Loading rankings…
Loading rankings…
AI agent evaluation focuses on the agent — did it complete the task? But agents are directed by humans. SigRank measures the operator — the human directing the AI agent — with public, content-free evals.
Agents are tools. Operators wield them.
Agent evaluation measures the tool. Operator evaluation measures the wielder. Two operators using the same agent can have 100× different Yield. SigRank makes that difference visible.
Measures whether the AI agent achieves its goal end-to-end. Tests task completion, reliability, and autonomy.
Measures how effectively the human operator directs the AI agent. Tests cascade architecture, context reuse, and input efficiency.
Agent evaluation frameworks treat the human as a constant — the same operator, the same prompts, the same context. But in reality, the operator is the most important variable. Two operators using the same agent, the same model, and similar prompts can have 100× different Yield. The difference is the cascade architecture — how the human structures context reuse, output extraction, and input minimization.
SigRank measures that architecture using four token pillars — input, output, cache-read, cache-write — and computes Yield (Υ) = (cache_read × output) / input². The measurement is privacy-preserving: token counts only, never prompt content, never code.
Agent evaluation and operator evaluation are complements, not competitors. Agent evaluation tells you whether the tool works. Operator evaluation tells you whether the human is using it well. A great agent with a poor operator produces poor results. A poor agent with a great operator can still produce good results. You need both measurements to understand the full picture.
The public evaluation layer for AI operators — ranked by Yield (Υ).
The four layers of AI evaluation and where the operator layer fits.
Public operator evals vs private output evaluation.
Public operator evals vs private LLM observability.