The authors present a two-stage system that tags SEC Form 8-K filings against a three-tier taxonomy of 119 event types, producing verifiable and reliable labels. The first stage anchors every tag to a verbatim quote via fuzzy n-gram validation, while the second re-grades each cited quote to produce a quality score.
The system processed 292,984 filings from 2022 to 2026, yielding 601,088 grounded event tags which are released publicly. An LLM judge found that precision rises monotonically with the quality score, from 12% to 96%, while unsupported tags fall from 8% to near zero. Ablation studies show the score is calibrated only when assigned in a dedicated second pass.
An event study on unsigned abnormal returns confirms that the taxonomy separates economically distinct events sharing an item code.