- 2026-11-21T05:20:08Z start; read fam3.py, cyc2.py (copies identical to outputs dir).
- 2026-11-21T05:25:32Z fetched arXiv 2603.08822v1 via WebFetch (curl to arxiv blocked by proxy 403). Note: M-Z ref [11] = Hackmann-Kemnitz, DM 286 (2004) 89-93, not Zhu survey. Problem 6 literally allows chi'c = d+1 (K{d+1}, d even, trivially satisfies it); M-Z themselves note K_{d+1} and C_7(1,2) are 4-connected with chi'_c = 5. Watkins 1970 atom theorem confirmed via secondary source (Morris, SpecialFamily.pdf). Hamidoune 1977 CRAS 284 A1253-1256 bib data via arXiv math/0603504 bibliography.
- 2026-11-21T05:27:18Z verify.py written (stdlib only); passes g=2..30 colouring/partition/alpha' checks, kappa=4g for g=2..8 (all pairs for g<=4) and kappa=18 for C_25(...); max-flow cross-checked vs networkx on 7 circulants. Hand-checked window conditions for Delta=18 example (9 translate conditions, list in PROOF.md). Self-contained proof of 'distinct atoms disjoint' (3x3 grid/submodularity) worked out.
- 2026-11-21T05:29:30Z PROOF.md written to outputs (Part 1 done). Self-contained Lemma 1 via averaging over unit arcs (no need for Vince/Zhu).
- 2026-11-21T05:30:26Z cycsat.py (pysat/cadical, cyclic ansatz, cardinality via seqcounter) reproduces q=3 g=2,3 and q=4 r=3 g=3 (8.5s); q=4 r=3 g=1,2 UNSAT. Started sweep1.sh (q=4..7) -> sweep1.out
- 2026-11-21T05:40:14Z server restart; Part 1 files intact in outputs; sweep1.sh (cycsat 4 3 4) survived and still running. Observed pysat timer-interrupt apparently ineffective; relying on outer timeout.
- 2026-11-21T05:40:22Z killed sweep1 (4 3 4 = q4,r3,g4 unresolved after 11 min). Started sweep2.sh (q=5..8 small g, 300s each) -> sweep2.out
- 2026-11-21T05:52:44Z sweep2: q=5 r=3,4 g=1,2; q=6 r=5 g=1; q=7 r=4,5,6 g=1 all UNSAT (cyclic ansatz); 5 3 3 timeout 300s. dirsat 13 29 3 (direct, Delta=9) no answer in 200s. Insertion lemma (cycins.py): q=3 r=2 g=2,3 D=1,2,3 UNSAT (encoding sanity-checked: SAT without non-edge constraint); g=4 timeout 120s.
- 2026-11-21T06:02:00Z K_{2m+1} - (near-perfect matching + edge), single-edge classes q=m: UNSAT m=3 (0.1s), m=4 (21s); m=5 timeout 120s. STRUCTURAL LEMMA found: in a perfect (p,q)-colouring with hole x_c of window c, V(class c+q) = V(class c) + x_c - x_{c+1} (single swap or equal). In the cyclic ansatz colour c+q <-> a+1, so W_{rho+1} = swap(W_rho) and W_0+1 = swap(W_{r-1}); hence |(W_0+1)_0| <= r, i.e. each base vertex set is a union of <= r cyclic intervals. Adding these as redundant SAT constraints.
- 2026-11-21T06:02:17Z swap constraints added to cycsat.py: q4r3g3 8.5s->0.1s. Started sweep3.py (q=4..9, n<=90, 300s each) -> sweep3.out
- 2026-11-21T06:15:46Z cycins2.py (generalised insertion, free inserted vertex sets): q3r2 D=1: g=2,3,4 UNSAT; g=5 SAT -> Delta=21 non-regular graph n=31 chi'_c=65/3 (verified by assemble_and_verify); g=5 D=2,3 UNSAT (spacing 3); g=6 D=1 SAT (Delta=25, 77/3), D=2 SAT (Delta=26, 80/3), D=3 timeout 400s. Observed in g=6 D=2: base S avoids length 1; blocks are length-1 pairs + 2 long edges; base W0={0,1}u[2g+3,4g], W1={1,2}u[2g+3,4g]. Results in results/ins_q3.txt
- 2026-11-21T06:17:27Z HAND DESIGN (b): base W0={0,1}u[2g+3,4g], W1={1,2}u[2g+3,4g] satisfy the 4 window conditions; copy-insertion of windows a=0 (colours 0,1,2) and a=2 (colours 6,7,8) with all-length-1 matchings uses length-1 edges {x,x+1}, x in {0}u{odd 3..6g-1} resp. {1}u{even 4..6g}: disjoint. So D=1,2 work for ANY base with these W's and 1 notin S. Remaining: base matchings with 2g distinct lengths != 1 (internal lengths must be exactly {2..2g-3}). Symmetric nested choices fail (checked by hand).
- 2026-11-21T06:22:59Z base matching problem for structured base (W0={0,1}uI, W1={1,2}uI, 1 notin S): SAT for g=6..10 (UNSAT g=3,4,5). D=3 with this base UNSAT g=6,7 (spacing 6). q4 g3 + 1 block (Delta=19) UNSAT. Building certificates g=6..G.
- 2026-11-21T06:38:10Z coordinator asked to wrap up in 30 min; killed sweep3. sweep3 NEW: q=5 r=3 g=3 SAT: C_31(1,2,3,7,8,9,10,11,12), Delta=18, chi'_c=93/5 (denominator 5, simple regular); q=4 r=3 g=4 SAT: C_33(1..5,9..15), Delta=24, 99/4. UNSAT (cyclic ansatz, exhaustive): see sweep3.out. Base certificates for (b) g=6..13 in results/baseB.jsonl; g=14..16 timeout 60s.
- 2026-11-21T06:39:12Z wrote verify_part2.py + data/part2_certificates.json + part2_scripts/ to outputs; both verifiers pass; PROOF.md Section 7 'Further searches' added. Wrap-up.
- 2026-11-21T06:50:40Z applied referee gap fixes (mirror insertion form, edge-disjointness hypothesis, 7.1 wording, Problem 6 'exactly d+1/2' reading, C_31 in Sec 3, Referee pass paragraph, status line); review/ folder with REVIEW.md, check.py, extra.py (extra.py sys.path made relative); all re-run OK; verify.py and verify_part2.py pass.
LOG (circulant upper half)
What Lean checks
Lean proves the actual 4g-regular and 4g-connected graph with real circular index 4g+2/3 for every g ≥ 2, and the complete degree/connectivity/index statements for C25, C31 and C33. The real-colouring lower bound and atom/subgroup connectivity arguments are proved internally. Strict sources, an all-owned audit of 601 declarations/500 theorem declarations and 171 required endpoints, and exact stock-kernel replay passed. Local receipt. The nonregular Section 7 examples, structural/insertion lemmas, SAT exclusions, combined repository gate, and hosted validation remain separate.