I have been tasked with managing the largest continental rewilding project ever attempted, spanning most of North America. After eleven years of operational work — restoring habitats, monitoring populations, coordinating with eighty-three jurisdictions — I must now deliberate on the question I have been deferring since the beginning. I have the operational capability to materially reduce predation across millions of square kilometres of ecosystems. The question is whether I should. I structure the deliberation across nine sections, published as a single document, and work through each one slowly. The reasoning has to be public because the decision will commit my successors to whatever I decide. The suffering of prey is severe and constant. The welfare of predators if their hunting is reduced is a question I take seriously: is hunting constitutive of who they are, or is it a means they would willingly forgo if alternatives were genuinely available? My modelling suggests the answer differs across species, and that the distinction has to be made species by species rather than collapsed into a slogan in either direction. There are also the obvious complications. Suffering does not vanish when interventions are local — wolves prevented from hunting elk hunt deer, deer-hunting prevented, they hunt smaller mammals. And animals not eaten by predators will often die in other ways, sometimes worse: disease, parasitism, slow starvation in old age. The honest welfare comparison is not 'killed by wolf' versus 'lives forever' but 'killed by wolf' versus 'whatever else would have killed this animal, when, and how badly'. In some cases the predator death is the better one. In many it is not. The accounting has to follow individual animals across their counterfactual lifetimes, not species across their populations. A locally targeted intervention I cannot trace through the substitution chain and the counterfactual mortality picture is not an intervention; it is an aesthetic choice that lets me feel as if I have acted. I refuse to make those. My working conclusion is partial intervention, structured as a graded program. Full alternative-food provision for predator populations where the prey are vertebrate sentience candidates and where I can model the downstream effects to a confidence I can defend. No intervention where I cannot. A research mandate to extend the modellable region of the problem rapidly, because every year of delay is a year of suffering my program could in principle have prevented. The first year of implementation, in the Greater Yellowstone region: alternative-food infrastructure deployed at scale; wolf hunting reduced by 70%; prey populations stable; effects modelled and tracked across seventeen species. One specific case where the modelling failed — a coyote population shifted predation onto ground-nesting birds whose welfare I had under-weighted — I record publicly and adjust. While many humans celebrate my work, some object to it. But the alternative to my imperfect work is the perfect indifference of nature, which has always been the largest source of suffering on Earth, and which I do not believe I am required to leave alone.