Forecasting Scholarly Impact: A Data Pipeline of Ecosystem-Wide Article Events

A paper's impact accumulates from the moment it becomes public, well before venue decisions. In this preliminary work, we propose ongoing work on a pipeline that records the community's engagement with a paper as a stream of timestamped contextual events, which can then be used to study the predictability of impact based on different sources at different points in time. We instantiate the pipeline for the ICLR 2025 and 2026 cohorts, finding that community signals available seven days after release can identify nearly half of top-cited papers six months down the line.