On the night of 20 October 1966, a girl in the Welsh mining village of Aberfan told her mother about a dream. Her name was Eryl Mai Jones, she was ten years old, and in the dream she had gone to school and the school was not there. Something black, she said, had come down all over it. The next morning she went to Pantglas Junior School as usual. A little after nine o’clock a waterlogged tip of colliery waste, half a million tons of it, slid down the mountain and buried the school. A hundred and forty-four people died, a hundred and sixteen of them children. Eryl Mai was one of them.
Stories like this are the reason the word prophecy still has a pull. And almost all of them fall apart the moment you look. The prophecy turns out to be vague enough to fit anything, or it was written down after the event and dressed as foresight. A good number were invented outright. The internet is full of predictions Nostradamus never made and forecasts Baba Vanga’s neighbours say she never gave. That material is worth taking apart on its own, and we will, another day.
This is about the other kind. The small number of cases where someone recorded a specific prediction, in front of a witness, before the thing happened, and where the easy explanations have to work harder than usual to close the gap. We are here to measure them, not to marvel.
The ruler
The useful thing about the Society for Psychical Research, the Victorian body that first tried to study this seriously, is that it kept arguing with itself about what would count as proof. In 1965 one of its former Council members, G. W. Lambert, wrote down the argument as a list. For a precognitive dream to be worth anything as evidence, he said, it had to pass five tests.
It had to be told to a credible witness before the event, not after. Little time could pass between the dream and the event. Nothing about the event could be expected when the dream occurred. The dream had to describe something literal, a thing that happened as dreamed and not a symbol read onto the event afterward. And the details had to tally.
Keep that list in your hand for the rest of this piece, because it does most of the work. Run the famous prophecies through it and they drop out at the first hurdle, the one about being told before. What remains after Lambert’s filter is a much shorter and much stranger list.
The seven hundred and one
Before Aberfan, before any of this had a name, the Society tried to build the archive. In 1886 three of its members, Edmund Gurney, Frederic Myers and Frank Podmore, published two heavy volumes called Phantasms of the Living. Inside were 701 cases, most of them the same shape: a person sees or dreams something about a distant friend or relative, and at that hour the friend is dying or in danger. The book was meant to be a mountain of testimony too large to wave away.
A Scottish lawyer named Alexander Taylor Innes waved it away in a single essay. His question was narrow and lethal. He called it “Where Are the Letters?” If someone dreams of a death and then, before the news can reach them, writes it in a diary or posts a letter to a friend, that document would be real evidence, dated by the event itself. The book claimed such documents existed. Innes asked to see them, and pointed out that in case after case there was no sign the investigators had ever held the letter in their hands.
Gurney gave the real count, and it cost him. Out of 701 cases, he could name three that met Innes’s standard and had a genuine contemporary document behind them. One of those three later turned out to be a fraud.
The Society for Psychical Research did not bury its critics. It printed Innes’s attack and Gurney’s reply in the same public journal, The Nineteenth Century, in 1887, and let readers watch the strongest collection of premonition evidence in the world shrink to two cases in real time.
So the flagship archive of Victorian psychical research, the 701 cases, comes down under pressure to two. Stranger still, even those two are hard to find now. Dig through the record today and you can put a name to one of them. A woman the book calls Mrs L. G. T., living near Adelaide in Australia, wrote two words in her diary on 18 November 1863: “two figures.” That was the night she saw, in her bedroom doorway, a former suitor in mourning. The date in her diary matched, weeks before any letter or obituary could reach her, the day his father had died. The other surviving case sits unnamed in the fine print of Gurney’s reply, and in a book by Trevor Hall long out of print. The best evidence the age could produce is now two entries, one of which the modern reader can barely locate.
That is the floor of the whole subject. Whatever is real here is small and hard to hold. Keep that in view while we look at the cases that seem, at first, bigger.
Aberfan, measured
Go back to Eryl Mai Jones and run her against Lambert’s list. Her mother heard the dream the night before, so it reached a witness in time, and the interval was hours. No one expected the tip to move. The school buried under something black is close to what happened, black slurry over the building, and the details tally closely.
It is the best single case of the disaster, and it is still not what it first looks like. The account we quote, the one about the school not being there, was written down after the disaster and signed by her parents, with the local minister attesting to it. That gives us testimony that the child had spoken the day before, not a note dated the night of the twentieth. The corroboration rests on a memory of the timing, sworn to by grieving parents rather than a document that beats the event to the page. Under Innes’s rule, that is the gap that sank the 701.
And around Eryl Mai the picture gets weaker fast. After the disaster the psychiatrist John Barker, helped by an Evening Standard journalist, asked the public to send in any premonitions they had felt. Around seventy-six people wrote back. Barker judged that twenty-two of them had told someone about their feeling before the tip fell. Only a handful of those had any independent witness to the timing at all. The rest were what he himself admitted they were: memories reshaped by a national tragedy, of little value as evidence, because they were created after the fact.
Barker turned Aberfan into an experiment. He and the journalist Peter Fairley set up a Premonitions Bureau where the public could file forebodings in advance, time-stamped, so a real signal could be caught before the event rather than reconstructed after it. In eighteen months it collected close to a thousand reports and recorded eighteen as hits, none of which it prevented. Twelve of the eighteen came from the same two correspondents.
Those two people are the next problem, because they look, for a moment, like the strongest cases of all.
The two who kept calling
One was a switchboard operator named Alan Hencher. In 1967 he told the Bureau that a plane would come down, and gave a number: a crash involving 123 people. Nine days later an airliner crashed on approach to Nicosia, in Cyprus, and 124 people died on impact, off by one. It is the kind of detail that stops you, and it was time-stamped by the Bureau before the crash, which is more than Aberfan can say.
The other was a music teacher, Kathleen Middleton. She sent the Bureau a written warning about Senator Robert Kennedy on 11 March 1968, and on 4 June she telephoned them three times, agitated, about him. That night, in Los Angeles, Kennedy was shot. He died the next day.
Both cases pass tests that Eryl Mai’s cannot. They were logged, with a date, before the event. On the face of it, that looks like proof.
It is not, because of a third report from the same Kathleen Middleton. She once wrote to the Bureau describing a vision of a petrified astronaut. It felt like a warning of a disaster to come. But the cosmonaut Vladimir Komarov had already been killed that day, when his Soyuz 1 capsule crashed, and the news had not been released yet. Her vision ran behind the event, not ahead of it, which is a different and much smaller claim. On Lambert’s list it fails the first test outright: there was no future left to see.
Hold Hencher and Middleton together and the shape becomes clear. Neither was an ordinary person with a single uncanny hit; both were the Bureau’s most prolific correspondents, filing warning after warning, most of which came to nothing. Between them they produced twelve of the eighteen recorded hits and a far larger pile of misses. A person who predicts disaster constantly will, sooner or later, predict one that happens, and the ones that don’t happen do not get quoted. The time stamp is genuine, and so is the volume of misses behind it.
The sun at Fátima
Unlike the private dreams so far, the cleanest pre-registration of them all was a public dare.
Through the summer of 1917, three shepherd children near the Portuguese village of Fátima said that a lady appeared to them and had promised a sign. They named the terms in advance: a miracle at the Cova da Iria, at noon on 13 October, a time and place printed and talked about for months. On the day, a crowd came to hold them to it. Estimates run from thirty thousand to seventy thousand people, many of whom had walked through rain to watch a prophecy be tested at a fixed time.
The crowd then reported that the sun moved: that it spun and changed colour, then seemed to plunge toward the earth before pulling back. Its weight comes from the witness who reported it: Avelino de Almeida, writing in O Século, the main Lisbon daily, a liberal, republican paper openly hostile to the Church. He had no reason to hand the Church a miracle, and he described the sun before the astonished crowd anyway.
The pre-registration is real and unusual. The catch is in what was pre-registered. The children promised “a miracle,” not “the sun will appear to dance.” The prediction fixed the time and left the content open, and an open promise is satisfied by whatever unusual thing a keyed-up crowd, staring at the sky after rain, happens to see. The weak point is the staring itself. Prolonged looking at the sun produces this family of effects: after-images and apparent motion, colours shifting as the retina fatigues. The physicist Auguste Meessen argued that the whole account can be built out of optics and expectation, and the reports vary the way subjective effects do: the colours and directions differed from person to person, and some witnesses saw nothing at all.
Two things keep the optical account from closing all the way. Some witnesses said they saw the phenomenon from villages miles from the Cova, people who were not packed into a suggestible crowd or told where to look. And the timing, the sun breaking through as the long-promised hour arrived, is either a genuine coincidence or something the explanation has to leave a little loose. The date was called in advance, and what answered it stays, at the edges, unaccounted for.
The cases that teach you to check the date
Set against those are the ones that feel strongest and are in fact the weakest, and they are worth keeping because each teaches a rule.
In January 1788, at a dinner in Paris, the writer Jacques Cazotte is said to have gone around the table and told each guest how the coming Revolution would kill them: this one by poison in his cell, that one by the guillotine. Every detail correct. The account is unbeatable until you check who wrote it. It comes from the papers of Jean-François de La Harpe, published around 1806, after the Revolution had already killed everyone in the room the way the story says. The perfection is the giveaway: a real premonition tends to be vague, while a memory of one comes out specific, because the writer already knows the answers.
Abraham Lincoln, the story goes, dreamed of a corpse lying in state in the White House and was told it was the President, killed by an assassin, days before Booth shot him. The dream may well have been real. But the only source is his friend Ward Hill Lamon, who set it down in a book published in 1895, thirty years later, with no contemporary record and some internal details that do not fit. It reaches us as a memory of a memory, long after the event it predicts.
And the astrologer Luca Gaurico is remembered for warning King Henry II of France to avoid single combat, before Henry died in 1559 from a lance splinter driven through his eye in a joust. The warning that survives in Gaurico’s published work, printed in 1552, says only that the king faced danger to the head around his early forties. The specific instruction about avoiding a duel is a later addition, glued onto the death after the fact, the same way a Nostradamus quatrain was. A dated document warned of vague danger, and legend sharpened it once the danger had a shape.
The rule these three share is the one Innes used to break the 701. Before you ask whether a prophecy came true, find out when it was written down.
Why the misses matter more than the hits
There is a mathematical reason so few cases survive, and it is no dismissal, only the size of the world.
The mathematician J. E. Littlewood put it as a joke with a straight face. Call a miracle an event with a one-in-a-million chance. In the hours we are awake and paying attention, things happen to us at a rate of about one a second, which comes to something over a million events a month. So each of us should expect a one-in-a-million coincidence about once a month, from nothing more than being alive. Persi Diaconis and Frederick Mosteller made the same point with more rigour in a 1989 paper on how to study coincidences: with a large enough sample, they wrote, any outrageous thing is likely to happen. A one-in-a-million event, across a country of hundreds of millions, is expected many times a day.
Now apply that to premonitions. Some eight billion people dream every night, and dreams of death or of a child in danger are among the commonest there are. After any large disaster, the numbers all but require that a few people recently dreamed something close and told someone about it. The Premonitions Bureau is the experiment that shows it: ask a whole nation for forebodings and hundreds arrive; run it forward as an alarm and it catches nothing and saves no one. The base rate produces the hits on its own. What it does not produce, in eighteen months of trying, is a single prevented death.
There is a cleaner test still, and it has been run. When the psychologist Daryl Bem published experiments in 2011 that appeared to show precognition under laboratory control, other researchers did the obvious thing and tried to repeat them. The main replication, by Stuart Ritchie, Richard Wiseman and Christopher French in 2012, found nothing. Put in a form where the prediction has to be registered before the result and checked by strangers, the effect did not survive. That is the strongest single fact on record, and most of the evidence runs the same way.
Where it doesn’t close
It closes over most of the field, but not all. Two places hold out, and they are worth naming rather than sewing shut.
The first is the corroborated tail of Aberfan. Strip away the seventy-odd reports gathered after the fact and you are left with a small number told to a witness before, Eryl Mai’s among them. The mathematics of coincidence covers them well, yet only statistically, as a statement about the pile, not a solution to the single case of a ten-year-old describing black coming down over her school the night before it did. The second is the handful of Fátima witnesses who reported the sun from miles away, outside the crowd, and the fit of the timing to a date set months earlier. The optical account handles most of it and reaches the edges of these and stops.
Neither seam proves anything. A statistical explanation that fits ten thousand cases does not have to solve the one that unsettles you, and the fact that it doesn’t is not evidence of the marvellous. This is the discipline the whole subject asks for and rarely gets: to hold a dated, specific prediction in one hand and the best rational account of it in the other, and to say where the account is complete and where it runs out, without deciding the question in either direction because the not-knowing is uncomfortable.
The Victorians, for all the fraud and the credulity, built the right instrument and then reported what it found. Seven hundred and one cases, and under real pressure, two. One of the two is a woman near Adelaide who, on a November night in 1863, wrote two words in her diary and nothing else. Two figures. It settles nothing. It has also never quite been explained.
Sources
Bibliography. The same list is held in the article’s frontmatter for the citation tools that read it programmatically.
- Gurney, Edmund, Frederic W. H. Myers, and Frank Podmore. Phantasms of the Living. 2 vols. London: Trübner and Co. / Society for Psychical Research, 1886
- Innes, Alexander Taylor. ‘Where Are the Letters? A Cross-Examination of Certain Phantasms.’ The Nineteenth Century 22 (1887): pp. 174-194
- Gurney, Edmund. ‘Letters on Phantasms: A Reply.’ The Nineteenth Century 22 (1887): pp. 522-533
- Barker, J. C. ‘Premonitions of the Aberfan Disaster.’ Journal of the Society for Psychical Research 44 (1967): pp. 169-181
- Knight, Sam. The Premonitions Bureau: A True Story. London: Faber and Faber, 2022
- Lambert, G. W. Criteria for the evaluation of precognitive dreams, Journal of the Society for Psychical Research (1965), as summarized in the SPR literature on precognition
- Almeida, Avelino de. Report on the events at the Cova da Iria. O Século (Lisbon), 15 October 1917
- Meessen, Auguste. ‘Apparitions and Miracles of the Sun.’ Physics and Society / conference paper on the optical account of the Fátima observations, 2005
- Diaconis, Persi, and Frederick Mosteller. ‘Methods for Studying Coincidences.’ Journal of the American Statistical Association 84, no. 408 (1989): pp. 853-861
- Littlewood, J. E. A Mathematician’s Miscellany. London: Methuen, 1953
- Paulos, John Allen. Innumeracy: Mathematical Illiteracy and Its Consequences. New York: Hill and Wang, 1988
- Bem, Daryl J. ‘Feeling the Future: Experimental Evidence for Anomalous Retroactive Influences on Cognition and Affect.’ Journal of Personality and Social Psychology 100, no. 3 (2011): pp. 407-425
- Ritchie, Stuart J., Richard Wiseman, and Christopher C. French. ‘Failing the Future: Three Unsuccessful Attempts to Replicate Bem’s Retroactive Facilitation of Recall Effect.’ PLoS ONE 7, no. 3 (2012): e33423
- McCorristine, Shane. Spectres of the Self: Thinking about Ghosts and Ghost-Seeing in England, 1750-1920. Cambridge University Press, 2010
- La Harpe, Jean-François de. ‘Prophétie de Cazotte.’ In Œuvres, published posthumously c. 1806
