RED-2400: A Public Benchmark of Algorithmically-Rejected Trading Events with Outcome Labels

RED-2400 is a public benchmark of 6,659 algorithmically-rejected trading events drawn from a live Solana decentralised-exchange filter stack, observed continuously between 2026-04-10 and 2026-05-02 (UTC). Each rejection event is linked to its post-rejection price and liquidity trajectory (169,122 outcome observations across 1,836 graveyard-tracker snapshots). Outcome labels follow the five-tier classification rule of Kamat (2026c, SSRN abstract_id 6638259, companion dataset Zenodo DOI 10.5281/zenodo.19987697): saved (windowed), saved (early-death), missed, flat, unclassifiable.

Filter labels are anonymised to filter_1 through filter_8 in descending order of rejection volume. Source-collector identifiers are anonymised to source_a and source_b. Liquidity and 24-hour volume are log2-binned to preserve heavy-tailed distributional shape while blocking threshold inference. Filter-internal model outputs (signal scores, weights, calibration thresholds) and the priceChange* derivatives are removed. SHA-256 checksums for the three CSV files are provided in checksums.txt.

This is the first window of a planned dataset series; subsequent windows will extend the time horizon and enable regime-stratified analysis.