Comparison Table

The Comparison Table widget lays out pinned trials as columns and their objectives, parameters, and user attributes as rows, with the best value in each row highlighted. It is the numeric counterpart to Radar Comparison: where the radar shows the qualitative shape of a handful of candidates at a glance, the table shows their exact values side by side, ready to quote or export.

Workflow

1. Pin the candidate trials to compare (📌 in the Trial Table or the trial detail modal). Each pinned trial becomes one column.
2. Read down each row to compare a single objective, parameter, or user attribute across every pinned trial; the best cell in each row is highlighted.
3. Unpin trials to narrow the comparison, or pin more to widen it — the table always reflects the current pin set.

Reading the table

• Best-cell highlighting is per row and direction-aware. For an objective row, the highlighted cell is the minimum or the maximum depending on that objective's optimization direction — not always "the biggest number." Parameter and user-attribute rows have no optimization direction, so they are shown without a best-cell highlight.
• Categorical parameters are shown as their label, not coerced into a number — a row for a categorical parameter simply lists each pinned trial's chosen category.

Caveats

• Values are shown raw, with no normalization. Unlike Radar Comparison, which min-max normalizes every axis to make units and scales comparable on one chart, this table intentionally shows the real numbers — it is the place to check exact figures, not silhouettes. If you need a scale-free comparison of overall shape across many axes at once, use the Radar Comparison widget instead.
• The table is the numeric ground truth; the radar is the qualitative shape. Use them together: skim candidate profiles on the radar to shortlist, then confirm the actual numbers behind that impression here.