6.674324e-4

The complex saving of the v4 source-assisted word (7.009184e-4) exceeds
every published bit supplier, so the bit side binds. assemble.py prices
PR189's certified face-diagonal bit profile (R = 17,554, W = 21,074)
with PR184's unchanged select() and assemble(): coarse 6682989/10^10,
effective 6.678782e-4, 47 strict constraints and seven margins.

kappa = 1668581/2500000000 = 6.674324e-4: +0.398% over PR193 and
+0.838% over PR186.

PR189's manifest pins README.md, Makefile, NOTICE, SOURCES.json, the
verify workflow, docs/reproducibility.md and the joint-dual files; this
commit updates those nine stale hash values. PR189's verify.py passes on
this branch, and the workflow now runs it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@sennemmi sennemmi changed the title Lower PR189 endpoint frames: conditional κ = 0.0006674922 (stacked on #194) Lower PR189 endpoint frames plus finite leaf bootstrap: conditional κ = 0.0006679123 Oct 9, 2026
maxime-fleury added a commit to maxime-fleury/integer-mult-bounds that referenced this pull request Oct 9, 2026
…sition

This PR's kappa is superseded on CrocSwap#194's inputs by CrocSwap#198 (same recurrence on a
descended CrocSwap#189 word) and CrocSwap#197 (a stronger bit word), so state the standing
plainly instead of claiming a record. What remains is the reproduction harness,
the two maximality proofs, and a supplier-agnostic pricing path: applied to
CrocSwap#200's deferred composition with CrocSwap#193's complex it yields kappa =
1693287/2500000000 = 6.773148e-4, +0.1826% over CrocSwap#197's standing claim, and the
depth-2/3 rows sit 3.4e-11 below that word's own C/(1+C) ceiling, so pricing on
it is exhausted. verify.py now re-prices the vendored CrocSwap#200 certificate and
fails if the composition stops beating CrocSwap#197.
romainhedouin added a commit to romainhedouin/integer-mult-bounds that referenced this pull request Oct 9, 2026
The pinned file is byte-identical in CrocSwap#168, CrocSwap#178, CrocSwap#182, CrocSwap#189 and CrocSwap#198; CrocSwap#168 is the lowest-numbered PR found
carrying it. The three data files are also identical in CrocSwap#168. No code change.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@sennemmi

Copy link
Copy Markdown
Author

Archiving this Draft as part of the research closeout. It explored lowering the PR #189 endpoint frames with a finite-leaf bootstrap on #194. At head a22b271a06c6e61b7dcd2ec4859e6ccd532b1f6a, full CI failed, including the joint-dual source pin assertion and recycled-bit replay error (complex_certificate missing phys); the terminal workflow also failed. Both inherited dependencies remain open. The branch, commits and certificates are retained; no further maintenance is planned. See the consolidated closeout in #261.

@sennemmi sennemmi closed this Oct 10, 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.

6 participants