Radar Comparison

The Radar Comparison widget overlays pinned trials as polygons on a radar (star) chart.

One axis per objective (and optionally per parameter). It is the decision-phase companion to the Pareto and MCDM views. Once a handful of candidate designs has been narrowed down, pin them and compare their profiles in a single glyph. You can see which candidate is balanced, and which one buys its strength in one objective with weakness in another.

Workflow

1. Pin the candidate trials to compare (📌 in the Trial Table or the trial detail modal; up to 20 pins are kept). The widget shows every pinned trial as one colored polygon with a legend entry.
2. Choose the axes: objectives are always shown; enable Include parameters to add the numeric parameters. A radar chart needs at least 3 axes to be readable — with a 2-objective study, enable parameters or pin the widget next to a Pareto view instead.
3. Toggle Outward = better (default on) to orient all objective axes so that larger polygons are better (see below). With the toggle off, axes show plain normalized values.

Normalization

Each axis is min-max normalized over all completed trials of the study, so different units and scales become comparable radii in $[0, 1]$:

$$u = \frac{v - \min_a}{\max_a - \min_a}$$

With Outward = better enabled, axes of minimized objectives are flipped ( $u \mapsto 1 - u$) so that "further out" consistently means "better" on every objective axis, regardless of the optimization direction. Parameter axes have no better/worse direction and are never flipped. A degenerate axis (all values equal) is drawn at mid radius.

Caveats

• The enclosed area is not a score. Area grows quadratically with radius and depends on how the axes happen to be ordered; adjacent axes also visually correlate. Use the polygon shape for qualitative comparison and the MCDM widgets for a defensible ranking.
• Axis order is arbitrary (objectives in study order, then parameters). The same data with a different axis order produces a differently shaped polygon — do not over-interpret shape features that depend on which axes are neighbors.
• Min-max normalization is study-relative: a trial near the middle of a radar axis is average within this study, not in any absolute sense, and pinning/unpinning trials does not change the scale (the whole study does).
• For many trials or a continuous view of all trials, use Parallel Coordinates — the radar is intentionally limited to the pinned few.