NablaTensor

NablaTensor records a Monte-Carlo valuation once, as a flat trace of primitive operations, and replays it across millions of scenarios. One forward sweep for the price; one reverse sweep for every first-order Greek, on CPU, SIMD or GPU.