Sutra
The figure of merit · k · d² / n

Logical qubits k, distance d, physical qubits n. It rewards encoding more logical qubits and tolerating more errors, while penalising qubit overhead — the single number that captures whether a code is a good deal. Baseline: IBM's Gross code scores 12. Select a row to inspect the code, its lattice, and how its distance was verified.

exact = MIP-certifiedbound = BP+OSD upper bound
Code [[n,k,d]]k·d²/nvs GrossDistancequbits/log.Novelty
[[144, 12, 12]] · Gross12.0
1.00×
baseline12.0IBM 2024
[[196, 18, 24]]52.9
4.41×
bound10.9novel
[[192, 16, 20]]33.3
2.78×
bound12.0novel
[[196, 8, 26]]27.6
2.30×
bound24.5novel
[[192, 8, 24]]24.0
2.00×
exact24.0novel
[[168, 8, 22]]23.1
1.92×
bound21.0cyclic risk
[[180, 6, 26]]22.5
1.88×
bound30.0cyclic risk
[[198, 6, 27]]22.1
1.84×
bound33.0cyclic risk

Seven codes from a 60,000-candidate search; 14 of the top 50 beat Gross after rigorous re-verification. Distances are exact only where marked — fast decoders upper-bound the distance and can over-state it.

[[192, 8, 24]]
12×8 torus · weight 8 · random
2.00×
vs gross
24.0
k·d²/n
24.0
qubits / log.
30/24
d_X / d_Z
distance exactnovel

Distance — MIP exact — every one of 255 cosets solved (CBC). BP+OSD had over-estimated d=26; the solver corrected it to 24.

The certified result. Asymmetric distance (d_X=30, d_Z=24) — valuable under biased noise. The only code here whose distance is proven exactly.

Stabiliser neighbourhood · 12×8 torus
weight 8max L1 = 4 · heavy-hex ✓
L qubit (via A)R qubit (via B)X-check torus wrap
The honest tradeoff

A higher figure of merit buys less margin elsewhere. Our top codes win decisively on k·d²/n — and currently lose on the circuit-level threshold, which sits below today's hardware error rates. Closing that gap (a fault-tolerant schedule, full-scale runs) is the next phase. The tradeoff is the finding.

k·d²/n (best)up to 4.4× Gross
qubits per logical (best)10.9 vs Gross 12
circuit-level threshold≲ 3–5×10⁻⁴ vs Gross ≈ 5×10⁻³
vs IBM gate error≈ 1×10⁻³ — codes not yet hardware-ready

Threshold figures are preliminary (naive, non-fault-tolerant syndrome schedule); exact values await the next phase.

Why verification matters
claimed (raw search)
[[192, 20, 20]]
score 41.67 · d=20
verified
d = 8
score 6.67 · Δd = -12

Ranked #1 on the raw leaderboard at a claimed distance of 20 (score 41.67). Rigorous BP+OSD verification dropped it to distance 8 (score 6.67). We publish failures like this — an unverified figure of merit is worth nothing.

Faithful to the program's open release (code · data · demos). Every figure here is reproducible; distances are labelled exact (MIP-certified) or bound (BP+OSD upper bound).