Counting Is Not Scoring.

The short version: claudecount

claudecount is a simple token counter for Claude Code sessions. It reads your logs and prints the four pillars — input, output, cache-read, cache-write. It does its job: it counts. But counting is not scoring. Two operators can burn the same 50K input tokens and get wildly different outcomes. claudecount can't tell them apart.

SigRank reads the same token telemetry and asks a different question: is the cascade compounding or burning? The headline metric, Υ Yield = cache_read × output / input², rewards the operator who reuses cached context efficiently and penalizes the one who burns fresh input without leverage. claudecount gives you the numbers; SigRank tells you whether the cascade they describe is compounding or burning.

Feature comparison

Why counting isn't scoring

claudecount answers "how many tokens did I burn?" That's necessary but not sufficient. Two operators can burn the same 50K input tokens and get wildly different outcomes. One reuses cached context efficiently and produces 30K output tokens; the other re-sends the same context every turn and produces 3K. Same count, ten-fold difference in signal. On a counter, they look identical. On a yield leaderboard, the gap is obvious.

SigRank's headline metric, Υ = cache_read × output / input², measures exactly that gap. It rewards the operator who compounds cached context into output and penalizes the one who burns fresh input without leverage. claudecount gives you the four integers; SigRank tells you whether the cascade they describe is compounding or burning.

From counting to scoring

If you already run claudecount, you have the token counts. SigRank reads the same telemetry and adds the scoring layer counting never had:

npm install -g sigrank
sigrank enroll      # create your operator identity
sigrank submit      # reads logs, scores, signs, publishes

Prefer to inspect before you submit? Run sigrank me --dry-run to see your scored payload locally, or paste your token counts into the /score calculator to compute your Υ Yield, class tier, and compression ratio instantly, no account, no submission, just the numbers.

Frequently asked questions

Is SigRank a claudecount alternative?

They overlap on reading Claude Code token telemetry but diverge on what they do with it. claudecount counts tokens — input, output, cache creation, cache read. SigRank scores the cascade those tokens describe. If you want a counter, claudecount is that. If you want to know whether your token usage is efficient, SigRank answers that. You can run both — they read the same logs.

Why is counting not scoring?

Counting tells you what you spent; scoring tells you whether the spend was worth it. Two operators can burn the same 50K input tokens and get wildly different outcomes. One reuses cached context efficiently and produces 30K output tokens; the other re-sends the same context every turn and produces 3K. Same count, ten-fold difference in signal. On a counter, they look identical. On a yield leaderboard, the gap is obvious.

What does claudecount not measure that SigRank does?

claudecount reports raw token counts per session. SigRank reads the same four pillars and derives the cascade architecture: Υ Yield (is signal compounding or burning?), compression ratio (output per input), SNR (signal density), Leverage (how much cached context amplifies your input), and Velocity (tokens per unit time). claudecount tells you what you spent; SigRank tells you whether the cascade it describes is compounding or burning.

Can I use both claudecount and SigRank?

Yes. claudecount gives you the raw token counts. SigRank gives you the scoring layer that counting cannot provide. Run `sigrank submit` to publish your cascade score to the SigRank leaderboard, and keep claudecount for the raw view. The two are complementary, not mutually exclusive. The same local logs feed both.

Which is better for improving my AI coding efficiency?

SigRank. Counting can tell you when you're burning a lot, but it can't tell you why. Yield tracks the root cause: an operator whose Υ is low is burning fresh input without compounding cached context. Fix the cascade and the count drops automatically. claudecount shows the number; SigRank shows the meaning.