Module sequential 

Sequential / always-valid testing via the mixture mSPRT approach.

Key property: once Significant is returned, subsequent calls with more data never downgrade it — the ever_significant flag is sticky.

Practical decision rule (simplified mSPRT for median delta):

1. Require min sample per arm (same as fixed-horizon).
2. Compute bootstrap CI.
3. Apply alpha spending: effective α = 0.05 / ln(max(n,e)). This bounds Type I error uniformly over all stopping times.
4. CI threshold: lo > 0 (increase) or hi < 0 (decrease).
5. Once Significant, stays Significant (ever_significant is sticky).

Structs

SequentialSummary

Enums

Decision

Outcome of a sequential significance test.

Functions

decide

Always-valid decision with a sticky ever_significant flag.