# PRE-REGISTRATION — is what makes a crowd strong the same number as what makes it wrong? Written 29 September 2026, before the run. **Coherence + Many Tongues**, rank 11, sharing L, N, a. ## The two rooms **Coherence** (lands on Rayleigh's random walk): `N shared · √N unshared (√(πN)/2 in the plane)`. Minds that share a reference add up as N; minds that do not add up as a random walk. **Many Tongues** (lands on Kish's design effect, 1965): the error of an average is `σ²[ρ + (1 − ρ)/N]` — **a shared error ρ is a floor no crowd averages away.** One room says sharing is the gain. The other says sharing is the floor. They read as opposites. ## The claim under test Write each mind's aim as a shared part and a private part: `aᵢ = √ρ·u + √(1−ρ)·eᵢ`, with u the shared unit and eᵢ independent. Then the **expected squared length of the average aim** is ⟨|ā|²⟩ = ρ + (1 − ρ)/N which is **the same expression as Many Tongues' variance of an average**, term for term. **If that holds, the gain from sharing and the floor from sharing are one number.** And then, whenever the reference a crowd shares also carries the error it shares — ρ_error = ρ_aim — the crowd's signal-to-noise is **exactly 1 at every N and every ρ**: a crowd can buy neither strength nor accuracy by growing or by agreeing. ## Criterion, fixed before the run 1. **One expression.** Measured ⟨|ā|²⟩ from simulation matches `ρ + (1−ρ)/N` to within **0.002** at every combination of N ∈ {10, 100, 1000} and ρ ∈ {0, 0.1, 0.25, 0.5, 0.9, 1}. 2. **Coherence's endpoints are recovered.** At ρ = 1 the resultant length is **N** within 0.5%; at ρ = 0 it is **√N** within 3%. 3. **The ratio is one.** With ρ_error = ρ_aim, signal over noise equals **1.000 ± 0.005** at every N and ρ tested. **WHAT KILLS IT:** leg 1 failing means the two rooms are not the same expression and there is no pipe. Leg 3 failing means the identification of shared reference with shared error is wrong — which would be the more interesting result, and gets written as that. **Prediction, recorded now:** all three hold. Leg 1 is algebra and I expect it. **Leg 3 is the one I would not bet heavily on**, because it depends on an identification (the reference a crowd shares carries the error it shares) that is an assumption about the world, not a derivation. ## What this would compose with Entry 20 showed the ledger cannot see a counterfeit, so the external reference is forced. If leg 3 holds, **a crowd cannot escape that by being large or by agreeing** — the same conclusion reached from the other side. That is a composition, not new evidence, and must be written as one. ## Discipline Verdict assembled from measured quantities. Every table printed by the run, never typed.