OctoInformBeta

OpenAI preprint, September 2026 · §4. The exact correction and copy hierarchy · §4

Remark 4.6 [Order of the one-stage constants]: Natural numbers TT, KK, JJ, MM meeting the four one-stage conditions, TT chosen first, then KK and MM, then JJ

The one-stage constants of the copy hierarchy exist.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

One stage of the exact correction and copy hierarchy runs with four numbers fixed in advance: an iteration count TT, the number MM of copies of each vertex, the copies of a vertex xx being z+Mxz + Mx for 0≤z<M0 \le z < M, and two further integers KK and JJ. None of them depends on the input.

The analysis of a stage asks one condition of each, at the cutoff scale A=1/4096A = 1/4096, and the conditions fix an order of choice: TT first, by a condition on TT and AA alone, then KK and MM, each large compared with TT, and last JJ, large compared with MM and TT. The question is whether all four can be met together.

Scales and Parameter Constants

The analysis uses the cutoff scale AA, the constant δ\delta and the map η\eta from the reals to the reals given by A=14096,δ=A400=11638400,η(z)=z1000.A = \frac{1}{4096}, \qquad \delta = \frac{A}{400} = \frac{1}{1638400}, \qquad \eta(z) = \frac{z}{1000}.(1)

A positive integer TT is an admissible iteration count if (5/A)TT!≤11024.\frac{(5/A)^T}{T!} \le \frac{1}{1024}.(2)✦

The set 𝒞\mathcal{C} of one-stage parameter constants is the set of quadruples (T,K,J,M)(T, K, J, M) of natural numbers in which TT is an admissible iteration count and 128(1+32(T+1)η(δ))≤K+1,16(T+1)<M,64M(T+1)2≤δ(J+1).128\Bigl(1 + \frac{32(T+1)}{\eta(\delta)}\Bigr) \le K + 1, \qquad 16(T+1) < M, \qquad 64\,M\,(T+1)^2 \le \delta\,(J+1).(3)✦

The Claim

Theorem 1. There are one-stage parameter constants: natural numbers TT, KK, JJ, MM with (T,K,J,M)∈𝒞.(T, K, J, M) \in \mathcal{C}.(4)✦

What Is Checked

The claim is stated over the set 𝒞\mathcal{C} of (3), built from the cutoff scale AA, the constant δ\delta and the map η\eta of (1) and from the admissible iteration counts of (2).