This is a very sophisticated analysis, but it gets the comparison wrong for the question it is trying to answer. On-site generation needs to be compared with the mix of power infrastructure *that actually changes* when new load comes online. The average grid mix doesn't tell us that. A new combined-cycle plant is a useful benchmark, but it doesn't capture the full grid response either. Of course, the best comparison is a capacity expansion/dispatch model, but we'll save that for another day. For now, EPA eGRID non-baseload emissions give us another useful, and I'd argue better, benchmark. I redid the comparison on a generation-emissions basis, and the picture is much less one-sided. Under the assumptions shown, the same gas engine case is about 57% above the U.S. grid average but 12% below its nonbaseload benchmark. Different regions produce an even wider range of results. This chart excludes upstream fuel emissions, so it doesn't settle the lifecycle comparison. It shows how much the choice of grid benchmark changes the result. There are also questions about the project mix. Large combined-cycle capacity is under construction alongside planned data centers. See Homer City. That campus is grid-connected, so how it is classified matters. And the 2.3% national methane estimate isn't a measured leakage rate for each project's fuel supply. The analysis tests regional leakage rates. Those distinctions matter here, too.