swordfish: Efficient Forecasting of New Physics Searches without Monte Carlo

We introduce swordfish, a Monte-Carlo-free Python package to predict expected exclusion limits, the discovery reach and expected confidence contours for a large class of experiments relevant for particle- and astrophysics. The tool is applicable to any counting experiment, supports general correlated background uncertainties, and gives exact results in both the signal- and systematics-limited regimes. Instead of time-intensive Monte Carlo simulations and likelihood maximization, it internally utilizes new approximation methods that are built on information geometry. Out of the box, swordfish provides straightforward methods for accurately deriving many of the common sensitivity measures. In addition, it allows one to examine experimental abilities in great detail by employing the notion of information flux. This new concept generalizes signal-to-noise ratios to situations where background uncertainties and component mixing cannot be neglected. The user interface of swordfish is designed with ease-of-use in mind, which we demonstrate by providing typical examples from indirect and direct dark matter searches as jupyter notebooks.