v1.147.0

Added

• The run-order strip is optional. order="auto" | "always" | "never" on
  write_report, report_from_rows, report_from_results and every draw_* that draws
  the strip, and --order on bencher compare --report and bencher evidence-rows. With
  auto (the default) the strip shows only for a design that is not simple (several
  shards, paired runs, an interleaved, ABBA or mixed order, several seeds, or a within-arm
  drift of at least the margin); otherwise the caption states the design in one line.
• Each summary mark in an evidence graphic shows its value beside it, in text ink:
  each arm's median bar (more digits when two medians would read the same), each forest
  row's change, each waffle arm's rate (18/20 · 90 %) and the difference, the sweep's
  change per point, the shift plot's medians and named deciles, the power bar's MDE and
  Δ, and each diagnostics trend's drift. No interval is drawn; the CI stays in the
  caption and the alt text.

Changed

• A view of one metric gives its change in the metric's unit (scorecard panels,
  estimation, shift and power plots): base median × (ratio − 1), with the ratio CI
  converted the same way. A unitless positive metric stays in %, a binary one in pp, and
  views that share one axis across metrics (forest, sweep effect row, heatmap, table) keep %.
• The estimation plot (G3) has no effect panel: both arms, the ±Δ band and the median
  connector only.
• Each evidence graphic is as wide as its content and its text matches GitHub's body
  text. One metric is 490 px wide, several metrics or points 648 px (before: 800 px for
  all); the figure shows at 1.2 px per pt (ticks 13 px, labels 14 px, title 16 px, the big
  number 20 px, nothing under 12 px). A scorecard puts at most 3 arms panels in a row, in
  balanced rows.