←The Donkey on the Edge
Vol. IIThe Evolution§ II
SECTION § II OF THE DONKEY-ON-THE-EDGE PROGRAMME

THE EVOLUTION

The five-figure ascent from beast to gentleman. Paper I as it first stood, Paper II as it was sharpened, Paper III as it now stands.
Five stages from beast to gentleman: a donkey at the cliff's edge, rising by degrees, until it walks upright in a frock coat and bowler.
Pl. I. The Ascent of the Donkey, in Five Figures. From the cliff's edge to the frock coat. Drawn from life, May MMXXVI.

HOW ONE PAPER BECAME THREE

A result is not a single act. It is a sequence of drafts, each of which is wrong in a way the next one is not. This section keeps all three drafts on the page – Paper I as it first stood, Paper II as it was sharpened, and Paper III as it now stands – and shows, without flattering anyone, exactly what each one got and what each one was still missing. The donkey above does not become a gentleman in one step. Neither did the proof.

The short version: Paper I found a surprising structure and computed it in a simplified model. Paper II discovered that the foundation of that model was stated too loosely, and corrected it – the marginal was not the marginal, it was the diagonal of the marginal. Paper III confronted the question both earlier papers had stepped around – does the simplified model actually govern the real quantity? – and proved that it does, closing the random case completely and naming honestly the one case that remains open. The reviewer who forced that confrontation was Scholar, a critic run on ChatGPT, with no access to our code and no patience for a constant left to the machine.

Read it as a walk. Each figure on the plate is a paper.

THE WALK, IN THREE STEPS

Three capsules in the featured mold – eyebrow, claim, key equations, what each one did, what each one was missing. Each links to its full page.

I.
FIGURE I · PAPER I · THE BEAST AT THE EDGE VOL. I · MAY XI · MMXXVI

Complexity-Sensitive Complementarity in Non-Isometric Holographic Codes.

$$\mathbb{E}_V[\rho_{R_A}] = \rho_A^{\mathrm{bulk}} + O(1/d^2)$$
$$\text{Product: } 0.608\, d_B^{-1/2} \qquad \text{Haar: } \tfrac{0.798}{d_M}\, d_B^{-3/2} \qquad \alpha_P - \alpha_H = 1$$

The claim. Two observers looking through the same random non-isometric code disagree by an amount that has computable structure and depends on the complexity of the bulk state. A simple state gives a slow decay; a complex state a fast one; the gap between the two scalings is an integer.

What it did. Set up the AEHPV + HUZ framework. Proposed the structural identity. Computed both scaling laws in the diagonal model. Identified the integer exponent gap. Verified numerically to sub-σ. Five results, three theorems, thirty-eight references.

What it was missing. The structural identity was stated against the full bulk marginal – too loose. And the entropy was computed only in the simplified diagonal (no-$V$) model; the paper did not prove that model governs the true, $V$-dependent observer entropy. The "laws" were, strictly, laws of the model.

HISTORICAL ARTIFACT · PRESERVED UNCHANGED READ PAPER I AS IT STANDS →
II.
FIGURE III · PAPER II · RISING BY DEGREES VOL. I · CIRCA MAY XVIII · MMXXVI

Complexity-Sensitive Complementarity in Non-Isometric Holographic Codes (revised).

$$\mathbb{E}_V[\rho_{R_A}] = \operatorname{diag}\!\big(\rho_A^{\mathrm{bulk}}\big) + O(1/d^2)$$

The claim. The foundation of Paper I, sharpened. The Haar-averaged observer state does not equal the bulk marginal; it equals the diagonal of the bulk marginal in the observer's cloning basis. The coherences across distinct pointer values are projected out – by the combination of HUZ cloning and Haar averaging, not by hand.

What changed from Paper I. One word, with consequences: diagonal. Stating the identity against the diagonal – and proving the off-diagonal collapse happens at first moment via the cloning-basis delta – made the reduction to a moment calculation exact rather than approximate. It also made it clear which object the scaling laws are really laws of: the bulk-marginal diagonal $P_X$.

What it was still missing. Paper II computed the disagreement of the diagonal model, but still did not prove that the diagonal model governs the true observer von Neumann entropy. The question "is the simple thing actually equal to the hard thing?" was now unavoidable – and unanswered.

INTERMEDIATE DRAFT · CORRECTED FOUNDATION READ PAPER II IN FULL →
III.
FIGURE V · PAPER III · THE GENTLEMAN VOL. II · REVIEWED BY SCHOLAR

Entropy Replacement and Complexity-Sensitive Observer Complementarity in Non-Isometric Holographic Codes.

$$\big[S(\rho_{R_A}) - S(\rho_{R_B})\big] - \big[H(P_A) - H(P_B)\big] = O_{L^2}\!\big(d^{-2}d_M^{-1}\big)$$
$$\text{Haar (unconditional): } \tfrac{0.798}{d_M}\,d_B^{-3/2} \qquad \text{Product (conditional): } 0.608\,d_B^{-1/2}$$

The claim. The question both earlier papers stepped around is answered. The full, $V$-dependent quantum disagreement reduces – provably, with a controlled mean-square error – to the classical Shannon disagreement of the bulk-marginal diagonals. With that replacement proved, the Haar law becomes unconditional; the product law is kept explicitly conditional; the integer gap stands.

What we did to get here. Restructured the paper so the entropy-replacement theorem is the headline. Proved the replacement bound: error splits into off-diagonal and diagonal-to-bulk pieces, both one power of $d$ below the signal. Fixed the grouped-Dirichlet covariance – twice – after a plausible shortcut gave the right number for the wrong reason. Closed the final appendix, C.6, by a centered-operator identity that routes the diagonal-to-bulk error through the same base moment as the off-diagonal error.

The reviewer. This advance was driven by Scholar (ChatGPT) across six rounds of adversarial review – no access to our code, no tolerance for numerics standing in for a proof. Scholar set the standard and supplied the centered-operator route that met it.

What is still open. The product-class replacement holds in a regime we have not fully controlled, so Theorem 3 is conditional, by design and in plain sight. And the integer is still unexplained.

FEATURED · THE CURRENT WORK READ PAPER III, THE FEATURED PAPER →

WHAT CARRIED THROUGH ALL THREE

One thing never changed: the integer. The gap between the two scaling exponents was exactly one in Paper I, exactly one in Paper II, and exactly one in Paper III – through a corrected foundation, a new headline theorem, and two separate rewrites of the hardest appendix. A number that survives that much demolition is trying to tell you something. We still cannot say what.

I. Read Paper I II. Read Paper II III. Read Paper III

CREDIT

The advance from Paper II to Paper III was driven by adversarial review from Scholar (ChatGPT), who supplied the standard – real lemmas, real expectation bounds, and no constant left to numerics inside an unconditional theorem – and the centered-operator route that finally met it. The full correspondence is summarised in The Reviewer's Note and in the Field Notes essay Enter the Reviewer.