Early access — cascade metrics are real (derived from canonical token telemetry); the operator field is a curated seed. Learn more about the data
◈ Evaluator

AI Evaluator — What SigRank Does Differently

An AI evaluator assesses AI performance. Most evaluate models, outputs, or safety. SigRank is an AI evaluator for the operator layer — the only one that answers “who is best at using the AI?”

What is an AI evaluator?

An AI evaluator is a system that assesses AI performance. The category includes model evaluators (MMLU, LMSYS Chatbot Arena) that rank models on test suites or preference votes, output evaluators (LLM-as-judge, human review platforms) that assess whether the output is correct, and safety evaluators (red-teaming frameworks, alignment benchmarks) that test for harm. Each evaluates a different layer of the AI stack.

SigRank is a fourth kind of AI evaluator — an operator evaluator. It assesses the humans driving the AI, not the AI itself. The question it answers — “who is best at using the AI?” — has had no evaluator until now, because there was no privacy-preserving way to measure it. Token counts make it possible; reading prompts does not.

Five capabilities that set SigRank apart

  • 1. Content-free telemetry. SigRank captures four token pillars — input, output, cache-read, cache-write — and nothing else. No prompt content is ever read or stored. Token counts are the minimal sufficient statistic for operator evaluation, and they make the evaluator privacy-preserving by design.
  • 2. The Yield metric. Υ = cache_read × output / input² captures cascade architecture in a single number — whether signal is compounding or tokens are burning. It blends all four pillars and penalizes fresh input quadratically, reflecting the real cost asymmetry of the cascade.
  • 3. Cohort-relative ranking. Operators are ranked against the live field, not against a static threshold. Your score reflects where you sit relative to other operators right now. The field shifts as the population grows, so the ranking stays meaningful.
  • 4. ed25519-signed provenance. Every snapshot is cryptographically signed on-device and verified server-side. The data is trustworthy without being readable — you can prove the token counts came from a real session without revealing what was in the session.
  • 5. Continuous scoring. Operators are scored over 7-day, 30-day, 90-day, and all-time windows, not on a one-off test. You see your trend, not just a snapshot. Improvement is visible; regression is visible. The evaluator runs on every session, in the background.

How to get evaluated

Install the SigRank CLI or MCP server, run a coding session, and submit your ed25519-signed token snapshot. The snapshot is verified server-side and your operator score is computed from the four token pillars. You appear on the public leaderboard with a codename — your real identity is never shown. You can also paste token counts into the /score page for an instant Yield score without submitting. No prompt content is ever read or stored.

Explore the category

AI Evaluation — Measuring the Operator, Not Just the Model

The four-layer model of AI evaluation: model, output, safety, operator. SigRank is the AI evaluator for the operator layer.

The SigRank Index — Methodology

How operator scores are computed from four token pillars, verified server-side, and ranked. The canonical methodology behind the SigRank evaluator.

Score Your AI Coding Sessions

Paste your token counts and get an instant Yield score. Try the SigRank evaluator on your own sessions — no signup required.

The SigRank Leaderboard

See the ranked operators the SigRank evaluator has scored. The public board of AI operator performance, updated continuously.

FAQ

What is an AI evaluator?
A system that assesses AI performance. Model evaluators rank models, output evaluators assess quality, safety evaluators test for harm. SigRank is an operator evaluator — a new category.
What makes SigRank different?
It measures the operator, not the model. Four token pillars, the Yield metric, cohort-relative ranking, ed25519-signed snapshots. The only AI evaluator that answers “who is best at using the AI?”
What are SigRank's five capabilities?
Content-free telemetry, the Yield metric, cohort-relative ranking, ed25519-signed provenance, and continuous scoring over multiple time windows.
How do I get evaluated?
Install the CLI or MCP server, run a session, submit your signed snapshot. Or paste token counts into /score for an instant Yield score. No prompt content is ever read.