What survivesRev. 2026-09-07
Most of what you have been told did not survive being checked.
23 claims the self-improvement industry repeats, each with a verdict, the single number that decides it, and the citation. Where the honest verdict is that nobody knows, it says that instead.
Three of them come out well and are below, in full. That matters more than the ones that fail: a document where everything is wrong is a position, not a reading. The twenty-one days figure is a plastic surgeon’s memoir. Ego depletion failed across 23 labs. The 23 minutes appears in no paper at all. Spacing, if-then plans and paying people to stay clean all hold up, and two of those three are close to unused.
It carries a date and it gets revised. When a row changes, everyone who asked for it is told what changed and why. That is the whole of what ever arrives.
The ladder
- holdsReplicates. The popular version is roughly right.
- conditionalReal, and the popular version drops the condition it depends on.
- weakThe effect is smaller than the claim, or rests on less than the claim implies.
- failedTested and did not hold.
- unsourcedTraces to no study at all.
Three of the 23
It takes 21 days to form a habit.
unsourcedmedian 66 days, range 18 to 254
The 21 comes from a plastic surgeon's 1960 memoir, watching patients get used to a new face. He wrote 'a minimum of about' and the hedge was dropped in the retelling. When it was actually measured, automaticity took a median of 66 days and varied by an order of magnitude between people and habits. Buried in the same study: missing a single day did nothing to the curve.
- Lally, van Jaarsveld, Potts and Wardle, 2010. Median 66 days to automaticity, range 18 to 254 across 96 people (European Journal of Social Psychology)
- Maltz, 1960. Psycho-Cybernetics. An observation of patients, not a study
Willpower is a finite resource that gets used up.
failed23 labs, 2,141 people, indistinguishable from zero
The muscle-that-tires model failed a coordinated multi-lab replication. Every technique built on protecting your reserve is built on nothing. Set the default instead: organ donation runs 42% consent under opt-in and 82% under opt-out, same people, one setting.
- Hagger et al., 2016. Ego depletion across 23 labs and 2,141 people came out not different from zero (Perspectives on Psychological Science)
- Johnson and Goldstein, 2003. Organ donation consent, 42% opt-in against 82% opt-out (Science)
Spreading study out beats doing it in one go.
holds839 assessments across 317 experiments
One of the largest and least controversial effects in the whole of learning, and almost nobody spaces on purpose. The gap should scale with how long you need it: roughly 20 to 40% of the retention interval for something a week away, 5 to 10% for something a year away. For a test next week that is every day or two. For something you want to keep, monthly.
- Cepeda, Pashler, Vul, Wixted and Rohrer, 2006. Distributed practice, 839 assessments in 317 experiments (Psychological Bulletin)
- Cepeda et al., 2008. Optimal gap scales with retention interval, over 1,350 people (Psychological Science)
The other 20
Give an address and they arrive at it, along with the revisions when a row moves.
Who wrote this
The same reading is what the rest of it is built on. Forvel is a coach that holds you to one thing at a time, in the words you typed rather than a summary of them, and has no streak to break: three of the rows above are the reason it works the way it does. See what it is. Nothing here depends on that, and this document is the same document either way.