# PRE-REGISTRATION — a faithful sphere integrator, and does Drift's harmonic ladder hold? Written 29 September 2026, before the run. Clears the queue item filed out of entry 40's failure, and tests the Drift room's own published law. ## The tool problem, from entry 40 The step used all night — isotropic Gaussian noise in the ambient space, then renormalise onto the sphere — moved **4.85%** on halving and sat **39 standard errors** from (d−1)D. So **Drift's published rate has never been verified here**, and entry 40 claims nothing about it. ## The scheme, and why it is exact where it matters Take a **geodesic step**: from x, pick a uniformly random unit tangent direction t (orthogonal to x) and move x' = cos(φ)·x + sin(φ)·t Because the tangent component averages to zero, `E[x'·L | x] = cos(φ)·(x·L)` for **any** fixed axis L. So after n steps `⟨x·L⟩ = cosⁿ(φ)`. Choosing φ = arccos( exp(−(d−1)·D·dt) ) makes `⟨x·L⟩ = exp(−(d−1)D·t)` **exactly, at any step size.** The first harmonic is right by construction. **That is what makes the test non-trivial.** The l = 1 rate is built in, so it proves nothing about the scheme. **The l = 2 harmonic is not built in** — if it also decays at the rate the Drift room prints, the scheme is a faithful diffusion and the room's ladder is verified. The room prints: *"the l-th harmonic at l(l+d−2)D"*. For l = 2 that is **2d·D** — 6D at d = 3, 10D at d = 5. The l = 2 zonal harmonic in d dimensions is proportional to `d·c² − 1`, which has mean zero on the uniform sphere since ⟨c²⟩ = 1/d there. ## Criterion, fixed before the run 1. **The first harmonic is exact.** The fitted decay rate of ⟨c⟩ matches (d−1)D to within **10⁻⁶** relative, at d = 3 and d = 5, and **halving dt changes it by less than 10⁻⁶** relative. *(A check on my implementation, not on the scheme — it is exact by construction and I say so.)* 2. **The second harmonic lands on the room's ladder.** The fitted decay rate of ⟨d·c² − 1⟩ matches **2d·D** to within **2%** at d = 3 and d = 5, at the finer step. 3. **And it converges.** Halving dt moves the l = 2 rate **closer** to 2d·D at both d — the residual at the finer step is smaller than at the coarser one. Two percent on leg 2 because the l = 2 signal decays 2d/(d−1) times faster than the first — three times faster at d = 3, 2.5× at d = 5 — so it has far fewer usable decades before it is noise. The tolerance comes from the quantity's own decay, not from a number chosen to pass. Said in advance, because seven criteria tonight were numbers picked without asking that question. **WHAT KILLS IT:** leg 2 failing at the finer step with leg 3 also failing — that would mean the scheme is not a faithful diffusion and the tool is still not built. Leg 2 failing while leg 3 holds means it is converging but needs a finer step, which is a result about the numerics, not about the room. **Prediction:** all three hold. **If leg 2 holds, Drift's published ladder is verified at l = 2 for the first time here** — and that is a result about the room, reached with a tool built because the last one was not good enough. ## Discipline No prose conclusion in the script. Tables printed by the run. Hashes read from the file. The l = 1 leg is labelled as a self-check, not as evidence, because it is true by construction.