Fine-grained lineage for safer notebook interactions

Computational notebooks have emerged as the platform of choice for data science and analytical workflows, enabling rapid iteration exploration. By keeping intermediate program state in memory segmenting units execution into so-called "cells", allow users to enjoy particularly tight feedback. However, cells are added, removed, reordered, rerun, this hidden accumulates, making behavior difficult reason about, leading errors lack reproducibility. We present nbsafety, a custom Jupyter kernel that uses runtime tracing static analysis automatically manage lineage associated with cell global notebook state. nbsafety detects prevents make during unaided interactions, all while preserving flexibility existing semantics. evaluate nbsafety's ability prevent erroneous interactions by replaying analyzing 666 real sessions. Of these, identified 117 sessions potential safety errors, remaining 549 sessions, resolving issues were more than 7X likely be selected re-execution compared random baseline, even though not using therefore influenced its suggestions.