## 1. Precise restatement **Setup and hidden assumptions:** - θ ranges over all of ℝ. - The reading is Y = θ + E, with E independent of θ, E having a fixed law, and E|E| < ∞. - A reporting scale f is *mean-calibrated* if E f(θ+E) = f(θ) for every θ. - The *calibration error* is sup_θ |E f(θ+E) − f(θ)|, a uniform sup-norm. - "Arbitrarily nearly linear" means: for any δ > 0, the two scales can be strictly increasing with ‖f − id‖∞ < δ and ‖f′ − 1‖∞ < δ. **Claim.** There exist two such scales f₁, f₂ with these properties: - **(A) Exact case.** Exact calibration on both scales implies E = 0 almost surely. - **(B) Tolerance case.** For every ε > 0 and every M, there is an error law with Var(E) > M whose calibration error is at most ε on both scales. In fact the error is 0 on f₁. ## 2. Step-by-step check and independent derivation **Scales.** Take f₁ = x + a sin x and f₂ = x + a sin(√2x). - The derivatives are f₁′ = 1 + a cos x ≥ 1 − |a| and f₂′ = 1 + √2a cos(√2x) ≥ 1 − √2|a|. - So both are strictly increasing for |a| < 1/√2. - The deviations satisfy |fⱼ − x| ≤ |a| and |fⱼ′ − 1| ≤ √2|a|. - Hence both scales are uniformly close to the identity in value and derivative. ✓ **Exact case.** Expanding the f₁ condition: E f₁(θ+E) − f₁(θ) = E[E] + a[ sinθ·(E cos E − 1) + cosθ·E sin E ]. - The functions 1, sin θ, cos θ are linearly independent, even on any interval. So the condition holding for all θ gives E[E] = 0, E[cos E] = 1 and E[sin E] = 0. - Since cos E ≤ 1, the equality E[cos E] = 1 gives cos E = 1 almost surely, i.e. E ∈ 2πℤ almost surely. - The same argument for f₂ gives cos(√2E) = 1 almost surely, i.e. E ∈ √2πℤ. - The two lattices meet only at 0: 2πm = √2πk forces k/m = √2 unless m = k = 0. - So E = 0 almost surely. ✓ The remark about convexity is a side comment. It presumably contrasts with the Jensen-type route, where a strictly convex scale forces degeneracy. It plays no role in the proof. **Tolerance case.** Take E = ±2πn with equal probability. - E[E] = 0, cos E = 1 and sin E = 0, so f₁ is exactly calibrated. - For f₂, the error is a[ sin(√2θ)(cos(2√2πn) − 1) + cos(√2θ)·E sin(√2E) ]. - By symmetry E sin(√2E) = 0. - So the sup error is |a|·|cos(2π√2n) − 1| = 2|a| sin²(π√2n) ≤ 2π²|a|·‖n√2‖², where ‖·‖ denotes distance to the nearest integer. - By Dirichlet's theorem, ‖n√2‖ < 1/n for infinitely many n. Hence the error tends to 0 along a sequence with n → ∞. - Meanwhile Var(E) = 4π²n² → ∞. ✓ Both parts re-derive correctly. ## 3. Attempts to break it **Commensurate frequencies.** If the frequencies were commensurate, e.g. sin x and sin 2x, exact calibration would only force E ∈ 2πℤ. Large noise would then remain possible. The claim says "can force", so this is consistent with it; irrationality of the frequency ratio is essential. **Large tolerances are trivial.** For any symmetric, zero-mean E, the calibration error is at most 2|a|. So tolerances of order a are vacuous anyway. The nontrivial content is for ε ≪ |a|, which is handled by the Diophantine approximation step. Still, "any positive tolerance" is true as stated. **Restricted θ.** If θ were restricted to a bounded interval, the coefficient matching still works, because 1, sin and cos remain linearly independent there. **Moments.** The finite-first-moment assumption is used for E[E]. The counterexample laws have finite variance. **Dependence on a.** The exact implication holds for every a ≠ 0, so there is also a discontinuity at a = 0. That is consistent with the claim. I found no gap. ## 4. Novelty Mathematically, part (A) is the classical fact about characteristic functions: if |φ(t)| = 1 at two incommensurable points t, the distribution is degenerate (lattice arguments; see Lukacs). Here φ(1) = φ(√2) = 1, which forces a point mass at 0. Part (B) is Dirichlet/Kronecker approximation of n√2 to integers. The measurement-theoretic framing, "calibration conventions impose perfect measurement only exactly", may be a new presentation. The content, however, is a direct application of known results. ## 5. Fairness of the summary The one-line claim is accurate: - It says "can force", an existence claim. - It specifies additive, state-independent error. - "Same scales" and "any positive tolerance" match the construction. Caveats it leaves implicit: - Tolerance is measured in sup-norm over θ. - The frequencies must be incommensurate. - Tolerances comparable to a are trivially satisfiable. None of these amounts to an overstatement. {"verdict": "correct", "novelty": "special case or variant of known results", "fair_summary": "yes", "confidence": 5, "one_line": "Both the exact-degeneracy argument (cosine/characteristic-function lattice intersection at incommensurable frequencies) and the Dirichlet-approximation counterexample with unbounded variance check out; the content is a measurement-theoretic repackaging of classical facts."}