# PRE-REGISTRATION — slow turns, attempt 2: the limit tested as a limit Written 29 September 2026, after attempt 1 failed, before this run. ## ⛔ Attempt 1 — FAILS, and it is the same error as the veil `PREREG_slow_turns_hide_2026-09-29.md` required the invisible fraction at **τλ = 0.01** to be within 3 standard errors of **0.50**. Measured **0.477750**, off by **8.91 SE**. The verdict stands. **0.50 is the τ → 0 limit, not the value at τλ = 0.01.** At that window the walk takes nine steps, so some paths near the equator cross out of the dark and the fraction is strictly below a half. **I tested an asymptotic value at a finite argument — the twelfth time in two days, and the same fault as entry 31's veil.** ## What attempt 1 established and does not need re-running - **Leg 1 held:** at an aligned start the invisible fraction is exactly **0** at every window. - **Leg 2 held, and it is the finding:** at τλ = 5 the fraction is **0.248950** against entry 32's **0.25**, at **0.49 standard errors**. **Entry 32's quarter is the fast-turning limit.** - **Leg 4 held:** monotone in the window, 0.4778 → 0.4528 → 0.4105 → 0.3172 → 0.2490, no reversals. ## The limit, tested as a limit The deficit from a half should scale with the angular step, which for diffusion goes as **√τ**. Attempt 1's own numbers say so: deficits 0.02225, 0.047225, 0.0895 at τλ = 0.01, 0.05, 0.20, whose log-log slope is **0.464**. **So the test is the exponent and the shrinkage, not the value at one window.** And saturation is reached **exactly** rather than by walking to it: the saturated distribution on the sphere **is** the uniform one, so a uniform sample is a saturated sample. That removes 80,000 integration steps and any question of whether saturation was reached. ## Criterion, fixed before the run 1. **The start is genuinely saturated.** For the uniform sample, ⟨c⟩ is within **3 standard errors** of 0 and ⟨c²⟩ within 3 SE of **1/3**. Both must hold, and this is checked before anything is inferred. 2. **The deficit scales as √τ.** Fitting log(0.5 − f) against log(τ) over τλ = 0.0025, 0.005, 0.01, 0.02, 0.04, the exponent is within **3 standard errors of the fit** of **0.5**. 3. **And it shrinks as that implies.** The deficit at τλ = 0.0025 is below **half** the deficit at τλ = 0.01 — since √(1/4) = 1/2, this follows from leg 2's exponent and is a separate check that it does. Tolerances are the fit's own standard error and a ratio derived from the exponent. **No number is chosen here.** Said because eleven criteria in two days were numbers fixed without asking what they imply. **WHAT KILLS IT:** leg 2. If the deficit does not scale as √τ, the approach to a half is not diffusive and the reading of why short windows hide more is wrong. **Prediction:** all three hold. **This attempt adds only the limit**; the finding — that the invisible fraction runs from a quarter at fast turning up toward a half at slow — is attempt 1's, and its leg 2 is where it rests. ## Discipline Attempt 1's FAILS is published beside this. No prose conclusion in the script.