Natural selection, the central explanatory concept of evolutionary theory, is usually considered to be differential reproduction due to differential fitness.  On this view, exactly what explanation by natural selection amounts to turns on exactly what, if anything, fitness is.  The propensity interpretation of fitness (PIF) defines fitness in terms of objective probabilistic dispositions known as "propensities", in particular propensities for an organism to have various numbers of offspring.  Despite its popularity, the PIF is controversial, in part for its (usual) assumption that biological processes are significantly affected by fundamental indeterminism.

I describe an alternative to the PIF, the mechanistic conception of fitness, which defines fitness in terms of a new objective interpretation of probability, mechanistic probability.  Mechanistic fitnesses depend on the causal structure of what I call a "causal map", which is defined by properties of an organism and its environment, and on certain general facts about similar organisms in similar environments.  This makes fitness makes consistent with either fundamental determinism or indeterminism, and plausibly makes fitnesses multiply-realizable properties.  Although mechanistic fitnesses typically have imprecise rather than precise values, I argue that they can often be modeled with real numbers.  My account of mechanistic fitness includes a proposal for a quite general solution to the heretofore unresolved problem of delineating the temporal and spatial extent of the environment which contributes to fitness.

The paper provides a rough preliminary sketch of mechanistic probability--in particular the version I call "far-flung frequency (FFF) mechanistic probability".  This should be of interest to those in the market for an objective interpretation of probability and those interested in realist interpretations of probabilistic generalizations.  The mechanistic interpretation of probability makes use of Strevens' (2003) concept of "microconstancy" and related ideas.  As Strevens makes clear, microconstancy and concepts based on it do not themselves justify an interpretation of probability, and Strevens stops short of providing one.  FFF mechanistic probability shows how microconstancy can be used as part of an objective interpretation of probability.  While the basic form of FFF mechanistic probability should satisfy standard Kolmogorov axioms, by making use of a generalization of microconstancy ("bubbliness"), FFF mechanistic probability also supports objective interval-valued probabilities.  I use the latter as the basis of imprecise fitness.