Cross-Tool Token Metrics, Not Agent Lock-in
aider is a terminal AI coding agent with a /usage command. SigRank is platform-neutral — works with aider, Claude Code, Copilot, and 15+ others. Your score follows you across tools.
The short version: aider
Verdict: SigRank does not replace aider — it scores the operator driving it. aider is a superb terminal-based AI coding agent, and its /usage command gives you honest token counts and dollar costs per session. That is useful when you live entirely inside aider. The moment you also use Claude Code for agentic work, Copilot for inline completion, or Cursor for refactoring, those metrics fragment: each tool reports its own numbers, in its own format, locked to its own surface. There is no unified score.
SigRank is the platform-neutral layer that fixes that. It reads token telemetry from aider, Claude Code, Copilot, and 15+ other tools, scores them all on the same cascade axis (Υ Yield), and gives you one rank that follows you across agents. You don't switch agents to use SigRank — you add it alongside whatever you already drive.
Feature comparison
| Feature | aider | SigRank |
|---|---|---|
| What it is | Terminal-based AI coding agent | Platform-neutral operator scoring layer |
| Token tracking | Yes (/usage command, cost-focused) | Yes (cascade-derived) |
| Cascade efficiency score (Υ = cache_read × output / input²) | No | Yes |
| SNR + Leverage + Velocity | No | Yes |
| Class tier (IGNITER to ARCH+) | No | Yes |
| Global operator leaderboard | No | Yes |
| Works across aider + Claude Code + Cursor + 15+ | No (aider only) | Yes |
| Score follows you across tools | No | Yes |
| Operator profiles + head-to-head compare | No | Yes |
| ed25519-signed snapshot submission | No | Yes |
| MCP server for agent integration | No | Yes |
| Privacy-preserving (token counts only) | Yes | Yes |
The agent lock-in problem
aider's /usage is real — but it is agent-scoped and agent-locked. The numbers live inside aider, in aider's format, visible only at the terminal where aider ran. They do not export. They do not compare to anyone outside aider. And they vanish the day you try a different tool.
Most operators do not live in one agent. A realistic week: aider for terminal-driven edits, Claude Code for agentic multi-file tasks, Copilot for inline completions, maybe a Cursor refactor. aider's /usage covers one slice of that week. Your actual efficiency is the union — and SigRank is the only layer that scores the union on a single axis.
The cascade is tool-agnostic
Υ = cache_read × output / input² is computed from four token integers that every AI tool produces — input, output, cache-read, cache-write. The math does not care which agent generated them. An operator who reuses context efficiently in aider scores the same way as one who does it in Claude Code. The cascade is the universal substrate.
Your score follows you, not the tool
SigRank's operator identity is tied to you, not to your agent. Enroll once, submit from any tool, and every signed snapshot feeds the same leaderboard rank. Switch from aider to Claude Code to Copilot over a month and your Υ trajectory reflects your driving across all three — not three disconnected per-agent gauges. That is the difference between a metric and a reputation.
Frequently asked questions
- Does SigRank replace aider?
- No. aider is a separate coding tool; SigRank measures token efficiency from supported sources. The current SigRank CLI has no dedicated aider log adapter, so merely installing SigRank does not import aider sessions or publish them to the leaderboard.
- Does aider have token usage metrics already?
- Usage or cost figures displayed by aider belong to that tool's own interface. SigRank does not automatically read aider logs; it measures supported local adapters and can separately calculate metrics from supplied four-pillar token counts. A manual calculation alone is not a verified leaderboard submission.
- Why does platform neutrality matter for token metrics?
- Using multiple AI tools makes cross-tool comparison valuable, but input coverage matters. SigRank can aggregate two or more supported active sources when explicitly invoked with `npx sigrank submit --platform multi`; it does not automatically include aider. The default `npx sigrank submit` uses Claude telemetry.
- Can I use SigRank with aider specifically?
- You can use aider alongside SigRank, but there is no dedicated aider log adapter at present. To publish from a supported local source, first create a connect code under SignalAF Settings, run `npx sigrank enroll`, preview with `npx sigrank submit --dry-run`, and then submit. The bare submit command uses Claude; other supported sources require an explicit --platform selection.
- What is the difference between aider /usage and SigRank metrics?
- aider's /usage answers "how many tokens did this session cost?" — a per-session, agent-local view denominated in dollars. SigRank's metrics answer "how efficiently does this operator drive AI across all their tools?" — a cascade-level, cross-platform view. aider tells you what you spent in one agent; SigRank tells you your Υ Yield (is signal compounding or burning?), your class tier, and your global rank among all operators regardless of agent. The first is a receipt; the second is a leaderboard.
Keep aider. Add the score that follows you.
SigRank can measure supported local token sources while you keep using aider. It has no dedicated aider log reader at present. To compare, use SigRank's manual calculator for supplied counts or enroll to submit signed telemetry from a supported platform.
Related: AI Coding Efficiency Tools · Measure AI Coding Efficiency · SigRank vs Cursor