## 1. Precise statement and assumptions The claim is correct under the following interpretation. - The true state \(\theta\) ranges over \(\mathbb R\). - The instrument reports \(Y=\theta+E\), where the **same probability distribution** of \(E\) applies at every \(\theta\). - The noise has a finite absolute first moment: \(\mathbb E|E|<\infty\). A finite variance is not assumed. - Choose a fixed, **nonzero** amplitude \[ 00\)” means the **absolute, uniform transformed-mean error** \[ \max_{j=1,2}\sup_{\theta\in\mathbb R} \left|\mathbb E[f_j(\theta+E)]-f_j(\theta)\right| \le \varepsilon. \] The precise conclusions are: 1. Exact calibration on both scales implies \(E=0\) almost surely. 2. For these **same fixed scales**, for every \(\varepsilon>0\) and every finite variance threshold \(V\), there exists centered, finite-support noise satisfying the tolerance condition and having \(\operatorname{Var}(E)>V\). The scales really can be arbitrarily close to identity in the stated sense: \[ \sup_x|f_j(x)-x|\le a, \qquad \max_j\sup_x|f'_j(x)-1|\le \sqrt2 a. \] Also, \(f'_j(x)\ge 1-\sqrt2 a>0\), so both are strictly increasing. ## 2. Check of the argument, followed by an independent derivation ### Checking the supplied argument For \(t\in\{1,\sqrt2\}\), write \[ f_t(x)=x+a\sin(tx), \quad m=\mathbb E E,\quad c_t=\mathbb E\cos(tE),\quad s_t=\mathbb E\sin(tE). \] The calibration residual is \[ R_t(\theta) =m+a(c_t-1)\sin(t\theta)+a s_t\cos(t\theta). \] This formula is valid: the first-moment assumption handles the linear term, and the trigonometric terms are bounded. Because \(1,\sin(t\theta),\cos(t\theta)\) are linearly independent as functions of \(\theta\), exact calibration gives \[ m=0,\qquad c_t=1,\qquad s_t=0. \] The supporting text omits the sine expectations, but that omission causes no problem: they are not needed for the conclusion. Next, \[ \mathbb E[1-\cos(tE)]=0. \] The integrand is nonnegative, so it vanishes almost surely. Hence \[ E\in \frac{2\pi}{t}\mathbb Z \quad\text{almost surely}. \] For the two frequencies, this means \[ E\in 2\pi\mathbb Z\cap \sqrt2\pi\mathbb Z. \] That intersection is indeed \(\{0\}\), by irrationality of \(\sqrt2\). For the approximate construction \(E=\pm2\pi n\), with equal probabilities, the stated residuals are also correct: \[ R_1(\theta)=0, \] and \[ R_{\sqrt2}(\theta) =a\sin(\sqrt2\theta)\bigl[\cos(2\pi\sqrt2 n)-1\bigr]. \] Consequently, \[ \sup_\theta|R_{\sqrt2}(\theta)| =a\left|\cos(2\pi\sqrt2 n)-1\right|. \] The only arithmetic assertion left implicit is that arbitrarily large suitable \(n\) exist. That is true, as shown explicitly below. ### Independent proof of the exact conclusion One can avoid coefficient comparison entirely. Fix either frequency \(t\), and evaluate calibration at \[ \theta_+=\frac{\pi}{2t}, \qquad \theta_-=-\frac{\pi}{2t}. \] Since \[ \sin(t(\theta_++E))=\cos(tE),\qquad \sin(t(\theta_-+E))=-\cos(tE), \] the two calibration equations become \[ m+a\,\mathbb E\cos(tE)=a, \] \[ m-a\,\mathbb E\cos(tE)=-a. \] Adding and subtracting gives \[ m=0,\qquad \mathbb E\cos(tE)=1. \] Doing this for \(t=1\) and \(t=\sqrt2\) forces \(E\) onto the two lattices above, whose intersection is \(\{0\}\). Thus \(E=0\) almost surely. In fact, exact calibration at these four particular state values already suffices; calibration at every real state is stronger than necessary. ### Independent proof of the tolerance conclusion Use the integer solutions of Pell’s equation \[ p_k^2-2q_k^2=1, \qquad p_k+q_k\sqrt2=(3+2\sqrt2)^k. \] Here \(q_k\to\infty\), and \[ \delta_k=p_k-\sqrt2 q_k =\frac{1}{p_k+\sqrt2 q_k}\longrightarrow0. \] Let \(E_k\) equal \(\pm2\pi q_k\), each with probability \(1/2\). Calibration on \(f_1\) is exact. On \(f_2\), \[ \begin{aligned} \sup_\theta|R_{\sqrt2}(\theta)| &=a\bigl[1-\cos(2\pi\sqrt2 q_k)\bigr]\\ &=a\bigl[1-\cos(2\pi\delta_k)\bigr]\\ &\le 2\pi^2a\,\delta_k^2 \longrightarrow0. \end{aligned} \] Meanwhile, \[ \operatorname{Var}(E_k)=4\pi^2q_k^2\longrightarrow\infty. \] This proves the required quantifiers: for any positive tolerance and any variance threshold, choose sufficiently large \(k\). ## 3. Attempts to break the result The theorem survives scrutiny, but several conditions matter. ### Zero amplitude would invalidate the exact conclusion If \(a=0\), both scales are identity. Exact calibration then requires only \(\mathbb E E=0\), not zero noise. Thus “sufficiently small” must mean **small and nonzero**. ### Incommensurate frequencies are essential to this construction Replacing the second scale by \(x+a\sin(2x)\) would not work: centered noise \(E=\pm2\pi\) would be exactly calibrated on both scales. Each periodic perturbation individually allows noise on a nontrivial lattice. The irrational frequency ratio eliminates the common nonzero lattice points. ### Calibration at arbitrary isolated states is not enough For example, at \(\theta=0\), any symmetric integrable noise is exactly calibrated on both scales, because both reporting functions are odd. The stated all-state assumption is sufficient, although—as shown above—particular finite sets of states can also suffice. ### The tolerance concerns means, not individual measurement accuracy In the approximate examples, \[ |Y-\theta|=2\pi q_k \] with probability one. Individual readings are extremely inaccurate even while the transformed means are uniformly well calibrated. This is not a contradiction. The calibration criterion measures bias after transformation, not absolute error, mean squared error, concentration, or closeness of distributions. ### A numerical check Take \(a=0.1\) and the Pell pair \((p,q)=(99,70)\). With \(E=\pm140\pi\), \[ \sup_\theta|R_{\sqrt2}(\theta)| \le \frac{2\pi^2(0.1)}{(99+70\sqrt2)^2} <5.04\times10^{-5}, \] while \[ \operatorname{Var}(E)=19600\pi^2\approx193{,}444. \] Calibration on \(f_1\) remains exact. There is therefore no missing mathematical step: the result has a complete proof. ## 4. Novelty The mechanism is a combination of standard facts: 1. **Characteristic-function lattice characterization.** For a real random variable \(E\), \[ \mathbb E e^{itE}=1 \quad\Longleftrightarrow\quad E\in(2\pi/t)\mathbb Z\ \text{almost surely}. \] The proof is precisely the nonnegative-cosine argument used here. Two incommensurate frequencies have only the zero common lattice point. 2. **Diophantine approximation and recurrence of irrational rotations.** Arbitrarily large integers \(n\) with \(\sqrt2 n\) close to an integer follow from classical approximation results. Pell’s equation gives an explicit sequence in this particular case. 3. **Fourier analysis of convolution.** Averaging a sinusoid under additive noise multiplies its complex exponential by the noise characteristic function. That is the basic calculation behind the calibration constraints. I would classify this as a **special case or variant of known results**, potentially with a useful measurement-theoretic presentation. These observations do not establish whether this exact formulation has previously appeared. ## 5. Fairness of the one-sentence claim The sentence fairly summarizes what is proved: - “can force” correctly describes an existence construction, not a claim about all nearly linear scales; - “arbitrarily nearly linear” is justified in uniform value and derivative; - “those same scales” is correct: the amplitude need not shrink as the tolerance shrinks; - arbitrarily large variance is achieved by finite-support noise, so it does not depend on pathological infinite-variance distributions. The main interpretive qualification is that “calibration” means **uniform transformed-mean unbiasedness**, not a general guarantee of measurement quality. The claim explicitly says “mean calibration,” so this is a scope clarification rather than an overstatement. Also, the unbounded-variance conclusion is not a discontinuity of variance under a specified convergence of noise laws; it says these two residual constraints alone provide no finite variance bound. {"verdict":"correct","novelty":"special case or variant of known results","fair_summary":"yes","confidence":5,"one_line":"For fixed nonzero sinusoidal perturbations with incommensurate frequencies, exact transformed-mean calibration forces zero noise, while Pell approximations yield uniformly vanishing calibration errors with diverging finite variances."}