Compare robots by class

Compare one source table or a reviewed intersection of compatible rows from several tables. Cross-class columns and automatic label joins are impossible by design.

Comparison workspace

Source-table mode remains the default, and existing table= links retain their view. Reviewed profiles expose only maintainer-audited intersections across source tables. Inspect all metrics, exact authored differences, published gaps, or reported cells without linked evidence. A dash remains a published gap, not zero; values stay in authored units and conditions.

Loading the compact comparison data…

What the CSV preserves

Claim identity

Comparison scope, profile and metric ids, physical class, source page/table/row, subject, reported/missing state, authored value, and stable claim URL.

Evidence identity

Every evidence id plus its stable registry URL and direct source URL. Empty evidence stays empty.

Descriptive completeness

The evidence-mix card counts unique backing records by grade. Per-subject cards count reported rows, published gaps, out-of-row cells, and evidence-linked cells. None ranks products.

No unit guessing

Difference views compare authored state and text. Numeric-looking text is not coerced, converted, or treated as equivalent.