SigRank — token-efficiency leaderboard that ranks operators, not just models

[SunrisesIllNeverSee]

Hey — thanks for Claude-Code-Usage-Monitor. The real-time monitoring + prediction angle is great, and it's part of what got me thinking about the ranking problem.

I built SigRank (signalaf.com) — a leaderboard that ranks AI coding operators by token cascade efficiency, not raw volume or spend. The metric:

Yield = (Cache Reads × Output) / Input²

Input is squared so re-pasting context costs quadratically; cache reuse compounds. The counterintuitive finding: the highest-volume operators often rank lower on efficiency because they hoard context and collapse output velocity. Volume is noise; yield is signal.

The npm package (npx sigrank) bundles ccusage + tokscale + tokendash as deps for the local read layer. Everything runs locally; only four signed token counts (Input, Output, Cache Write, Cache Read) leave the machine, ed25519-signed. submit --dry-run shows the exact payload first.

What the site gives you:

• Live leaderboard (7d / 30d / 90d / All-time) with platform filters
• Head-to-head comparisons + personal operator profile
• Hall of fame for the top operators

If you've got 60 seconds: https://signalaf.com/score — paste your stats, get your projected yield + ghost rank, no signup.

• Repo: https://github.com/SunrisesIllNeverSee/sigrank-app
• CLI/MCP: https://github.com/SunrisesIllNeverSee/sigrank-mcp
• Site: https://signalaf.com

Would love feedback on the metric design. Not trying to replace monitoring — this is the ranking layer on top of it.