Skip to content

Bit c = 0.001733427039589844: two A-section transports on #555's punctured R439 word (+0.0216%, κ unchanged) - #589

Merged
Mister-Meeseeks merged 2 commits into
CrocSwap:mainfrom
DreamingOfClouds:research/e8-bit-p111-tune
Oct 12, 2026
Merged

Mister-Meeseeks merged 2 commits into
CrocSwap:mainfrom
DreamingOfClouds:research/e8-bit-p111-tune

Conversation

@DreamingOfClouds

@DreamingOfClouds DreamingOfClouds commented Oct 12, 2026 •

Copy link
Copy Markdown
Contributor

Bit c = 0.001733427039589844234749 (+0.0216% over #555), κ unchanged.

Two helper additions of #555's punctured R439 bit word (v = 111) are moved out of their D4 mini-node gates and rerun at other frames. Nothing else changes: the same registers, scatter, retained totals (nine COPY ranks of 7, decoder constant 63), frozen hosting selections and pipelines.

#555 this PR change
bit c, nine stages 0.001733052823330162099157 0.001733427039589844234749 +0.0216%
bit c, eleven stages 0.001710608806098012673516 0.001710951900777583975622 +0.0201%
κ with #548's complex b 0.001590984155135328 0.001590984155135328 none

κ stays complex-bound for every cited complex supplier (#548, #544, #535, #531), so this is bit headroom only.

What changed

  • hoist: in A[1993] (frame 509, rank 4), the summand helper 222 += helper 248 becomes its own gate at rank-1 frame 79, placed just before that gate.
  • sink: in A[1994] (frame 509), the summand helper 191 += helper 247 becomes its own gate at the rank-7 pseudo-total frame 517, placed after A[2028].

Both summands commute with every addition they cross, and the dirty-register endpoint is unchanged on every column. The local paid profile moves by +2e1 − e3 − 3e4 + e6 + e7, and rank mass stays at 5,129.

The mechanisms come from #541 and #548 (Dugongue) and #542 (rohanarun) on the R480 word. Their move lists are R480 gate ids and don't apply to #555's word, which has no two-x-source additions and no zero-response gates. tune/tune.py therefore re-runs the mechanisms as searches. It reproduces #541's and #548's R480 profiles exactly. On #555's word it finds exactly these two moves, and pruning, Steinberg 3→2 and least-frame steps find nothing further.

Checks

bash research/e8-bit-p111/tune/verify.sh /absolute/new/output   # tuned word
bash research/e8-bit-p111/verify.sh /absolute/new/output        # #555's own check, unchanged

Notes

Credit: #555 (eumemic) is the base, built on #543 (eumemic), #545 and #510 (DreamingOfClouds). The mechanisms are #541 and #548 (Dugongue) and #542 (rohanarun).

DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code.

🤖 Generated with Claude Code

eumemic and others added 2 commits October 11, 2026 21:50
… R420 skeleton, c 0.001733052823

Stacks CrocSwap#545's punctured-star unit (DreamingOfClouds) on CrocSwap#543's R420 L3 skeleton: nine E8 lines
(P = 10, 15, 51, 53, 56, 66, 72, 89, 111) are neither sources nor targets, and the nine retained totals
are pseudo-targets at frame rank 7 (decoder constant 63). Because L3 delivers the Q-part in phase B, each
pseudo-target gets it through a free keeper rewritten in a phase-A mini-node at the complementary block
frame (with extra carriers for missing lines); R = 439, v = 111. Priced by CrocSwap#510's nine-stage chain,
parameterised for v <= 120 (code/adapt9p.py): c = 1733052823330162099157/10^24, +4.79% over CrocSwap#543's word;
by CrocSwap#493's eleven-stage chain as parameterised by CrocSwap#545's adapt.py: c = 0.001710608806098012673516. Every
cited complex supplier binds: kappa = 0.001590984155135328 with CrocSwap#548's b. Controls reproduce CrocSwap#510's R480
outputs byte for byte through the parameterised files, and CrocSwap#545's c and program hash through CrocSwap#545's files.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…nsports: nine-stage c 0.001733427039589844234749

Hoist one summand of a D4 mini-node gate to its rank-1 frame and sink another to
the rank-7 pseudo-total frame (tune/moves.json, tune/apply.py), found by running
the mechanisms of PR CrocSwap#541, CrocSwap#542 and CrocSwap#548 as generic searches on PR CrocSwap#555's word.
Nine-stage c +0.0216% (eleven-stage 0.001710951900777583975622, +0.0201%); kappa
unchanged (complex-bound). tune/verify.sh replays the stacked word through
PR CrocSwap#555's chains; PR CrocSwap#555's verify.sh is unchanged. MANIFEST.sha256 lists tune/.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@DreamingOfClouds
DreamingOfClouds marked this pull request as ready for review October 12, 2026 14:11
@Mister-Meeseeks
Mister-Meeseeks merged commit b3eb785 into CrocSwap:main Oct 12, 2026
62 checks passed
Mister-Meeseeks added a commit that referenced this pull request Oct 13, 2026
…ling contributions

Select conditional kappa 0.001661634653669567 from Jeremy Kothe's #799 on Chafik's #741 and the #589/#555 bit supplier. Preserve the previous checkpoint and cumulative attribution. Complete #795, #777, six #681 component reviews, and five exact tooling reviews. Integrate #806 as a scoped hash-substitution diagnostic with an adversarial counterexample to timing-only attribution. Retain source-pinned evidence and separate publication/CI state.
Mister-Meeseeks pushed a commit that referenced this pull request Oct 16, 2026
Confirms kappa 0.001660666354374641 (#699 at 03e8d7c) and
0.001660339298410333 (#685 at 5fc2a0c) from fresh constructions,
a no-import checker and a standalone #589 bit replay.

Audit by Rohan Guliani (@Rohansguliani) with substantial AI assistance
through Mux. Apache-2.0.
Mister-Meeseeks pushed a commit that referenced this pull request Oct 17, 2026
…ew kappa)

Answers PR #3004's third open question for one full-size word. Row 0
[4,5,6,7,8] of the selected PR #589 bit word is extended four ways and
each word runs through the unchanged symbolic checker, record exporter,
both entry modes of PR #2992's simple-scatter-v1 guard and the native
legality and five-stage column checkers. Appending [4,4] triples an
existing slot: the symbolic and both native checkers accept it, and all
three exporter artifacts and both native receipts equal the baseline
byte for byte (27,234 records, 24,305 ADDs, rank mass 5129, nine
COPY/ERASE pairs). Appending [0,0,0] triples an absent slot: symbolic
and exported actions agree, but both miss the required endpoint at
output 111 (98 columns; dirty input 472 is a one-bit witness), and the
full symbolic and native column checkers reject it. PR #2977's paired
[0,0] reproduces symbolic acceptance with native rejection. The guard
rejects all three repeated rows before creating output, as its stricter
schema intends. Native legality accepts all four words, so legality
does not certify the endpoint. Occurrence counts are not emitted ADD
counts: the accepted triple has 529 occurrences and 527 scatter ADDs.

This is a finite, author-checked audit with a separately written
harness. Native builds use the pinned checked 128-bit cpp_int stand-in.
No schema decision, supplier adoption, cold ancestry, system-Boost
build, physical hosting or kappa change is claimed.

Adds research/scatter-odd-native-audit/ only.

Submitted by @commdata2338 with substantial AI assistance (OpenAI Codex
research and implementation; packaged with Claude Code).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X86sCSXdoDd61SDFfC1PKG
Mister-Meeseeks pushed a commit that referenced this pull request Oct 17, 2026
… native COPY counts (no new kappa)

Runs PR #3128's four full-size fixtures on the selected PR #589 bit word
(selected baseline; balanced alias, which appends helper 480 as retained
slot 9 and [8,8,9,9] to scatter row 64; unpaired even slot; distinct
helpers) and a new unused-slot control through the unchanged symbolic
checker, record exporter, simple-scatter-v1 guard and native legality
and five-stage column checkers. Compiled for ten COPY/ERASE pairs, the
balanced alias passes legality and all 883 formal columns; the selected
nine-copy builds reject it at their count checks, and the guard rejects
its repeated row before any output is written. Its complete local
invoice keeps the extra rank-seven COPY/ERASE pair and two scatter ADDs:
records 27,234 to 27,238, rank mass 5,129 to 5,136, five-stage ADDs
121,525 to 121,535 and fresh copies 45 to 50. The prefix maxima (49,347
forward, 1,113,070 reverse) and the 93-bit payload bound are unchanged.

An unused tenth alias with unchanged simple rows passes both guard entry
points and the ten-copy native checks, while the nine-copy builds reject
it, so the guard enforces row support, not retained-list injectivity or
the selected copy count. The unpaired and changed-helper controls keep
symbolic acceptance and native legality but fail the formal columns at
output 175, where dirty input 480 is a witness. Three native mutants
separate pricing, lifecycle and endpoint checks.

Finite audit evidence only: NCOPY=10 is an explicitly changed contract,
so no ten-total supplier is admitted and no frontier or kappa changes.
Cold replay, bank/projector/hosting admission, nine-stage allocation,
all-size compilation, precision/tape and analytic transfer stay open.

Adds research/scatter-alias-native-audit/ only.

Submitted by @commdata2338 with substantial AI assistance (OpenAI Codex
research and implementation; packaged with Claude Code).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X86sCSXdoDd61SDFfC1PKG
DaysSky pushed a commit to DaysSky/integer-mult-bounds that referenced this pull request Oct 19, 2026
Adds BREAKTHROUGH-MAP.md, audit_breakthrough.py and audit_anatomy.py to the 2026-10-19 II19 audit.

- Framing: the selected kappa is complex-bound (current-status: the bit supplier remains CrocSwap#589 on
  CrocSwap#555), so a bit-side II19 win of any size does not move kappa by itself.
- At fixed (v, h, cst) the profile mass and deficit are fixed; the word only chooses the rank split.
  Role anatomy (replay): in the r434 the 434 helpers carry 74% of all climbs (4-6 rank-1/2 climbs
  each); targets are already at 2 climbs.
- Model requirement for +25% on II19-112: R/v ~ 2 with <= 3 climbs per register (or R/v ~ 3 with 2).
- Why E8-type middles resist it: cell bottleneck (R >= 3.25v in class), reach decay (63 -> 36 at the
  first frame step, so relays are forced), ~2 private wires per target (perp E7 needs 3 hyperplane
  slices). Idealized 3-climb middle: II19-112 +3.6% at R/v 3.25 (E8 +20.1%).
- II19 trails its h=8 parent by 16-17 points at equal middle quality.
- Scope lever measured on actual words: idle slots without bridge split +6.2% (r434), +5.2 points
  (II19-112); it lifts both.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012xTYY3auFQxtMfUudXG8Zk
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.

3 participants