Moving the Goalposts
A reader's summary of the informal fallacy of changing the test for success once the old test has been passed — its sporting metaphor, Popper's 1934 warning about ad hoc rescue, the 1970 Lakatos distinction that keeps honest revision from being condemned with it, the “AI effect,” and why medicine answered it with a public register.
The term at a glance
“Moving the goalposts” (or “shifting” them) borrows its image from team sports: if the posts are relocated mid-match, the same shot that would have scored no longer does. In argument it names the equivalent move — raising or altering the criterion for what counts as proof, success or an adequate answer after the first criterion has been met. Logic primers file it with the informal fallacies, close to the ad hoc rescue of a failed claim and to No True Scotsman, which redefines a category rather than a standard.
Origin
The metaphor is plainly sporting, and it is a twentieth-century idiom in English; the sources I could check do not settle the first figurative use, so this page does not name a date for it. The idea underneath it is much older than the phrase. Any arrangement that depends on an agreed test — a wager, a negotiation, a trial — is open to having the test quietly redefined, and philosophers of science gave the scientific form its sharpest treatment in the 1930s.
History and context
Karl Popper's 1934 Logik der Forschung, published in English as The Logic of Scientific Discovery, made refutability the mark of a scientific theory and listed the “conventionalist stratagems” by which a theory is kept alive against contrary evidence: new auxiliary hypotheses, redefined terms, doubts cast on the measurement. In 1956 Festinger, Riecken and Schachter published When Prophecy Fails, their field study of a group whose predicted flood did not arrive, and the behavior that followed — a reinterpretation that rescued the belief — became the standard illustration, taken up in the theory of cognitive dissonance. Imre Lakatos's 1970 essay then asked when such repair is legitimate, and in the same decade the artificial intelligence field coined its own version of the pattern. The internet did not invent the move; it removed the friction that once limited how long an argument could keep it going.
Main ideas
The criterion changes after the result is known
The defining feature is timing, not content. A claim is made, a standard for proving it is stated or implied, the standard is met — and then a new, harder standard appears. Nothing about the new bar need be unreasonable; what makes it a move is that it was not on the table when the first one was agreed.
Popper called the scientific version a conventionalist stratagem
In The Logic of Scientific Discovery (1934), Karl Popper described how a theory can be shielded from refutation by adding ad hoc hypotheses or redefining its terms after a failed prediction. He did not forbid revision outright; he held that such rescues lower a theory's testability, and that a theory which can absorb any outcome has stopped saying anything about the world.
Lakatos separated honest revision from the degenerating kind
Imre Lakatos's 1970 essay on research programmes gave the fair reply to Popper: scientists always adjust auxiliary assumptions, and sometimes rightly. His test is whether the adjustment predicts something new that is then borne out (a progressive shift) or only patches the hole it was made for (a degenerating one).
The failed prophecy is the textbook case
Leon Festinger, Henry Riecken and Stanley Schachter's 1956 book When Prophecy Fails followed a small group whose leader predicted a flood on December 21, 1954. When it did not come, the group did not abandon the belief; it announced that its faith had spared the world. The test had been set by the believers and was re-scored by them afterward.
Artificial intelligence has a standing version, the 'AI effect'
Pamela McCorduck's 1979 history Machines Who Think noted that each time a computer mastered a task once thought to need intelligence, observers decided that task had never really required it. A line credited to Larry Tesler, quoted by Douglas Hofstadter the same year, compresses it: intelligence is whatever machines have not done yet. Chess before and after Deep Blue's 1997 win over Kasparov is the usual example.
Online, it rides on the cost of asking
In a comment thread, a demand for a source costs one line; supplying it costs an hour. When the source arrives and the response is a demand for a better one, the exchange can continue indefinitely without anyone having to say what would count as enough — the dynamic that Brandolini's Law describes and that sealioning exploits.
Pre-registration is the institutional fix
Medicine's answer is to write the goalposts down before the game. From 2005 the major medical journals' editors' group (ICMJE) required clinical trials to be registered in advance, with their primary outcome named. Later audits, such as Ben Goldacre's 2015 COMPare project, found that many published trials had still quietly swapped outcomes — evidence that the temptation is persistent and that the register is what makes it visible.
Critique
- Not every new standard is a moved goalpost. Evidence can show that the original test was badly chosen — a proxy that turned out not to measure the thing — and saying so is correction, not evasion. Lakatos's distinction exists because science depends on this kind of revision; the useful questions are whether the change was explained openly and whether it makes new, checkable claims.
- Vague claims had no posts to move. Often the original standard was never stated, and each side privately assumed a different one. Calling the later, clearer demand a shift can punish the person who is finally being specific; the cleaner remedy is to ask both parties what would settle it before the next round.
- The accusation is itself easy to abuse. Like other named-tactic labels, it can end a conversation instead of describing one: any follow-up question can be answered with “you're moving the goalposts,” which is a way of declaring the matter closed without checking whether it is.
Impact
The practical lesson of the idea is to fix the test before the result: state what would change your mind, then see whether it happens. It sits beside steelmanning and Hitchens's Razor as a discipline for arguing in good faith, and it is the counterpart of the bait-and-retreat pattern in motte-and-bailey arguments. In a thread where one reply costs a minute and a rigorous answer costs an afternoon, naming the criterion early is the cheapest defense available to either side.