FA-74810 / Experiment statistics / Member archive

Switchback carryover washout: Every block start is washed out · case 05

Data after repeated-arm blocks is discarded, wasting sample for no carryover.

Member previewVariant 5 · 3 implementations · 8 checks per implementation

Case contract

Time is split into blocks of block_len minutes; block b runs schedule[b % len(schedule)]. Events [minute, value] in the first washout minutes of a block are excluded only when that block's arm differs from the previous block's arm (block 0 has no previous block). Return [mean A, mean B, excluded count] with None for an arm without events.

Why this case matters

Switchback experiments on marketplaces must discard carryover right after each switch.

One recorded failure

Sample boundary fixture

This sample comes from the broken implementation of a controlled reproducer.

Boundary fixture	Actual	Expected	Outcome
no washout when the arm repeats	`[null, null, 2]`	`[5.0, null, 1]`	Failed

MEMBER ARCHIVE

The complete case is available to members.

This record includes three runnable implementations, regression fixtures, execution results, and source hashes.

Member access is invitation-based. Sign in with your invited account to inspect the sources.