# PRE-REGISTRATION — over-trust, attempt 4: with the verified tool, and stationarity checked first Written 29 September 2026. Fourth attempt at entry 40's open question, now that entry 41 has built and verified the integrator it was waiting on. ## What changed Entry 40's three attempts used a step that moved **4.85% on halving** and sat **39 standard errors** from (d−1)D. Entry 41 replaced it: a geodesic step with `φ = arccos(exp(−(d−1)D·dt))`, whose first harmonic is exact by construction and which measured **0 of 12 cells beyond 3 standard errors**. **Attempt 3's one failing cell was τλ = 0.01, where the window held only ten steps of the old, untrustworthy integrator.** That is the cell to settle. ## ⛔ And a structural point I should have seen three attempts ago Past saturation the walk's distribution on the sphere is **uniform**, so the process is **stationary** — and **nothing about a stationary process can depend on its age.** If the distribution at 4/λ and at 8/λ is the same distribution, then σ²_self measured at those two ages must agree, as a consequence, not as a coincidence. **So leg 2 below is largely a consequence of leg 1, and I say so rather than presenting it as independent evidence.** What attempt 3 was really measuring at τλ = 0.01 was its own discretisation. ## The window floor, derived rather than chosen A window's scatter is a measurement only if the window contains enough steps that the scatter is not dominated by single-step geometry. At dt = 0.005 and λ = 0.1, a window of τλ = 0.01 spans **20 steps**; τλ = 0.05 spans **100**. **This run uses τλ ∈ {0.05, 0.1, 0.25, 0.5}, every window at least 100 steps**, and reports τλ = 0.01 separately as the cell that failed before, with its step count stated. ## Criterion, fixed before the run 1. **Stationarity at both ages.** ⟨c⟩ is within **3 standard errors** of 0 and ⟨c²⟩ within 3 SE of 1/d, at t = 4/λ and t = 8/λ. Both must hold, at d = 3. 2. **Age-independence.** For each τλ in {0.05, 0.1, 0.25, 0.5}, the over-trust factor at t = 4/λ and t = 8/λ agree within **3 standard errors of their difference**. Zero of four may disagree. **Stated as largely following from leg 1.** 3. **Monotone in the window.** The factor decreases at every step across τλ = 0.05, 0.1, 0.25, 0.5, with no reversals. 4. **Reported, not scored.** The τλ = 0.01 cell at both ages, with its step count, for comparison against attempt 3's 10.19 standard errors. **WHAT KILLS IT:** leg 1 failing would mean the walk has not saturated by 4/λ and the whole framing is wrong. Leg 2 failing **with leg 1 holding** would be genuinely surprising and would mean age matters even in a stationary process, which would point at a bug in my measurement rather than at a finding. **Prediction:** all three hold. **This is the fourth attempt at this question and the criteria are set so that a pass is close to uninformative** — leg 2 follows from leg 1, and leg 3 already held in attempt 3. **What it settles is narrow: whether the old failing cell was an artifact.** The write-up says that plainly rather than presenting a fourth-attempt pass as a discovery. ## Discipline All three earlier verdicts stay published. No prose conclusion in the script.