Moving non-incremental innovations from the pilot scale to full commercial scale raises questions about the need and implementation of public support. Heuristics from the literature put policy makers in a dilemma between addressing a market failure and acknowledging a government failure: incentives for private investments in large scale demonstrations are weak (the valley of death) but the track record of governance in large demonstration projects is poor (the technology pork barrel). We reassess these arguments in the literature, particularly as to how they apply to sup- porting demonstration projects for decarbonizing industry. Conditions for the valley of death exist with: low appropriability, large chunky investments, unproven reliability, and uncertain future markets. We build a data set of 511 demonstration projects in nine technology areas and code characteristics for each project, including timing, motivations, and scale. We argue that the literature and the results from the case studies have five main implications for policy makers in making decisions about demonstration support. Policy makers should consider: 1) prioritizing learning, 2) iterative upscaling, 3) private sector engagement, 4) broad knowledge dissemination, and 5) making demand pull robust.