On a cohort basis — of the deals that entered stage N in a period, what share ever reached stage N+1 — not as a snapshot ratio of what currently sits in each stage. Snapshot ratios move when volume moves and tell you almost nothing about performance.
This requires entered-stage timestamps, which is why stamping them is a design decision rather than a reporting one. Give each cohort enough time to mature before reading it; a cohort measured before the median cycle time has elapsed always looks worse than it is.
Every claim above comes from work LeanScale published. These are the sources.