+ The Workbench read-model is functioning, but the committed report is stale and lacks: {missing || "no required tracks"}. + No MLX success is being inferred from absent data. +
+
+ {APPLE_UMA_ABSENCE_ACTION}
+
+ {mlx.reason}
: null} +| N | +p50 ms | +rows/sec | +parity | +copy boundary | +
|---|---|---|---|---|
| {entry.N ?? "n/a"} | +{numericLabel(entry.p50_ms)} | +{numericLabel(entry.rows_per_sec)} | +{boolLabel(entry.parity.parity_pass)} | +
+ {entry.copy_in_boundary ?? "copy-in not declared"}
+ {entry.copy_out_boundary ?? "copy-out not declared"}
+ |
+
Apple Silicon / UMA
++ Read-only evidence surface for Python, Rust, and MLX benchmark tracks. This view never runs benchmarks and never upgrades absent evidence into success. +
+