Skip to main content
Off the GPU box, every Run mocks: real CLI, real Ledger, fake scores. That makes a laptop the safe place to learn the loop. You need Python 3.12+, uv, and a clone of monte-nemo-platform.
1

Install dependencies

This creates the project .venv. All commands below run through uv run.
2

List the installed Environments

demo appears as mock-only. It exists to exercise the loop itself.
3

Create a Measurement

This freezes the Splits, the eval settings, and the Environment fingerprint. The name hello is yours to choose.
4

Run the Baseline eval

The CLI prints the Run plan and asks launch?. The mock Run takes about 30 seconds. This is eval@0, the score you compare every later Run against.
5

Train a smoke chunk

A smoke Run is a capped sanity pass. It is never evidence: it cannot be a Baseline, a chain parent, or a promote source.
6

Read the Ledger

You get the Run rows and the Lineage’s Baseline line. Smoke rows carry a [not evidence] marker. There is no delta yet: a delta appears after a real training chunk and its follow-up eval.
A real score needs a real box. When you are ready, read Run on a real box.