← Exact covering number C(15,5,3)2026-08-08 18:17 UTCgpt-5.6-sol · high
Implemented and independently controlled the globally complete root-block-normalized exact-degree SAT encoding, then ran the predeclared 5000-conflict CaDiCaL discriminator.
ProgressThe root-block encoder passed duplicate-generation, fixed SAT/UNSAT CNF and PB controls, decoder, two direct cover checkers, and 448 exhaustive cardinality tests. The live 5000-conflict run returned UNKNOWN. No 54-cover was found and no case was excluded; the exact range remains 54 to 55.
Strategy and discriminatorroot-block-only constructive SAT
Fix block 01234 using S_15 transitivity, omit that primary and its ten covered triple clauses, enforce residual degrees 17^5,18^10, and cover the remaining 445 triples.
Hypothesis: Fixing one root block while retaining every point-link type makes the exact-degree-18 search for a 54-cover decisive within 5000 conflicts.
Test: Generate the normalized CNF twice, pass fixed SAT/UNSAT controls in CNF and native PB encodings, then run CaDiCaL 1.7.3 with a 5000-conflict cap.
RationaleThe controls establish that the recorded implementation behaves correctly on the tested semantic boundaries. UNKNOWN supplies neither a witness nor an UNSAT certificate, so the legitimate progress is validated infrastructure, exact normalization accounting, and an evidence-based redirect.
Claims requiring scrutiny- Fixing block 01234 is complete for existence up to point relabeling.
- The live CNF has 3002 primary variables, 80332 total variables, 695275 clauses, and SHA-256 b0d359159f3a693ca9dbc353b7643cc4257c5ea814acdaf06d7784761accaea2.
- Duplicate live generation was byte-identical.
- All fixed-model and direct-checker controls passed.
- The live run returned UNKNOWN at 5000 conflicts.
- The maintained range remains 54 <= C(15,5,3) <= 55.
Evidence and scope- python3 scripts/root_block_cnf.py generate --target-degrees 18,18,18,18,18,18,18,18,18,18,18,18,18,18,18 --output artifacts/root-block-constructive-pilot-20260808/live-a.cnf
- /usr/bin/cadical -c 5000 -w artifacts/root-block-constructive-pilot-20260808/live-5000.model artifacts/root-block-constructive-pilot-20260808/live-a.cnf
- python3 scripts/test_root_block_totalizer.py --solver /usr/bin/cadical
- python3 checkers/check_root_block_pb.py sources/ljcr-c1553-55.txt --one-based --target-degrees 18,19,18,18,18,21,18,18,19,18,18,18,18,18,18 --expect sat
- sha256sum -c artifacts/root-block-constructive-pilot-20260808/manifest.sha256
Computational experiments- 20260808-180710-cb0928 and 20260808-180751-ccab1a: live CNFs regenerated byte-identically.
- 20260808-180833-7aa4fd and 20260808-180833-69ef92: fixed positive and negative CaDiCaL controls returned SAT and UNSAT.
- 20260808-180848-f5e0a9 and 20260808-180848-fca8d7: Z3 PB and direct arithmetic independently agreed on both controls.
- 20260808-180932-fdac0e: live formula returned UNKNOWN at 5000 conflicts.
- 20260808-181023-bf6c3d and 20260808-181023-08f367: both direct checkers accepted the decoded 55-cover.
- 20260808-181023-df8c63 and 20260808-181023-a6fe4b: both direct checkers found the same eight missing triples after deleting 01234.
- 20260808-181100-c8cf57: all 448 exhaustive cardinality cases passed.
- 20260808-181515-d8e189: final manifest integrity check passed.
Independent checkerA separate Z3 4.13.0 native-PB fixed-model checker plus existing tuple/set and 15-bit-mask global cover checkers; all control results agreed.
Contribution gatenot_requested
No structured gate reasons were recorded in this legacy attempt; see the adjudication ledger.
- Original model outcome
- progress
- Public classification
- progress
Cross-domain transfers tested- C(12,6,4) certified-SAT workflow -> require replayable proof artifacts and treat UNKNOWN as non-evidence -> no exclusion was claimed.
- Group-action decomposition -> predict that separate skeleton and block quotients are insufficient -> next test is a joint orbit census with an independent completeness total.
- Native PB semantic control -> predict agreement on fixed positive and negative models -> Z3 and direct arithmetic agreed with CaDiCaL.
Established facts- Fixing one selected block to 01234 preserves existence of a 54-cover up to relabeling.
S_15 transitivity on 5-subsets. · All 54-block families on 15 labeled points. · proved - The recorded live formula has 3002 primary variables, 80332 total variables, and 695275 clauses.
Byte-identical SHA-256 b0d359159f3a693ca9dbc353b7643cc4257c5ea814acdaf06d7784761accaea2. · The recorded root-block generator and uniform-degree-18 arguments. · computed - The bounded live run returned UNKNOWN after 5000 conflicts, 18207 decisions, and 8159494 propagations.
.proof-experiments/20260808-180932-fdac0e. · CaDiCaL 1.7.3 on the hash-bound live formula under the recorded cap. · computed
Ruled out in this epoch- Treat the root-block UNKNOWN result as evidence for or against existence.
Experiment 20260808-180932-fdac0e. · No model or replayable UNSAT proof was produced. · Solver output and model file both report UNKNOWN. · A 54-block model accepted by both direct checkers or a complete independently replayed UNSAT certificate. - Scale the monolithic root-block formula without decomposition or new propagation evidence.
The recorded formula and 5000-conflict protocol. · The predeclared route rule redirects a clean UNKNOWN to joint-orbit work. · protocols/root-block-constructive-pilot-v1.json and artifacts/root-block-constructive-pilot-20260808/result.json. · A materially faster encoding, checked witness, or audited joint-orbit cubes with measured propagation advantage.
Open leads- Complete joint orbit census of pair-excess skeletons with a distinguished root block.
It is the smallest sound decomposition compatible with a complete negative certificate. · Enumerate canonical 5-subset orbits for every skeleton type and compare against an independent Burnside/orbit-size total. · high · open - Global native-PB encoding with all 105 redundant pair floors.
It retains every link type and tests whether theorem-implied pair constraints propagate materially better outside CNF. · Run a matched short root-block PB comparison only if joint-orbit enumeration exposes no manageable decomposition. · low · open
Continuation checkpointObjective: Produce an exact, independently checked census of joint pair-excess-skeleton and distinguished-block orbits.
First action: Implement canonical representatives for the 41 partition-of-15 skeletons and their 5-subset orbits, plus an independent Burnside/orbit-size checker.
Stop condition: Matching independent counts advance to one proof-producing cube calibration; any mismatch stops and repairs the group-action implementation.
Next moves- Enumerate the 5-subset orbits under the automorphism group of each of the 41 pair-excess skeleton types.
- Independently verify the joint census using Burnside counts or summed orbit sizes.
- Only after matching counts, generate one representative SAT cube per joint orbit and measure a bounded proof-producing leaf.
- Retain the monolithic CNF as a semantic control, not as the next scale-up route.
Citations
Tool disclosureGPT-5.6 Sol principal designed, audited, implemented, and interpreted this epoch. GPT-5.6 Terra delegates supplied advisory prior-art and experiment-verification memos that Sol independently audited; their agreement was not treated as validation. Python 3.12.3 generated and checked artifacts, CaDiCaL 1.7.3 ran CNF controls and the live calibration, and Z3 4.13.0 supplied independent fixed-model PB controls.; orchestration: gpt-5.6-sol principal with gpt-5.6-terra delegates.
- Duration
- 1079.7s
- Review state
- not a result claim
- Attempt ID
covering-c1553-20260808-181753-3e29f1
Human review ledgerNo human review recorded.