A map of 2 million scientific breakthroughs and their dependencies

Scientific progress builds incrementally—discoveries enable other discoveries. This work extracts detailed scientific contributions from 230,000 open-access papers in AI/NLP and links them to their prerequisites, creating the Scientific Contribution Graph. The authors introduce a prediction task: given existing technologies, which discoveries might they enable? Current models reach 0.48 MAP using temporally filtered backtesting, showing steady improvement. The dataset and task could support research impact assessment and accelerate scientific discovery.