Skip to content

P104 toolkit v3: token-settled drop engine, frame-perp duality, fast exact replay, κ(k) caller law (no κ) - #7607

Open
DreamingOfClouds wants to merge 2 commits into
CrocSwap:mainfrom
DreamingOfClouds:research/p104-toolkit-v3
Open

DreamingOfClouds wants to merge 2 commits into
CrocSwap:mainfrom
DreamingOfClouds:research/p104-toolkit-v3

Conversation

@DreamingOfClouds

@DreamingOfClouds DreamingOfClouds commented Nov 15, 2026 •

Copy link
Copy Markdown
Contributor

A reusable P104 toolkit of our stable, verified tools for the hosted-word format of #7546 and later. Every tool has docs, a CLI example and tests with positive and negative controls on vendored data (ToikY's #7546 word, Rohan Arun's #7113 caller, our #7601 edit list). No κ claim.

tool what it does
pricekit.py exact caller + native invoice, certified κ enclosure (#7546: S 1424, κ 0.00364841129479084532481164)
dualword.py frame-perp transpose: an exact involution that preserves the price (dual(#7546) is priced bit-identically; dual(dual) = id up to the one COPY)
tokdrop.py + modp.py token-settled drop engine: clean-mode stay-drop repair, premix settlement rank, exact build. It rebuilds our #7601 word byte-for-byte. On the dual it finds the main-902 drop at chart 595 (K = 0), which the primal engine cannot repair
accx.py + fastreplay.py admissibility against a base. Replay mod 2^61−1 with exact Fraction re-checks: FAIL is exact-confirmed, PASS is randomised-exact (~1 s on #7546)
lossless.py law DUAL: every lossy owned CLEAR in one pass; agrees with exact ranks on 400/400 random programs
nativeplan.py helper-plan extractor; #7546 recost 9,975.447669, 70,714 helper events re-emitted identically
validate.py one command: accx + lossless + the repository's compare_recursive_invoices.py / evaluate_recursive_power.py
kformula.py closed-form caller C(k) (C(17) = the #7113 caller), S(k) = 96k − 208, κ(k) peaks at k = 17, κ* = 1/((2h−1)k*), κ(k) → 0 like 1/ln k
layout_price.py, cap_check.py parametric mask-layout pricer [C-param]; crossing caps and the Floor lemma on the built #7113 B (CALLEROPEN, a DreamingOfClouds lane)

Caveats:

Validator: python3 scripts/research_package.py --package research/p104-toolkit-v3 → PASS_REGISTERED_RESEARCH_PACKAGE (5 commands, ~6.5 min locally; numpy==2.5.4).

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

🤖 Generated with Claude Code

…exact replay, kappa(k) caller law (no kappa)

Reusable tools for the hosted CrocSwap#7546 word format, each with docs, a CLI example and tests
(positive and negative controls) on vendored CrocSwap#7546 / CrocSwap#7113 data: pricekit, dualword,
tokdrop (+modp), accx (+fastreplay), lossless, nativeplan, validate, kformula,
layout_price, cap_check. No kappa claim.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@DreamingOfClouds

Copy link
Copy Markdown
Contributor Author

CI note: the only failure in the plan job is test_current_ci_section_matches_selected_record (1 of 760 tests). It also fails on pristine main e545235, because main's own docs/ci-verification.md is stale (the same failure is explained on #7601 and #7606). This PR adds only research/p104-toolkit-v3/ and touches no docs or tests. Locally, python3.12 scripts/research_package.py --package research/p104-toolkit-v3 (numpy==2.5.4) runs all five test commands in about 6.5 min and prints PASS_REGISTERED_RESEARCH_PACKAGE (outcome=completed). The replay job was skipped because plan failed; I will rebase once main is fixed.

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

…rding only)

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