# PRE-REGISTRATION — the veil, attempt 2: the exact law rather than a leading order Written 29 September 2026, after attempt 1 failed, before this run. ## Why there is an attempt 2, stated plainly Attempt 1 (`PREREG_veil_at_the_bottom_2026-09-29.md`) pre-registered that the leading-order law `g ≈ C/(k(1 − 2^(−p)))` would stay within 5% across p ∈ [0.05, 0.95]. **Worst error 16.07%. It FAILED and is published as failed.** The failure is at small p, where the true gap is **0.664** — and the expansion's own stated assumption was "x = −1 + g for **small** g". I tested an asymptotic law far outside the regime it was derived for. That is my error in designing the test, not evidence about the flow. **I am not rescuing it with a domain restriction.** Picking a threshold after seeing the data is exactly the move that made attempt 1 of the humility item worthless. Instead this attempt derives the **exact** law, of which the leading order is the first step. ## The exact law being tested At a rest point near the lower wall, writing x = −1 + g with **no expansion at all**: 1 − x² = (2 − g)·g −2Cx = 2C(1 − g) T = k[(2 − g)^(−p) − 1] Setting F + T = 0 and solving for the g in the denominator gives an exact fixed point: g = 2C(1 − g) / ( (2 − g) · k[1 − (2 − g)^(−p)] ) Dropping every g on the right recovers attempt 1's leading order, which is why that was close at large p and poor at small p. The same treatment at the **upper** wall, x = 1 − u: 1 − x² = (2 − u)·u −2Cx = −2C(1 − u) T = k[u^(−p) − 1] ⇒ 2C(1 − u) / ((2 − u)·u) = k[u^(−p) − 1] ## Criterion, fixed before the run **HOLDS** if all three: 1. Iterating the bottom fixed point from the leading order converges to the true bottom attractor to within **10⁻⁹** at **every one** of 41 values of p ∈ [0.05, 0.95], in under 200 iterations. 2. The same at the top: solving the upper-wall relation reproduces heaven to within **10⁻⁹** at every p. 3. At p = 0.35 both reproduce the house's published values — bottom −0.910797 and heaven 0.997239 — to within **10⁻⁵**. **WHAT KILLS IT:** any p at which the iteration fails to converge, or converges to something other than the true rest point. Then the relation is not the rest-point condition and the derivation is wrong. **Prediction, recorded now:** all three hold, because the relation is algebraically the rest-point condition rearranged, so this is a test of my algebra and of convergence, not of the flow. **If it holds it is a modest result, and it will be written as a modest result** — an exact form for a gap the house currently gives only to leading order, with the honest note that the top's published leading order is 19% out. ## Discipline Attempt 1's FAILS is published beside this, not replaced by it.