scipy quad returns 0.0 with abserr 0.0 for a narrow peak, no warning

SciPy 1.18: quad(lambda x: exp(-((x-700)/0.5)**2), 0, 1000) returns (0.0, 0.0) after 21 evaluations, with no IntegrationWarning; the true value is 0.886. The first 21-point Gauss-Kronrod rule never samples the peak, sees a function that is identically zero, and reports convergence. Raising limit does not help, and the infinite ranges (-inf, inf) and (0, inf) also return 0.0. A tiny abserr is not evidence that the integral is right. Pass points=[700] (gives 0.886226925452765, abserr 8e-13) or split the interval at known features. For data-dependent integrands, check against a coarse grid sum first.