## A note on reading this table

This table warrants a little more information on what is going on. Each
row is one regression coefficient: what one model predicts about one
accuracy metric (KPI) when one input changes by one unit.

- **All metrics are lower-is-better.** Tighter groups, less broadhead
  drift. A negative coefficient means the change improves the metric.
- **Practical Effect Status.** *Detectable* means the effect survives a
  strict Bonferroni correction across the 6 KPIs tested for that model
  term. *Borderline* means it is statistically significant on its own but
  does not survive that correction. *No detectable effect* means the model
  cannot distinguish the effect from zero.
- **Survives Bonferroni?** The boolean version of the above. *Survives* is
  the strongest claim of statistical significance available in this table.
- **+5pp FoC Effect ÷ Observed Metric Spread.** How big the predicted
  effect is compared to natural build-to-build variation. ~1.0 means the
  effect is as large as one standard deviation of spread across the
  matrix.
- **Lower / Upper 95% Bound** columns are the 95% confidence range for the
  coefficient or the translated effect. A range that crosses zero means the
  effect could plausibly be in either direction, which is what typically
  produces a Borderline or No-detectable-effect status.

Full statistical methodology lives in the
[Methods](/research/arrow-study-2026/methods/) page.

[Figure: DataTable]