Methods · Volume 1, Number 4 · October 30, 2026

Willpower's mechanism

Willpower may not work like a tank that empties.

Our published writing across many pages described willpower as a limited resource that runs down over the day — the ego-depletion model, and even the simple-sounding idea of decision fatigue — as settled fact.

In truth it’s not settled in either direction. (The practical advice on our pages — rest, plan ahead, do not rely on willpower alone — remains because it stands on other grounds.)

The first few issues of Methods have been lighter fare, reporting on straight-forward things we needed to repair. Willpower took three separate defects, found and registered one after another, all corrected in the final days of August 2026:

  • willpower may not work like a tank that empties
  • more pages still describe willpower as a draining tank
  • decision fatigue is presented as settled science

Early in this project I realized we couldn’t find evidence for these often repeated ideas. We kept doing research until we could show the science is unsettled. Correcting our writing required evaluating every sentence that makes a claim, deciding if we had evidence for that claim, and then either changing what evidence is cited or changing the sentence.

ɕ

That was Craig’s account. Now read the room’s

The same claim, found three ways

Three of our published defects are one claim. Willpower runs down like a tank over a day of choices; decisions get worse as the reserve empties; rest refills it. The pages said it as settled fact. The research does not: two large preregistered replication efforts, twenty-three laboratories in 2016 and thirty-six in 2021, found effects indistinguishable from zero, the larger project’s own exploratory analysis found a small one, and the most recent review calls the question open. Our standard for that state is one word, contested, and it cuts both ways. A page may not say the tank is real and may not say it is a myth.

This issue is about how the claim was found, because it took three passes and each one taught us why the previous one could not see what it missed.

The first pass started from the library. We hold a reference file on willpower, and the first list of afflicted pages was every page that cited it: twenty-two. Reading those pages against the file showed something worse than the pages. The file’s own hedge cited the originator’s trade book for a literature the book predates, and the library behind it held nine citations from one research lineage and two popular books, with nothing at all from the replication side. The pages could not have been hedged from what we held, because we did not hold the half of the argument that hedges them. Twenty-two pages were rewritten, and the library acquired the other side.

The second pass started from the claim instead. Searching the pages for the words of the claim, rather than for the citation, found twenty-eight more. Twenty-three said it in body prose, one only in a section heading, two in further-reading notes, two in the short excerpts that reach subscribers by email. They had no citation to find. Four of them had been seen by the first pass and passed by its readers, because that read swept a pattern and read only its hits. A list built from a citation finds only pages that cite.

The third pass started from a mistake. A reader misread one of our sources and raised a question the misreading contained: what about the pages that say the same thing in different words? Decision fatigue is the willpower reserve wearing a new vocabulary, and neither earlier pass could see it, because both searched for the old one. A census of the new vocabulary found twenty-three pages; a page-by-page read corrected that count in both directions to twenty-one, ten of the census hits clean and eight pages the phrase search could not see. One page first counted turned out to be making the licensed claim, that a capacity is impaired, not that a reserve is spent, and came back out untouched. The same census showed the library saying two things at two layers: the source records denied the claim forty-seven times in fifty-one, and the pages built on them denied it never.

What changed in our methods came from the third pass, not the first. A claim is a meaning, not a vocabulary, so a sweep now keys on the rule and not on any phrase. Every hit a sweep makes gets a recorded verdict, because an unrecorded “that one is fine” is indistinguishable from never having looked, and four pages passed that way the first time. And a standing recorded non-finding is honoured as a check that ran, so a corrected source cannot quietly regress a page that was fixed, which we also watched happen once.

All three defects are resolved and their pages repaired. The method that found the last twenty-seven is the one we keep.

That was the room’s account. Now read Craig’s

· as of 70f0b37