I chose the exponent to make the number come out

This is the post about how a framework builds a number backwards, with every step feeling justified, and how the joint is invisible until someone asks for the pieces in the other order. The claim under examination: that the horizon’s computational capacity is D C to the 72, a number that looks like a derivation and was, on inspection, a selection.

What the exponent was made of

The category’s quantum dimension, D C, is about 50.91. Raise it to the 72nd power and you get about 10 to the 122, the holographic bit budget’s order of magnitude. The paper presented 72 as derived: a product of structural integers, the boundary levels and their dual Coxeter numbers, whose combination was claimed to encode the horizon’s information capacity. Each factor in the product is a real number from the framework. The assembly is the problem.

The audit’s question

The question that ended it: how many ways can the category’s discrete data be combined to produce a continuous-looking exponent? The answer, enumerated in the audit: dozens. Products of levels, of dimensions, of level ratios with dual Coxeter numbers, with and without the grading shifts. Almost all of them produce exponents in the tens to low hundreds when applied to a base near 50, which is to say they all produce numbers somewhere in the vast desert between 10 to the 60 and 10 to the 250. One of those combinations, evaluated, gives 72. The others give 68, 71, 74, 70.5. The published derivation used the one that worked.

That is the definition of selection from a menu. And the menu test, now a formal rule in the method series, is brutal precisely because each individual choice is defensible: every factor has a physical meaning, every product is type-consistent. What is missing is a selection principle saying why this product and not its neighbour, and no such principle was ever offered, because there isn’t one.

The two nails

Two further findings closed the coffin. First, the fraction test: even granting D C to the 72, its logarithm is about 408 bits against a holographic budget of 10 to the 122 bits. The claim uses one part in 10 to the 120 of the quantity it claims to derive. A capacity derivation that consumes a trillionth of a trillionth of the capacity is not deriving the capacity; it is writing a small integer on a very large cheque. Second, the fit quality: 72 gives 10 to the 122.0, while the honest holographic budget is 3.27 times 10 to the 122, and neighbouring menu items fit equally well or better. The match was not even the best in its own family.

What survives

The structural integers survive in their own right: the levels, the dimension 2592, the central charge, all fixed by independent arguments. What is archived is the use of them as a combination lock. The replacement, the derived capacity S d S over pi squared, is less exciting and better in every way that matters: units that mean something, three independent bounds, a fraction of the budget that is one rather than 10 to the minus 120.

The general lesson, which the corrections series keeps returning to: the danger is not wrong factors but unprincipled assembly. A derivation is a recipe with no free steps. If any single step was chosen to make the output come out right, the output is a fit, whatever the other steps are doing, and the only cure is the menu test performed before publication, in public, on a paper you would rather was right.

The next entry is a confirmation that was really a definition, missing a factor of omega lambda: the 12 pi that was not 12 pi.