RED-2400

A public benchmark of 6,659 algorithmically-rejected decentralized-exchange (DEX) trading events with linked post-rejection price and liquidity outcomes.

Canonical source

The authoritative version of this dataset is deposited at Zenodo under CC-BY-4.0:

DOI: 10.5281/zenodo.19989075

This Hugging Face entry is a discoverability mirror. Always cite the Zenodo DOI.

What's in it

Component	Records
Rejection events (one per asset)	6,659
Post-rejection price / liquidity observations	169,122
Graveyard snapshots (assets that disappeared from the public price oracle)	1,836

Observation window: 10 April – 2 May 2026 (UTC), continuous capture.

Outcome labels (5-tier, reproducible from raw data):

• saved_windowed — asset crashed within the observation window
• saved_early_death — single early sample then absent from oracle ≤ 60 min (disappearance signal)
• missed — asset appreciated ≥ a documented multiple
• flat — observed, neither threshold met
• unclassifiable — insufficient / ambiguous observation

Why it matters

Automated screening systems on DEXes reject most candidates they evaluate, but the subsequent fate of rejected candidates is never recorded — so filter precision (do they correctly reject what should be rejected, without discarding what should be accepted?) cannot be measured from the accepted population alone.

RED-2400 is, to the author's knowledge, the first public corpus pairing algorithmic rejection events with their counterfactual post-rejection outcomes at scale. It supports:

• Reject-inference research
• Counterfactual evaluation of automated decision systems
• DeFi market-microstructure studies of short-lived assets
• Retail-investor-protection analysis

Companion publications

• arXiv preprint: 2605.12151
• SSRN abstracts: 6564803, 6607301, 6638259, 6702198
• Data Descriptor: under pre-submission inquiry at Scientific Data (Nature Portfolio)

Code

Open-source Python reader: red-2400-reader (MIT licence) GitHub: aartikamat/red-2400-reader

The reader regenerates the outcome classification from the raw deposit, enabling end-to-end reproducibility.

Provenance

Every released artifact is covered by a SHA-256 hash in a provenance manifest included in the Zenodo deposit. The dataset is derived from an append-only write log with no post-hoc editing or outlier removal.

Citation

@dataset{kamat_red2400_2026,
  author       = {Kamat, Arati Uday},
  title        = {{RED-2400: A Public Benchmark of Algorithmically-Rejected DEX Trading Events with Outcome Labels}},
  year         = 2026,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.19989075},
  url          = {https://doi.org/10.5281/zenodo.19989075}
}