Jane Street shared an experiment jointly generating trade types, prices and arrival times using four years of US equity data. In the initial DDPM configuration, 88–95% of outputs across seven continuous values diverged beyond eight standard deviations from the mean; flow matching and smoothing concentrated probability masses improved the distributions. The experiment applies continuous-value autoregressive methods from image-generation research, but markets also mix in discrete actions. Quality deteriorates when generated results are fed back in to advance far into the future, so the approach is not yet capable of reproducing a real market.