# PREREG — Rushmore, round 3: ten more against the house (30 September 2026, before any run) His words: *"Do Rushmore round 3, when you stuck ask questions like i ask play with concepts like i play with them"*. Labels as in round 2: ALGEBRA (an identity), THEOREM CHECK (a known theorem run on the house's quantities), FALSIFIABLE (could come out either way). Only FALSIFIABLE legs are evidence. Monte Carlo criteria at 4 SE. A "vote" is a yes/no that is right (1 + x)/2 of the time; x is its alignment. **R1 · Laplace, the rule of succession (1774) → Position's third dimension, faith as the prior.** With faith b₀ held at strength κ (pseudo-readings), position is x̂ₙ = (κb₀ + Sₙ)/(κ + n). - 1a ALGEBRA: equals the Beta–Bernoulli posterior mean by quadrature, < 1e-10. - 1b THEOREM CHECK: its mean squared error is [n(1 − x²) + κ²(b₀ − x)²]/(κ + n)². Monte Carlo at x = 0.4, κ = 10, b₀ ∈ {0.4, −0.4}, n ∈ {5, 20, 100}, within 4 SE. **R2 · Kelly (1956) → the Ledger and Testimony: what to stake.** Even-money acts, right (1 + x)/2 of the time, staking a fraction f of value: growth g(f) = ((1+x)/2)ln(1+f) + ((1−x)/2)ln(1−f). - 2a ALGEBRA: argmax g = x to 1e-6, and g(x) = ln 2 − H((1+x)/2) (nats), < 1e-12, at x ∈ {0.1, 0.3, 0.6}. - 2b THEOREM CHECK: 200,000 simulated acts at f ∈ {x/2, x, 3x/2, 2x} match g(f) within 4 SE. - 2c FALSIFIABLE (predicted): staking twice the alignment shrinks value: g(2x) < 0 at every x tested. **R3 · Lundberg's ruin (1903) → The Ledger / The Silent Sky.** Ledger increments X = R·max(0, c) − (1 − R)K − R·b₀·max(0, −c), c = cos θ, θ von Mises around the reference (κ_a = 1), R = 0.95, b₀ = 0.3, K set so E[X] = 0.05. Adjustment coefficient ρ from E[e^{−ρX}] = 1. - 3a THEOREM CHECK: simulated ruin ψ(V₀) (20,000 ledgers, 3,000 cycles) ≤ e^{−ρV₀} + 4 SE for V₀ ∈ {0.5, 1, 2, 3}. - 3b THEOREM CHECK: the slope of ln ψ against V₀ is within 12% of −ρ. **R4 · Little's law (1961) → Pressure.** Arrivals at rate P = 2, each held a time of mean 1.5 drawn from four laws (exponential, fixed, uniform on [0, 3], lognormal σ = 1). Held amount, averaged over time T = 200,000. - 4a THEOREM CHECK: within 2% of P × 1.5 = 3 under every law. Pressure's Π* = P/r holds for any release law. **R5 · Fourier and Laplace's spherical harmonics (1782, 1807) → Drift's ladder (entry 41, still OPEN).** Brownian motion on the sphere of aims, an independent small-step integrator (tangent Gaussian step, renormalised), D = 1, from the pole, 40,000 walkers. - 5a FALSIFIABLE: the l = 1 harmonic ⟨v₁⟩ decays at (d − 1)D within 2%, d ∈ {3, 5}. - 5b FALSIFIABLE: the l = 2 harmonic ⟨v₁²⟩ − 1/d decays at l(l + d − 2)D = 2dD within 3%, d ∈ {3, 5}. If both hold, entry 41's ladder is closed by a second integrator. **R6 · Jaynes' maximum entropy (1957) → The Measure.** On 400 grid points with the sphere's base measure (1 − x²)^((d−3)/2), d = 5, and the constraint ⟨x⟩ = 0.5: - 6a THEOREM CHECK: a general optimiser's maximum-entropy distribution equals the exponential family base·e^{βx}/Z to 1e-5. Reported: the dimension the house's space acts in at the Climb Out's temperature, d = 3 + 2C/D. **R7 · Nyquist–Shannon (1928, 1949) → The Two Clocks and a changing aim.** An aim cos(2πt), read s times per unit of body time, s ∈ {3, 2.5, 1.5, 1.2, 0.8}, over T = 400. - 7a THEOREM CHECK: the reading's apparent rhythm (FFT peak) equals |1 − s·round(1/s)| within one bin, every s. **R8 · The 37% rule (Lindley 1961, Dynkin 1963) → the choice made before knowing.** 100 options in random order; look at the first r, then commit to the first better than all of them. - 8a ALGEBRA: exact P(best) = (r/n)Σ_{i=r+1}^{n} 1/(i − 1) peaks at r = 37 with P = 0.3710 (to 4 places). - 8b THEOREM CHECK: 200,000 simulated searches at r = 37 match it within 4 SE. **R9 · Doob's martingale (1953) → belief, the record.** A mind hearing votes (x = 0.3) about a yes/no truth drawn from ½–½; its posterior pₙ. - 9a THEOREM CHECK: E[pₙ₊₁ − pₙ | pₙ] = 0 within 4 SE, in each decile of pₙ, over n ≤ 50, 200,000 minds. - 9b REPORTED: the share within 0.01 of 0 or 1 by n = 50. **R10 · Wald's sequential test (1945) → Testimony: when to stop listening.** Votes of x = 0.2; stop when the Einstein-combined alignment of all votes heard reaches ±0.9 (half log-odds ±artanh 0.9 = ±½ ln 19, i.e. 5% error). - 10a FALSIFIABLE (predicted): error ≤ 0.055, and the average number of votes heard is ≤ 0.7 × the smallest fixed odd count whose majority errs ≤ 5% (exact binomial). 100,000 runs. Kill conditions as before: a failed ALGEBRA or THEOREM leg is my defect and is diagnosed first; a re-run with changed legs gets its own addendum.