Repository navigation
Seven new results, referee fixes and validator hardening - #72
Merged
Merged
Conversation
The Delft trial digest, the published Hensen et al. values and the SHA-512 of the committed NIST pulse list are now code constants, so an edited data file cannot authenticate itself. The beacon line reports hash-chain consistency, not authenticity. Evidence paths may not escape the repository root. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Many valid properties checked internal arithmetic but not the instance, so hand-built certificates validated: zero TV-ball and confidence radii, forged chromatic lower bounds, suboptimal robust codes, dropped candidates, swapped graph enumerations, best-case adversarial paths, forged frontiers and oracle values, trimmed drift grids, truncated vertex lists with spurious rows, unbound polyhedral duals and Farkas certificates, empty network decoders, non-Huffman prefix codes, values attached to the wrong channel, and Boolean query values that were never recomputed. Affected validators now rerun the deterministic exact builder on the stored instance under the stored caps and require field-by-field equality (simtheory.rebuild.matches_rebuild), or recompute the missing quantity directly. tests/test_validator_forgeries.py reproduces every forgery from the audit as a regression test. Inequality-only receipts are documented as arithmetic checks, and the Farkas normalization comparison is fixed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two-sided pathwise KT sandwich, the Clarke-Barron constant with a non-asymptotic sandwich, the NIST beacon power frontier and Le Cam converse, BSC coarsening (extends ST-M83), and the persistent-latent KL cap, saturation identity and gap bound. Standard results (data processing, chain rule, Clarke-Barron, Le Cam two-point, KT) are labelled as such; cross-references re-anchored to ST-T271, ST-F52 and ST-M83; genuine counterexamples to the sketch's zero-gap 'iff' added; beacon n and k re-derived from the pinned pulse file; the second-order term stays an open problem (ST-O02). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…-T281 to ST-F56) Exact certificates for signalling, measurement-dependence and detection frontiers (Pironio, Puetz et al. and Garg-Mermin credited; the latter typed a model result), the uniform-setting statistical strength, and the Kelly/Wald bounds, all marked standard where they are. The Delft finite check is a negative result: no constant-bet mixture on the win indicator reaches 1/alpha at any prefix (ceiling 7.78). 'Preregistered' is removed, the CHSH correlator definition is fixed, and data load through the pinned real-data loader. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Dithering boundary, commensurable-schedule non-identifiability, the Diophantine Theta(1/D^2) separation for 0 and pi/4, and known-visibility rates. Qualifiers restored (4 | D, D even, declared D, exact cosines, fixed-count achievability, restricted-cosine converse), float64 treated as an illustration only, prior art credited, and the false 'every D divisible by 4' sketch recorded as a negative result. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Arbitrary-center area converse, the sharp pi/32 CHSH constant, the pi^3 ceiling (Lutwak, standard), exact tiling-schedule asymptotics and the proved optimality of the regular three-direction schedule. The covering bound is credited to L. Fejes Toth (centrally symmetric), not Fary; the in-family constant is identified as classical and the open problem narrowed to arbitrary emissions; the trine product schedule shows the optimum is compatible with a CHSH violation. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Stochastic reparametrization of the two-view fiber, the sharp k = 2 identified interval, the weight floor, and product-gauge rigidity, with PR #64, Henry-Kitamura-Salanie, Hall-Zhou and Huang-Sidiropoulos-Swami credited. Adds a proved necessary condition (pairwise-incomparable row supports) and an exact no-anchor point-identified example, replacing the 'anchors are necessary' overclaim; k = 3 grid values are label-free and each grid point is checked to reproduce T. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Exact deficiency of a visibility-contracted Bell simulator, shown to hold against arbitrary joint-record kernels, the worst-setting value for unequal magnitudes, risk transfer (standard), and the setting-wise deterministic gap. Setting-preserving and central-kernel qualifiers are explicit; a verified setting-relabeling map (23/160 < 3/14) bounds the scope; Le Cam and Torgersen credited; the optional LP line is gone, so the receipt is exact. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Exact b-bit value curves, the dyadic dichotomy and the two-sided Theta(2^-b) gap without nondegeneracy, the Knuth-Yao price of the optimal set (Knuth-Yao reproved and credited), and coin alphabets with L >= 2. Prior work on k-uniform strategies credited, the reduction-invariance lemma added, and the committed instances cross-checked against the repository solvers, including an ST-M27 radius sweep. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
alpha(K) is Gutman-Pena's TV-norm facial distance (Prop. 1) and the zero-sum restriction of the Kleinberg-Sandler l1 condition number; Pena and Rodriguez are scoped to the Euclidean norm. Standard results are marked in nonclaims, ST-T251 is an exactly computed lower certificate, the 'phase boundary' heading is retitled, and the math audit records the second audit. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-gauge remark after Theorem 5 with an exact counterexample (abstract, limitations and P1-T7 updated; Kruskal cited), section retitled 'Persistent latents and the sampling hierarchy', SSA plus SIA item attributed, Dorr and Arntzenius cited, Theorems 1 and 3 typed standard, bibliography case and arXiv fixes, ledger crosswalk to repository claims with a test, texttt, shorter abstract, qedhere. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Push runs only on main and the concurrency group is keyed by event. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
465 claims in 60 manifests; new lanes in the map; the Delft anytime-valid negative result; pinned receipts and validator rebinding in the verify description. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Integrates seven refereed research results as native lanes, applies every referee fix, hardens the certificate validators that an audit showed could be forged, applies the second audit's attribution fixes to the mixture-conditioning lanes (#61, #62) and the paper (#69), and stops push and pull_request CI runs from cancelling each other.
Claims: 397 in 53 manifests -> 465 in 60 manifests (304 theorems, 93 model results, 60 finite checks, 8 open problems).
uv run python -m simtheory verify(full, not--quick) passes locally.New lanes and claim IDs
docs/evidence-scaling-dichotomy.mddocs/bell-escape-route-frontiers.mddocs/finite-precision-renderer-identifiability.mddocs/bell-disk-renderer-memory.mddocs/two-view-orthogonal-fiber.mddocs/le-cam-deficiency-bell-simulator.mddocs/random-bit-price-shared-randomness.mdEach lane has a standard-library module under
src/simtheory/, tests undertests/, a committed receipt underreceipts/(checked byte for byte by a test), and a manifest with assumptions, nonclaims, repo-relative evidence and test citations. No result was excluded; nothing was marked unsound by its referees.Referee fixes (highlights)
Validator hardening
The validator audit showed that many
validproperties checked arithmetic but not the instance, so hand-built certificates validated (zero radii, forged chromatic lower bounds, dropped candidates, swapped enumerations, best-case adversarial paths, forged frontiers and oracles, truncated vertex lists, unbound duals, empty decoders, non-Huffman codes, values attached to the wrong channel, unrecomputed Boolean query values). Affected validators now rebind to their stored instance, mostly by rerunning the deterministic exact builder under the stored caps (simtheory/rebuild.py), andtests/test_validator_forgeries.pyreproduces every forgery as a regression test. Real-data receipts pin the Delft trial digest, the published values and the beacon pulse-list SHA-512 in code; evidence paths may not escape the repository. Inequality-only receipts are documented as arithmetic-only.Audit fixes for #61, #62, #69
2 rho_obstransfer.\texttt, shorter abstract,\qedhere.CI
tests.ymlnow runs on pushes tomainand on pull requests, with the concurrency group keyed by event, so a push run and the pull_request run for the same commit no longer cancel each other.🤖 Generated with Claude Code