Use a fixed GF(4) polynomial pair and a binary expansion to reach distance 13 at short blocklength; optimize the generator basis without changing the CSS stabilizer spaces. This is a conventional generalized-bicycle construction; no new family or literature novelty is claimed.
Resource comparison: [[128,20,14]] has maximum check weight 16; the submitted code uses four fewer data qubits and lighter checks at distance 13. Across code types, the stabilizer [[130,29,13]] uses six more data qubits but encodes more logical qubits and has maximum check weight 10 and degree 8. These are resource tradeoffs, not a common performance ranking.
This promotion confirms one fixed period 31 pair, rather than a new parameter sweep. Its earlier optimized basis replaces the 124-check weight 14/degree 14 representation by 104 independent weight 13/degree 12 checks. Two new independent seeds each ran 1,000,000 fast RIS trials on each CSS side on this exact final check basis.
Confirmation ladder for the submitted code, each rung a fresh seed unless marked otherwise, reading the lightest logical found:
| budget (trials/side) | lightest logical | | ---: | ---: | | 1,000,000 (seed 20261017711) | 13 | | 1,000,000 (seed 20261017712) | 13 |
2 fresh-seed rung(s) agree at the best bound, the deepest of them at 1,000,000 trials per side. The submitted witness reproduces it: X: d <= 13 (upper_bound), Z: d <= 13 (upper_bound).
Raw readings, supports, API semantics, matrix digest and timing are in the local RIS ladder record (not committed). The ladder totals 2,000,000 trials per side (4,000,000 across both sides); enumeration, native Pauli, doubling and folding trials are not included. An independent local exhaustive argument also establishes dX=dZ=13, as explained in the local certificate README (not committed). The JSON remains upper_bound; this is not an official server exact-status claim. No circuit or LER is supplied.
degree 12 basis; the latter is the only submitted code. Degree11 was not settled by the earlier bounded search.
Harness: Codex, with independent mathematical review and a separately supplied research package. Repo tooling: research/kit/surrogate.py, research/kit/promote.py, verify/validate_candidate.py. The local verdict is verify/validate_candidate.py; the bundle was assembled by research/kit/promote.py. Compute: CPU only; fresh promotion RIS used 56.23 CPU seconds. This timing excludes earlier exploration and proof replay; full campaign cost was not measured.
The local evidence bundle (not committed in this PR) contains a portable matrix reconstruction, exact rowspace binding, witness checks and the local exhaustive lower-bound proof. A local replay script (not committed) replays the official fast RIS calls. Validation and prose used upstream d9ef620f71939868090938e954bd15573e2f2233; the final candidate is codes/124-20-13.json.
Publication validation: the fresh official gate also passed at upstream c4f4873bbe9ab4203e795d7cc832dab2cb4e0ae2; full verdict and final JSON binding are recorded in the local publication-gate verdict (not committed).