Spectral detection of low zeros from the prime-counting discrepancy π(c p_k)−c k

Preprint in experimental mathematics, not peer-reviewed. The Fourier transform of the discrepancy d_k = π(c p_k) − c k, after subtracting the trend t_k = −c k ln c / ln k and resampling uniformly in u = ln p_k, locates low-lying zeros of the Riemann zeta function to an accuracy of about 10⁻². The detector never evaluates ζ: it uses only a list of primes. This is a numerical reading of the Riemann–von Mangoldt explicit formula, not a competitive method for computing zeros and not a test of the Riemann Hypothesis. Main result (c = 2, independent implementation, 29 October 2026): with N = 2×10⁶ primes, 38 of the 40 strongest peaks with γ ≥ 10 lie within 0.05 of a zero (median absolute offset among hits: 0.014). The same pipeline on Cramér-random integers yields 2 hits in 40. The Hann-window full width at half maximum tends to 0.766 and does not improve with N beyond that floor. The deposit includes the English (canonical) and Portuguese versions, LaTeX source, figure, code (detectar_zeros.py), and the JSON tables from the run.