It Was Not Subliminal. Eleven People Wrote In To Say They Could See It.
The insertion device was installed on 12 August 1957 and it was real. Measured on a calibrated rig, its shortest exposure is one forty-eighth of a second — a whole frame, and visible to anyone looking.
A months-long reporting project. Documents cited below are held in The Vault and available to readers.
FORT LEE, N.J. — The insertion unit is a Tachistolite Model 3. In June we put one on a calibrated rig in Rochester and asked it one question: how short a flash can you make?
Twenty point eight milliseconds. One forty-eighth of a second, over 340 trials, spread half a millisecond.
A projector at twenty-four frames a second shows every frame twice, behind a two-bladed shutter. One forty-eighth of a second is one of those illuminations: a whole frame, at full brightness, once. The gate cannot open for less. A bright flash on a dark surround stops being visible at a few milliseconds, and the machine's floor is several times above that.
Nothing at that drive-in in 1957 was below the threshold of perception. The equipment could not get below it.
What was installed, and when
The hire invoice is dated 9 August 1957: thirty-five dollars to install, thirty-one fifty a week for six weeks. The engineer's log has the unit in the projector on the night of 12 August. The device existed, it was fitted, it ran, and the man who ran it wrote down what it did.
The receipts
The circuit kept weekly concession returns. The six weeks before installation average $1,412.40. The six weeks with the unit running average $1,674.90.
A rise of 18.6 per cent, and for sixty-nine years the only number anybody quotes.
The cup
A supply invoice dated 19 August, in the test's second week, changes the popcorn cup from seven ounces to ten and the board price from twenty-five cents to thirty.
Per ounce that is a cut of 16 per cent. Cups sold rose 21.4 per cent over the test; popcorn dollars rose 44.
The heat
The Bergen County weather record has nine consecutive days at or above 90°F from 28 August into the first week of September, mean maximum 94.
They fall inside weeks three and four. Across that fortnight the soft-drink line rose 31 per cent. Popcorn, on the same nights, rose 4.
We gave twelve weeks of returns, the cup change and the daily temperatures to a statistician with the intervention date withheld, and asked her to model the series blind. Adjusted for cup and weather, the residual is plus 1.2 per cent, interval plus or minus 4.1.
She attached a second finding we are obliged to print as loudly as the first: on six weeks of one drive-in's takings, the design could not have detected anything smaller than about nine per cent. Our test cannot convict the machine or clear it.
The eight nights it was not in the projector
Estil Marchetti-Bowe's log runs to 42 nights and has the unit in the projector on 34. It came out for four while the lamphouse was rebuilt, three when the take-up spring went back to the hire company, and one wet Thursday.
The strongest concession week of the six is week four, at $1,791. The lamphouse rebuild covers four of its seven nights.
The best week of the famous experiment is a week the machine spent mostly on a bench.
Eleven letters
Bound in the circuit's patron correspondence file, between a complaint about mosquitoes and one about a car radio, are eleven letters written between 20 August and 14 September 1957.
They complain, in varying tempers, of a flicker in the picture with words in it. Two mention a headache. One asked for his money back and got it: the voucher book records one car admission, ninety cents, 4 September, initialled by the manager.
The report to the client
Lambert Sedgwick-Voss submitted ten pages to the circuit on 4 November 1957. It claims 57.5 per cent on popcorn and 18.1 on a soft drink.
We cannot construct 57.5 out of any series in the file. It does not mention the cup.
In 1962 he told a trade interviewer his sample had been too small to carry what he had said of it. That retraction has been quoted about a dozen times, against some four thousand repetitions of the 57.5.
Whether he knew about the cup, this newspaper cannot establish. The invoice was addressed to the cinema, not to him; his papers were destroyed by a burst pipe in 2004; his daughter says she does not know.
So we print that, rather than arranging the paragraphs to let a reader decide it for us.
What this desk takes from it
The nearest thing in our own vault is PX-1612, sixty-nine years later, and it fails in the same place: the thing being measured moved at the same time as three other things.
The founding experiment of the subliminal advertising industry involved no subliminal stimulus. Its documented human consequences are eleven annoyed letters, at least two headaches, and ninety cents refunded to a man in a car.
Dr. Sedgwick-Voss's reply, below, says our arithmetic is too blunt to have caught what her father claimed. She is right. The machine was real. The rise in takings was real. The only thing that was never there is the part that was supposed to be invisible.
The six o'clock edition
Three stories a morning, in your inbox before the coffee is made.
Sources & Method
We assembled the paper nobody had put on one table: six weeks of the circuit's own weekly concession returns and the six weeks before them, the projection engineer's nightly log, the equipment hire invoice, a supply invoice, the county weather record, the patron correspondence binders and the refund voucher book. Then we hired a metrology laboratory to measure a surviving unit of the same model on a calibrated rig, and gave the receipts to a statistician who was not told what the intervention was or when it happened.
Who we spoke to
- The circuit's records store, Secaucus, N.J., Weekly concession returns 1 July – 22 September 1957, supply invoices, the equipment hire invoice, the patron correspondence binders, the refund voucher book, and the researcher's ten-page report to the client. Read over four days by arrangement with the successor company; every page relied on is scanned and published with this story May 2026 The eleven letters had not been produced to anyone outside the company since a clerk filed them in 1957.
- Estil Marchetti-Bowe's nightly projection log, Projection engineer, 12 August – 22 September 1957; the unit's hours, faults and removals, night by night. Lent by his grandson, digitised at our expense and returned; no payment offered or made April 2026 Records the insertion unit in the projector on 34 of 42 nights, with a reason entered for every one of the eight it was out.
- Piotr Nkansah-Vogel, Instrument engineer, metrology laboratory, Rochester, N.Y.. Commissioned to measure a surviving Tachistolite Model 3 on a calibrated rig; his protocol was written and published before the unit reached him June 2026 Shortest achievable gate opening 20.8 milliseconds, spread 0.5, over 340 trials. He would not be drawn on what the 1957 unit did.
- Dr. Amara Lindqvist-Obi, Statistician. Given twelve weeks of concession returns, the cup change and the daily weather series with the intervention date withheld, and asked to model the series blind June and July 2026 Returned a residual of plus 1.2 per cent with an interval of plus or minus 4.1, and a note saying the design could not have detected anything smaller than about 9 per cent.
- Dr. Frances Sedgwick-Voss, Retired market-research statistician; daughter of Lambert Sedgwick-Voss, who ran the 1957 test. Interviewed by video call twice, then sent the full draft, the receipts and the laboratory protocol with three weeks to answer July and August 2026 Her reply is printed whole at the foot of this page and is the strongest argument against this story that we received.
Documents
- PX-1612 — Traffic-composition audit, a mid-sized advertising exchange accepted
What we could not confirm
- Whether the unit we measured behaves as the 1957 unit did. It is the same model from the same maker, serial 118 against the hire invoice's 94, and its shutter linkage carries a repair tag dated 1974. Sedgwick-Voss's own report refers once to the gate having been 'modified for the trial'. We have not found what the modification was, and we measured the machine that survives.
- Whether the researcher knew about the cup. The supply invoice is addressed to the cinema, not to him, and it is dated four days after his second site visit. His working papers were destroyed by a burst pipe in a storage unit in 2004. His daughter says she does not know. We cannot establish it, and we are not going to imply it by arranging the paragraphs so that a reader concludes it for us.
- Whether eleven letters are all the letters. The patron correspondence binders for August and September 1957 are complete and consecutively numbered. October's binder is not in the store, and a complaint written in the last week of the test would have been filed in it.
How Others Covered This
The same events, as reported elsewhere on the same day. We list what each outlet had that we did not, as well as what we had that they did not — including where we come off worse. Why we print this.
-
Signal & LedgerThe Industry's Founding Number Does Not Reconcile To The Till
Read it as a measurement-standards problem for readers who buy attention for a living, and ran our residual with its interval attached.
Had that we did not
The confidence interval, in the standfirst, which is more than this newspaper managed in its own headline.
Left out
The eleven letters. They are not a measurement problem, and they are the only people in the story.
-
ApertureThe Flicker
An essay on what a country wants a hidden persuader to be, and on the peculiar comfort of believing you can be reached without being asked.
Had that we did not
The observation that the legend needs the machine to be invisible, and that the machine's entire practical difficulty in 1957 was that it was not.
Left out
The receipts. The essay says the effect on sales was 'never measured'. It was measured weekly, in a ledger, for six weeks, and the ledger is a forty-minute drive from the essayist's desk.
-
They Buriedthis newspaperIt Was Not Subliminal. Eleven People Wrote In To Say They Could See It.
Put the receipts, the engineer's log, the hire invoice, a supply invoice, the weather and eleven letters on one table, then measured the machine.
Had that we did not
The eleven letters in full, the statistician's working, the refund voucher, and the interval on our own result.
Left out
Our headline says eleven people could see it. Our own figure says our test could not have detected the effect the researcher claimed even if it had been there. Both are true; only one of them is in the type that gets read. — V. Ashcombe-Doyle, standards editor
Right of Reply
They Buried contacted Dr. Frances Sedgwick-Voss, statistician, daughter of Lambert Sedgwick-Voss on 6 July 2026 by video call, again on 24 July with the full draft, the concession returns, the laboratory protocol and three weeks to answer. Replied 14 August 2026 by email, asking that it run entire and unsoftened. Printed unedited.
I will give you the cup before you ask for it. My father should have put the cup in his report and he did not, and I have said so twice in print, in 1998 and in 2011, before your newspaper existed. If he knew, it was dishonest. If he did not know, it was incompetent, and he was not an incompetent man, which is a sentence I have had sixty years to sit with. You have not caught him at anything I have not already conceded.
Now let me tell you what is wrong with your piece, since you asked me to.
Your own statistician told you the design could not have detected an effect smaller than about nine per cent. Your interval is plus or minus 4.1. Those two facts together mean you did not run a test of my father's claim. You ran a test that was incapable of resolving it, and got the answer such a test always gets, and then you wrote the phrase 'indistinguishable from nothing', which every reader alive will read as 'nothing'. That is the identical slide you have spent a thousand careful words accusing him of. He turned a small number into a large one by leaving something out. You have turned an unknown into a zero by leaving your own power calculation in the fifth section under a subheading.
On the machine. You measured a unit built in the same year, in 2026, with a shutter linkage that somebody rebuilt in 1974, and you have printed one forty-eighth of a second as though it were a fact about the summer of 1957. It is a fact about a machine in Rochester. My father's report says the gate was modified for the trial. You do not have the modification, so you measured the thing unmodified and put his name next to the number. I notice you say all this yourself, honestly, in your third Unverified line, where it can do no damage to your headline.
On the eleven letters, I am genuinely grateful, and I would like copies, and I did not know they existed. But consider what they are. Roughly sixty thousand cars came through that lot in six weeks. Eleven people wrote in. A cinema gets eleven letters about the sound. That eleven people saw a flicker is not the same finding as your headline, which implies the message was legible to the audience; not one of your eleven correspondents reproduces the words correctly, and one of them thinks the picture said EAT POPCORN when the slide, which is in the same file you read, said HUNGRY? EAT POPCORN.
And on my father, briefly, because you were decent about him and I want to be accurate rather than fond. He made money for about two years. He then spent thirty-one years being introduced at dinner as the man who invented a fraud, including by people who were at that moment selling attention by the thousand. In 1962 he told an interviewer that his sample was too small to support what he had said, which is a retraction, and which appears in your story as one sentence, set down after the arithmetic, where nobody will trip over it. He was a salesman who wanted to be a scientist and got the order of those two words wrong.
You have written a careful piece about a careless man. My complaint is that you finished it with the same confidence he did.
Published unedited under our right-of-reply guarantee.
How was this story?
We publish the result, whatever it is. Reader verdicts appear on the front page and in our newsroom metrics.
Readers' Letters 0
Printed at once under the name you give and read by the desk afterwards; anything unfit is removed, with a note saying so, and nothing else is ever deleted — only corrected. Letters that changed something in the story carry a mark saying so, and there are 5 of those across the archive.