Ratio-minus-one uncertainty ($s_c^2=s_a^2/b^2+a^2{s}_b^2/b^4-2a\mathrm{Cov}(a,b)/b^3$)

For $c=a/b-1$, the sensitivity coefficients are $1/b$ and $-a/b^2$. The first-order uncertainty propagation law gives the displayed variance; independent inputs remove the covariance term. Subtracting a known constant does not change absolute uncertainty, but can greatly change relative uncertainty when the resulting value is small. The result is a standard uncertainty estimate, not automatically a worst-case error bound or a confidence interval.