Publication of the upload-frequency study: report, methodology, binned aggregates, and a de-identified row dataset with a self-contained replication notebook.
Headline (pre-registered, read once on a sealed 20% holdout): posting more does not lift each video (a small −10% per doubling, too small to act on). Total reach scales roughly one-for-one with output, and that result survives trimming the biggest videos. All correlational; direction unestablished.
Reproduce from dataset/replicate.qmd.