Between March and May 2026 the Algorithm Game asked its audience for one exact thing six times: a word, a word of their choosing, a number, a number between one and ten, one of two words, and one sentence. Each time we counted how many people delivered it exactly.

For every single-token instruction the exact rate sat between two thirds and nineteen twentieths. For the one instruction that asked for a sentence, with an apostrophe in it, the exact rate was one fifth, and nine tenths of the same room said the sentence in some wording.

This paper puts the six counts side by side. It is the first paper in the series that is about the crowd rather than about a round.

Every round in the series had its own hypothesis. The question across them was never posted, which is why this paper is a synthesis and not an experiment: when a crowd is asked for one exact thing, what decides how many people deliver it exactly?

• Rounds included: 1 (YES), 2 (anything), 3 (a number for a slot), 6 (a number from 1 to 10), 7 (OPEN or TAKEN), 8 (I'M NOT THERE YET). Rounds 4 and 5 asked for actions off the platform and have no exact-token count.
• Act I transmissions with an exact token, counted from the archive as comments whose whole text is the token: 1 (the aura scan, 23 Oct 2025), 8 (the Sleep Cycle Test, 23 Oct), ENTER (Level 4 access, 21 Jan 2026), and the sentence I'M STILL HERE (Collapse Protocol, 14 Nov). These are stricter than the ledger counts, which accept a token with words after it.
• Exact: the comment text equals the token asked for and nothing else. For Round 3 and Round 6, any valid number counts as exact; the choice of number is a different question, answered in Seven. For Round 7, either word counts.
• Counting base: the ledger for each round, as filed on the round record. Rounds 1, 2, 3 and 6 count all comments including replies; Rounds 7 and 8 count top-level comments only. The bases are named in Table 1 and not mixed silently.
• Meaning rate: only Round 8 has one, because only Round 8 asked for something with more than one way to write it.

The obvious explanation is the number of places a comment can go wrong. YES has one way to be exact. A number has one. A sentence has a capital letter, an apostrophe, four spaces and a choice between I'M and I AM, and every one of them is a place where an ordinary phone keyboard produces something a strict rule rejects. The failure that showed up in Round 8 was punctuation, not defiance: 133 people added emoji or extra words, 287 refused, and the rest of the gap between 22 and 89 percent is spelling.

The spread among the single tokens is a second finding. YES, the lowest at 66 percent, was the first instruction the Game ever issued, to its largest crowd, on a thread that also carried the most liked refusal on record; 30 percent of that room wrote something other than the word. Round 7, the highest at 95 percent, came to a crowd that had by then been trained by six rounds, with a mechanism that only worked if you used one of the two words. The rate rose with the crowd's experience and with how much the token was needed to make the round work.

Act I agrees, from before the rounds existed. Three tokens from the first months (the digits 1 and 8, and the word ENTER) land at 76 to 87 percent exact under a stricter count than the ledger uses. The one Act I sentence, I'M STILL HERE, four words and an apostrophe, lands at 56 percent: below every token and above the eight-character sentence of Round 8. The rule holds across both acts and the two sentences order themselves by length.

For anyone who writes instructions for a crowd, the practical reading is short. If the thing you need is exactness, ask for a token. If the thing you need is meaning, ask for the sentence and count it by meaning, because counting it strictly will fail four in five of the people who did what you wanted.

• Two counting bases. Four rounds are counted over all comments and two over top-level comments. Replies are less likely to be exact tokens, so the first four rates are, if anything, understated relative to the last two. The direction of the finding does not depend on it; the exact spread does.
• Ten instructions is ten data points. The pattern across single tokens (experience and necessity raise the rate) is a reading of ten numbers, not a fitted model, and we present it as a reading.
• Two sentences only. The 22 and 56 percent strict rates are from two instructions, both with an apostrophe. A sentence without punctuation might land anywhere between here and the tokens.
• Two later tokens agree, from partial samples. The June 2026 re-entry posts asked for IN and PLAY; in the parts of those threads the archive holds (19,976 and 36,150 top-level comments, both incomplete), the whole-comment exact rates were 84.9 and 78.9 percent. They are not tabled because the samples are partial; they sit inside the single-token band.
• No independence. The same audience, growing and learning, took every round. This is a panel, not six samples.

One instruction in three lengths, run on three days: a word, a three-word phrase, and a sentence with an apostrophe, each with the same pass rule. If exactness falls with length the way this synthesis suggests, the three rates should step down in order, and the size of each step is the number this paper cannot yet give.

The round this paper reads is 10 posts, not one. Every post is below with the platform's numbers and what the comment archive holds of each.

The rounds are filed in the Chronicle as THE YES ROUND, ⟡029, THE CONVERGENCE, ⟡030, and THE NUMBER, ⟡035. The law the series entered, participation is cheap and precision is dear, is the claim this paper measures.