What are AI operator metrics?

AI operator metrics are measurements of how a human operates an AI system, distinct from measurements of model capability or task correctness. Upsilon's draft core uses token telemetry to define Yield, Leverage, Velocity, SNR, and 10xDEV.

What data does Upsilon need?

The core measurement layer uses four token telemetry primitives: input, output, cache write, and cache read. It does not require prompt text, response text, source code, or repository contents.

Are operator metrics the same as productivity metrics?

No. Operator metrics describe the operator layer. Productivity, task success, code quality, and business value are external outcomes that can be analyzed alongside operator measurements.

Category guide

AI Operator Metrics

Model metrics tell you about the model. Operator metrics describe the human layer that selects, directs, iterates with, and extracts work from that model.

Four portable telemetry primitives

Input (I)

Fresh input tokens.

Output (O)

Output tokens.

Cache Write (W)

Tokens written to context cache.

Cache Read (R)

Tokens reused from context cache.

Upsilon v0.1 draft core

Yield (Υ)

(cache_read × output) / input²

Compound token-flow relationship between context reuse and output relative to fresh input.

Leverage

cache_read / input

Reusable context amplification relative to fresh input.

Velocity

output / input

Output generated per unit of fresh input.

SNR

output / (input + output)

Output share of the direct input/output exchange.

10xDEV

log₁₀(cache_read / input)

Log-scale cascade summary under the reference null policy.

What operator metrics do not tell you by themselves

A token cascade does not independently prove correctness, code quality, task success, employee productivity, hiring suitability, or business ROI. Those are separate outcome layers. Good evaluation joins them without pretending they are the same measurement.

Canonical definitions

This page is a discovery surface. SignalAF maintains the canonical Upsilon Standard draft and public reference implementation.

Upsilon Standard →AI operator standard →Operator evals →

Loading rankings…