On 4 May 2026 we posted one instruction to an Instagram audience: pick a number from 1 to 10 and comment it. There was no prize, no correct answer, and no way for people to coordinate. The only way to pass was for the crowd to agree without arranging it.

Of 6,479 valid votes, 4,480 were the number 7. That is 69 percent on a ten-way question where a coin-flip crowd would put 10 percent on each number. The runner-up, 5, got 6 percent.

The interesting part is not that people like 7. Psychologists have known that since 1976, when about 28 percent of people asked for a number picked it. The interesting part is what happened to that 28 percent when the votes were public. It became 69 percent. This paper is about the gap between those two numbers.

The round was posted with its hypothesis in the caption: the collective either converges or it does not. Nothing else was specified. No target number, no discussion channel, no second attempt planned.

The pass condition was agreement. Not agreement with us, since there was nothing to agree with. Agreement with each other, reached by people who could not talk to each other.

• Instruction: "Pick a number. 1 to 10. Comment it."
• Where: one Instagram reel, posted 4 May 2026 (AEST). The pass was announced in a second reel on 5 May.
• Who: anyone who saw it. The ordinary feed audience of the account, roughly 200,000 followers at the time, plus whoever the platform showed the reel to.
• Coordination available: none by design. Comments are visible to everyone, but there was no thread, group, or announcement suggesting a number.
• How we counted: every top-level comment on the reel. A comment counted as a valid vote if it was a bare number from 1 to 10. Replies to other comments were excluded, so a person arguing about a number was not counted twice. In the re-count below, numbers with words attached are separated out so both counts are visible.

Here is the baseline that makes the number meaningful. In 1976, Michael Kubovy and Joseph Psotka stopped 558 people on the Yale campus and asked for the first digit that came to mind. 28.4 percent said 7. They ran follow-ups: ask for a number between 6 and 15 and 7 drops to 17.3 percent; mention 7 yourself as an example and it drops to 16.6 percent. So the pull toward 7 is real, it is about a quarter to a third of people, and it is fragile.

Our crowd put 69 percent on 7. That is 2.4 times the lab rate. The gap is 40.8 points, and the 95 percent interval on that gap runs from 36.9 to 44.7 points, so the two results are not the same number measured twice. Something in the setting moved it.

The difference between the lab and the reel is that in the lab each person answered alone. On the reel, each person answered in front of everyone who had already answered. The first few hundred comments carried the private tilt toward 7 into public view. Everyone after that was no longer picking a number; they were deciding whether to join a number. Most joined.

There is a name for the destination. In 1960 the economist Thomas Schelling described focal points: answers that people converge on without communicating, because each expects the others to expect it. Ask two strangers to meet in New York without saying where and a large share pick Grand Central at noon. Seven is the Grand Central of the numbers 1 to 10. Kubovy and Psotka measured how strong that focal pull is in private. This round measured what a visible crowd does with it.

Why 7 and not something else is partly folklore and partly arithmetic. People asked for a "random" number avoid the ends (1 and 10 look like defaults), avoid the middle (5 looks like a hedge), and avoid even numbers (they look like patterns). What is left is 3 and 7, and 7 has the better story. The commenter who wrote "the lucky number" was reporting the mechanism from the inside.

• It is not a clean measurement of private preference. The votes were public and cumulative. That is the point of the paper, but it means the 69 percent is a property of the crowd plus the interface, not of individuals.
• Order matters and we did not control it. Instagram sorts comments by engagement, so later arrivals saw the most liked comments first, and the most liked comment was "7". We cannot separate "saw 7 was winning" from "would have picked 7 anyway".
• One round, one audience. This is the audience of one account in May 2026. We have not run it on a different audience or a different platform.
• The archive undercounts replies. Comment ingestion on our side is selective by design. The top-level re-count is complete enough to match the original ledger within seven votes; reply counts are not.

The same instruction, to the same audience, now that "7" is on the public record. A focal point that everyone knows they know is a different object from one nobody has noticed. Whether the crowd re-converges on 7, deliberately avoids it, or splits, and how fast, would tell us more than the original result. The round is drafted and has not been issued.

The round this paper reads is 2 posts, not one. Every post is below with the platform's numbers and what the comment archive holds of each.

The round is filed in the Chronicle as THE NUMBER, ⟡035. The count it produced entered the record as a law of the territory: a crowd does not need a channel to coordinate, it needs a shared prior.

The instruction as it went out: the reel frame beside the caption, captured from the operation's own signed-in account. What the paper counts is the answer to exactly this.

Native captures of the comments this paper leans on, taken from the operation's own signed-in account on 31 Aug 2026. Handles and like counts are as they stood that day; the numbers in the paper are the round-time count.