Compare environments

Compare is one of a run’s tabs, alongside Environments and Side-by-side. It overlays the same metrics from multiple environments on shared charts, and answers the question the environments table can’t: not “what is each environment’s reward right now”, but “how do these environments’ curves differ”.

Like the other two, the tab only applies to reinforcement learning runs — it’s disabled on a run that logged no environments.

This is the environment-level counterpart to comparing runs — same idea, one level down. Compare runs to evaluate training configurations against each other; compare environments to look inside a single run at how its parallel worlds diverge.

How it works

The view has three areas:

Environments list (left) — the run’s environments with the usual table controls. Use the visibility control on a row to add an environment to the comparison.
Metrics tree (middle) — choose which metrics to plot. Show or hide it with the metrics tree toggle.
Charts (right) — one chart per selected metric, with a line per selected environment. Until you select one it shows “No environments selected — Select at least one environment to compare.”

Positioning in time

The RL Step / Episode toggle and the slider sit in the toolbar, exactly as they do on the environments table. They pin the point in the run that the charts are drawn for — see Episodes & steps.

Because that scale point already fixes the timeline, the chart x-axis mode isn’t offered here. The axis follows from the scale instead:

With an episode pinned, the x-axis is the environment step within that episode.
With an RL step pinned, the x-axis walks that step’s rollouts — its episodes concatenated and numbered in order.

Reading the comparison

Each environment keeps a consistent color across every chart, assigned the same way run colors are, and customizable from the row actions menu.
The chart tooltip’s Entity column shows the Environment rather than a run.
All the usual chart interactions apply — tooltip, zoom, highlight, fullscreen, and export.
Settings applies chart settings (y-axis scale, smoothing, reference lines) across the comparison.

Compare vs. Side-by-side

Compare overlays environments on shared axes — best when you want to spot divergence between a handful of them. With many environments the overlay gets crowded; use Side-by-side to give each its own column instead.