v1.115.0

Added

• BenchResult.explain_selection() / PluginRegistry.explain() (A2 Phase S2): the full plot-selection decision table — one row per registered plugin, chosen entries first, each rejected entry carrying the first gate that dropped it (named-only, not in plot_list, excluded, missing capability, shape-filter mismatch, superseded backend resolution). select() is now exactly the chosen subset of explain(), so the table is the authoritative record of why a plot did or did not appear.
• Scorecard orientation toggle — each category's table can now be read two ways, switched by a control in the page header (the choice persists via localStorage). The default metric across benchmarks view is unchanged (one column per metric, one row per benchmark); the new metrics within benchmark view transposes it (one column per benchmark, one row per metric) so a benchmark's metrics stack with their sparkline time axes aligned. Both orientations are rendered from a single set of per-benchmark cells through one shared cell macro; the switch is pure show/hide of the pre-rendered tables, so it needs no data round-trip.