The Cost of Distinction: Least Cost Fixes the Unit on the Countable Carrier

The size of the cost of a distinction need not be stipulated. A composition law on positive ratios forces the shape of the recognition cost to be J(x) = ½(x + x⁻¹) − 1 and fixes nothing about its scale, since every F_λ(x) = J(x^λ) satisfies the same law; selecting λ = 1 has been done by hand, either as a calibration on the completed line or as a value anchored at one ratio on the countable carrier, and that stipulation is the reason the honest description of the framework has been parameter-free up to a choice of unit. We replace the number with a principle, and on the countable carrier the principle is enough: the cheapest cost that charges anything at all is J, strictly so at every positive ratio, with no numeric value assumed anywhere.

Enough is a strong word and it has an exact price, three things stated openly rather than buried. The first is a selection postulate, that among admissible costs one takes the cheapest pointwise; we assume it and do not derive it. The second is a nondegeneracy condition, that the cost prices some distinction; without it the postulate selects a cost that prices none. The third is the six exponentials theorem, a published transcendence result that we use as a named hypothesis and do not prove, and on which the classification of the available scales rests. Trading a number for a principle plus two conditions plus one import is a gain only if the conditions are structural and the import is respectable, and the body of the paper is largely an argument that both hold.

The classification is what turns the postulate from a search into a selection. On the countable carrier the admissible scales are not a continuum: there is exactly one cost per nonnegative integer exponent, the character behind it being the sign times an absolute power, and nothing else. That is a theorem here, with the six exponentials input as its only assumption; the other thing the argument leaned on, Erdős's theorem that a monotone completely multiplicative function is a power, is elementary in the case we need and we formalized it rather than importing it. Among that family J is the strict pointwise least element on positive ratios once the exponent zero is set aside. The anchor condition and genuine leastness over the family turn out to hold of exactly the same member, at every anchor base, so the numeric stipulation was a global extremal condition written at a point, and the point was immaterial.

The word nondegenerate is not decoration, and we add it here having discovered we needed it. The exponent zero is a cost that charges nothing at every positive ratio, and we prove that every inhabitant of the ledger is nonnegative there, so that cost is the ledger's least element outright. Applied to the ledger as stated the postulate therefore selects a cost that prices no distinction at all. That is the same degeneracy we diagnose on the completed line, and the carrier does not escape it. What survives the correction is the difference between the two carriers, and it is still sharp: exclude the degenerate member on both sides, and the carrier's family has a floor while the line's has none, because halving the scale always undercuts.

We then correct the record on the residual gap, because this program has now been wrong about it four times. The obstruction was a demand that the multiplicative character behind a cost be valued in the carrier: it need not be, since r + r⁻¹ = 3 has no rational solution while the cost it defines is perfectly rational, its character being the square of the golden ratio. What a cost exposes is the trace, not the character. Under that weaker and correct hypothesis the exponent step goes through: we prove that a real above one with rational trace and any rational power is itself rational, hence that a rational exponent with rational trace is an integer, and the irrational case follows from the six exponentials theorem on three bases. The 1944 question of Alaoglu and Erdős is the two-base statement and is not the one needed here. The factorization step, which the first draft of this paper called the one remaining lemma, is now a theorem: every inhabitant of the anchor-free ledger factors through a real-valued completely multiplicative character, given by an explicit formula in three values of the cost, with monotonicity on the positive integers doing the work continuity does on the line. The companion paper carries that proof.

The two things this program thought it knew about the list of scales were both wrong. The exponent zero was missing, and so was every even exponent, both because the family had been described by the parameterization q → J(q^n) instead of by its character. Orientation reversal, which had been credited with excluding the even exponents, constrains the character at −1 and nowhere else. We reprove the selection over the corrected family, where it holds unchanged.