I built a zero-dependency Python calculus toolkit (~500 lines) where every operation — derivatives, limits, integrals, residues, pole detection (and more) — reduces to reading coefficients on Laurent polynomial objects. One evaluation of f(a+h) gives you all derivatives simultaneously. Limits on black-box lambdas work in standard float64, even for high-order indeterminate forms. Alpha release, looking for feedback on the approach. AGPL-3.0.