garden of forking paths
Many analysis choices, only one reported.
What it is
Gelman & Loken (2013) coined this term for a subtle form of p-hacking: every researcher makes dozens of small analytical choices (which covariates to include, how to handle outliers, which subgroups to define, which transformations to apply). Each choice is defensible in isolation, but the combination implicitly searches over a large 'garden of forking paths', inflating the false-positive rate even without conscious p-hacking.
Why a reviewer cares
Reviewers ask: would this result survive a different reasonable choice at any of these forks? If you trimmed outliers at 2.5 SDs, would 3 SDs give the same answer? If you covaried for age, what happens without? Sensitivity to a single reasonable choice flag is fragile evidence.
How to fix it
Pre-register the analysis plan, including the specific transformations, exclusion criteria, and covariates. Where pre-registration isn't possible, run multiverse analyses — present the result under every defensible combination of analytical choices, and let the reviewer see the distribution.
This is one of ~15 canonical methodology explainers Paper Review's red-team report links to. To get a full review of your manuscript, start a Paper Review — $5.