Experiment Registry
3 of 4 moved a coefficientMake lift real and learnable. Anecdotes are logged but never move a coefficient; missing holdouts are flagged loudly.
| What changed | Segment | Holdout | Lift (90% CI) | Evidence | Learned? |
|---|
| Interactive product tour on day 1 | self-serve | control-A | 8.0% [5.0, 11.0] | rct | updated |
| Annual-plan nudge at checkout | smb | control-B | 4.0% [1.0, 7.0] | holdout | updated |
| Cold outbound to lapsed trials | mid-market | missing | 6.0% [-2.0, 14.0] | observational | updated |
| Founder LinkedIn post | — | missing | 20.0% | anecdote | no |
Reflexivity
A repeated tactic’s lift is decaying — likely being copied. Half-life ≈ 1.8 cohorts.