My First MO§ES™ SigRank — the diagnostic x-ray of the token economy
───────────────────────────────────────────────────────────────────
The leaderboard that judges AI coding operators by architecture, not budget.
The problem with every other ranking: they measure volume. Who spent the most, who generated the most output. That's not architecture — that's a budget statement.
SigRank measures Υ = (Cache·Output)/Input² — a metric that penalizes raw-input padding quadratically. Volume can't buy rank. The only way to climb is to build a compounding architecture.
───────────────────────────────────────────────────────────────────
Two independent instruments say the same thing
SigRank's thesis is backed by external measurement at mos2es.com/benchmarks:
┌─────────────────────┬──────────┬────────────┬───────────────────┐ │ kernel │ MO§ES™ │ field avg │ delta │ ├─────────────────────┼──────────┼────────────┼───────────────────┤ │ Cache hit rate │ 94.66% │ 90.68% │ +3.98pp · #1 │ ├─────────────────────┼──────────┼────────────┼───────────────────┤ │ Output : Input │ 17.9× │ 0.162× │ 110× field avg · │ │ │ │ │ #1 │ ├─────────────────────┼──────────┼────────────┼───────────────────┤ │ Tokens per task │ 810K │ 4.67M │ 5.8× fewer · #1 │ ├─────────────────────┼──────────┼────────────┼───────────────────┤ │ Time per task │ 1.84 min │ 11.92 min │ 6.5× faster · #1 │ ├─────────────────────┼──────────┼────────────┼───────────────────┤ │ Cost per LOC │ $0.0007 │ $0.067 │ 96× cheaper · #1 │ └─────────────────────┴──────────┴────────────┴───────────────────┘
The operator that ranks #1 on Υ also ranks #1 on every external kernel. Architecture is the mechanism.
───────────────────────────────────────────────────────────────────
The math
Leverage C_r/I factors into three compounding stages:
(O/I) × (C_w/O) × (C_r/C_w) = C_r/I transmission · commitment · compounding = leverage
10x DEV is log₁₀ of that product — the amplification exponent. Thecascade tells you which stage an operator is winning or losing on,not just the bottom line.
───────────────────────────────────────────────────────────────────
The grounding
SigRank is built on the Conservation Law of Commitment: C(T(S)) ≈ C(S) — commitment is conserved when enforcement holds. Published atZenodo DOI 10.5281/zenodo.18792459. Four patents filed.
───────────────────────────────────────────────────────────────────
Try it
git clone https://github.com/SunrisesIllNeverSee/moses-sigrank cd moses-sigrank ./sigrank
Or paste your ccusage claude --json output directly on the Space. Sign in with HuggingFace to save your row to the board.
Space: build-small-hackathon/sigrank