# PRE-REGISTRATION — does one bottleneck width close the Ghost's residual? Written 29 September 2026, before the run. Closes, or fails to close, the residual entry 44 left open. ## Where this came from Entry 44 found the Ghost's published passage times are in simulation steps at dt = 0.01 — 40,042 and 414,010 against a computed 415.53 and 4,155.31, ratios 96.4 and 99.6. **A residual stayed open:** 3.8% at ε = 10⁻⁴ and 0.37% at ε = 10⁻⁶. **The website session supplied the shape.** The shortfall in natural time is **15.11 at ε = 10⁻⁴ and 15.19 at ε = 10⁻⁶** — the same absolute size while ε falls a hundredfold. So `π/√(aε)` is the leading term and what it omits is **ε-independent**, which is exactly why a 3.8% error becomes 0.37% when ε drops 100×. ## The exact passage, not the asymptote Near the saddle-node the flow is `ẋ = ε + a(x − x_c)²`. Passing from u₀ = −(x_c − x_start) to u₁ = x_end − x_c takes T = (1/√(aε)) · [ arctan(u₁√(a/ε)) + arctan(|u₀|√(a/ε)) ] As ε → 0 each arctan → π/2 and T → π/√(aε), the room's printed form. Expanding the arctans gives the ε-independent correction `−(1/a)(1/u₁ + 1/|u₀|)`, which is the 15. ## What is tested **Solve for the bottleneck half-width implied by each published time, and ask whether one width explains both.** Taking a symmetric window u₁ = |u₀| = w, the correction is `−2/(a·w)`, so w = 2 / (a · shortfall) **If the two ε give the same w, the residual is closed by one geometric quantity. If they give different w, it is not.** ## Criterion, fixed before the run 1. **The exact formula reproduces the published times.** With the w solved from ε = 10⁻⁴, the arctan formula divided by dt = 0.01 reproduces **both** 40,042 and 414,010 to within **0.5%** — that is, w is fitted on one point and must predict the other. 2. **One width, not two.** The w implied by each ε agree to within **5%**. 3. **The width is physically sensible.** w lies strictly inside the interval between the bottom and the ridge at h just below h_c, which entry 29 measured as merging at x_c = −0.664941 — so w must be well under 0.335, the distance from x_c to the ridge. **WHAT KILLS IT:** leg 2 failing. Then a single symmetric bottleneck does not explain both times and the residual stays open, which is what gets reported. **Prediction:** legs 1 and 2 hold and w comes out near 0.23. ⛔ **Leg 3 I am unsure of**, because the window the room's own simulation used is not stated anywhere I have found, so a sensible w is a consistency check rather than a confirmation. ## ⛔ What this cannot establish **What window the room's original run actually used.** Solving for w tells us what window would produce the published numbers; it does not tell us that window was used. If it lands sensibly that is evidence, not proof, and the entry says so. ## Discipline No prose conclusion in the script. Tables printed by the run. Hashes read from the file.