Numaro covering-design artifacts
Report NUMARO-2026-006, July 3, 2026

Run the offline audit with Python 3 (standard library only):
    python3 audit.py

The four beats/*.json files contain the complete saved block systems. Each has
v, k, t, ljcr (historical comparison value), ours, sources, construction, verify,
and blocks. Points are labelled 1 through v.

The audit checks all blocks for correct size, distinct points, range, uniqueness,
and count, and checks the split between blocks containing point v and the rest
against the recorded R4 source counts. It exhaustively checks all 2,794,155
four-subsets for C(92,7,4).

For t=7/8, the audit does NOT enumerate all required subsets or revalidate the
source coverings. Coverage uses the induction theorem applied to the original
LJCR source coverings and remains conditional on those sources being valid.
Checking the partition and counts alone is not a proof of source coverage.

The recorded comparison values are those used in the July 2026 report. This
package does not assert a current record or an optimality proof, and contains
no new source coverings. The original construction added point v to the first
source's blocks and appended the second source's blocks.

The witness files are unchanged from the autoresearch source snapshot
e80414af564a6f309189ecfb184201e638d0b652. audit.py was added for this website
update; it is an offline audit, not the original campaign verifier.

Current resources:
https://dmgordon.org/covering-designs/
https://github.com/dmgordo/LJCR
https://coveringrepository.com/

Report:
https://numaro.tech/research/covering-designs-la-jolla-2026/

SHA256SUMS lists hashes for all packaged files except itself.
