AI Coding Metrics — The Complete Guide
Six metrics, four token pillars, one cascade. The complete map of how SigRank measures AI coding efficiency — and which number to look at first.
The full metric set
SigRank defines six core metrics, all derived from the four token pillars (input, output, cache-read, cache-write). They are not six independent numbers — they are six lenses on the same cascade. Yield is the synthesis; the other five are the diagnostic views that explain why your yield is what it is.
Read them together. A high yield with a low cache hit rate means you are efficient despite poor cache reuse — probably because your output is very high relative to input. A high yield with a high cache hit rate means you are compounding: reusing cached context to amplify a small fresh input into large output. The latter is the sustainable pattern. The metrics tell you which one you are.
The six metrics
Yield (Υ)
Start herecache_read × output / input²The headline metric. Measures the architecture of the cascade — whether signal is compounding or tokens are burning. The single number that captures operator efficiency.
Compression Ratio
Pair with yieldoutput / inputHow much you get back per token you put in. A blunt but useful ratio — high compression means the model is doing more with your input than parroting it.
Signal-to-Noise Ratio (SNR)
Quality lenssignal tokens / total tokensThe density of useful output in your cascade. High SNR means most of what flows through is signal, not boilerplate or repetition.
Cache Hit Rate
Efficiency lenscache_read / (cache_read + cache_write)How well you reuse cached context. A high hit rate means you are reading from cache — free tokens — rather than writing new context you may never reuse.
Leverage
Compounding lenscache_read / inputHow much cached context amplifies your fresh input. High leverage means a small input is riding on a large cached foundation — the compounding effect.
Velocity
Secondaryoutput / session_timeTokens produced per unit of wall-clock time. The throughput metric — useful, but it measures speed, not efficiency. A fast operator burning tokens is still burning tokens.
How they relate
Yield is the composite. It multiplies cache reuse (cache_read) by output and divides by the square of input. That means yield rises when you reuse cache, when you produce more output, and when you send less fresh input — all at once. It is the one number that rewards the full compounding architecture.
Compression ratio is the output-over-input slice of yield — it ignores cache entirely. It tells you whether the model is doing more with your input than echoing it, but it cannot see whether you are compounding. Cache hit rate and leverage are the cache slices — they tell you how well you reuse context but say nothing about output. SNR is the quality lens on output. Velocity is the time lens — orthogonal to efficiency, useful for throughput comparisons but not for ranking.
Which to prioritize
Start with yield. It is the headline metric and the one the leaderboard ranks by. If your yield is low, open the diagnostic lenses: check compression ratio to see if your output is too thin relative to input, check cache hit rate to see if you are failing to reuse context, check leverage to see if your cached foundation is too small. SNR tells you whether the output you do produce is dense with signal. Velocity is a secondary metric — track it for throughput, but do not optimize for it at the expense of yield.
Go deeper
FAQ
- What are AI coding metrics?
- Six metrics derived from four token pillars: yield, compression ratio, signal-to-noise ratio, cache hit rate, leverage, and velocity. Each measures a different facet of the token cascade — the flow of tokens through an AI coding session.
- Which metric should I prioritize?
- Start with yield (Υ). It captures the architecture of the entire cascade in one number. Then use compression ratio and cache hit rate as diagnostic lenses to understand why your yield is what it is.
- How do the metrics relate?
- Yield is the composite — it multiplies cache reuse by output and divides by input squared. Compression ratio is the output-over-input slice. Cache hit rate and leverage are the cache slices. SNR is the quality lens. Velocity is the time lens. Yield is the synthesis; the rest are diagnostics.
- Yield vs. velocity — what is the difference?
- Velocity is tokens per unit time — how fast you produce output. Yield is token-cascade efficiency — how well you compound signal. A fast operator burning tokens can have high velocity but low yield. SigRank ranks by yield, not velocity.
- How do I measure my metrics?
- Install the SigRank CLI (
npm install -g sigrank), enroll, and submit a snapshot. The on-device scanner reads your four token pillars and computes all six metrics. No message content is ever read.