RuleUsed asWins by luck aloneStronger version
Threshold rule, 2 posts per versionused asBoth posts clear the line and the other version's don'twins by luckat most 12.5% (when each post clears half the time)stronger3 or 4 posts per version, or the same winner in a second pair
Block majority, 3 arms, 2 of 3 blocksused asAn arm beats both others in at least 2 of 3 blockswins by luckabout 78% (21 of 27 block outcomes produce a winner)stronger3 of 3 blocks (about 11%), or more blocks with a margin over the runner-up
Rank the top lanesused asTop 4 of 6 topic lanes by pooled scorewins by lucknot pass or fail: any lane is in the top 4 by chance 2 times in 3strongertreat it as a ranking to follow up, then confirm the winner with a 2-arm test
Two-proportion z-test, one lookused asThe default for yes or no metricswins by luck5%, by designstrongerfine as is; add a guardrail-only interim
The same test, 10 looks, stop at the first winused asWhat happens without a rulewins by luckroughly 15 to 20% (run the simulator)strongera planned sequential design, or one look
Read the amplification-style rule, an arm beating both others in 2 of 3 blocks, as a direction and not a finding: it picks a winner about 78% of the time even when all three arms are the same. A rule with a false-win rate above 10% caps your confidence at suggestive.