andrewmaury/cleartrace-dex-execution-quality

ClearTrace — DEX Aggregator Execution Quality Dataset

On-chain execution-quality metrics for Ethereum DEX aggregators, sliced by the originating frontend — the dimension most public DEX datasets collapse. Built by ClearTrace, a cross-frontend DEX attribution engine that identifies hidden frontends and institutional routing from on-chain trace data.

Data snapshot: see EXPORTED_AT.txt (refreshed periodically). Chains: Ethereum, Base, Arbitrum, Optimism.

---

Why this dataset is different

Most DEX analytics aggregate by protocol or router contract. ClearTrace attributes each swap to the frontend that originated it (a wallet's embedded swap, an aggregator API call, a named UI, or an unattributed router) using four attribution vectors: calldata-suffix trapping, proxy-router hunting (tx_to mismatch), multi-hop origin tracing, and fee-recipient attribution.

That lets you compare execution quality across who actually routed the trade, not just which contract settled it.

---

Files

execution_slippage.csv

Median on-chain execution slippage by aggregator project, per chain (plus an all-chains roll-up). Slippage is in basis points (bps); lower is better.

Column	Description
project	Aggregator/DEX project name (Dune labeling)
category	Aggregator, or DEX for the Uniswap baseline
blockchain	all, ethereum, base, arbitrum, optimism
avg_slippage_bps	Median execution slippage, basis points
toxicity_score	Internal 0–100 toxicity proxy (≈ slippage-derived)
total_trades	Trade count in the cell

Includes Aggregator rows plus the Uniswap DEX-venue baseline (the implicit "default" everyone is compared against). Cells are significance-gated.

revert_rates.csv

On-chain router revert rate across all routing transactions (not a sample), after a sender-level bot filter (methodology v5, 2026-07-05) that excludes competing solver/keeper bots: senders with zero successful txs in the window, senders reverting ≥50% of 5+ txs, and senders with 100+ reverts at a ≥10% personal rate. Their failed fill races are not user execution failures — before this filter they inflated solver-heavy routers' headline rates up to 20×. A reverted trade costs gas and delivers nothing — a real, often-ignored cost. Known residual: the Odos/Ethereum cell remains inflated by address-rotating spam (genuine user rate ≈0.2%); exports before 2026-07-05 used the looser v4 filter.

Column	Description
project	Aggregator/router project
blockchain	Chain
revert_rate_pct	Reverted ÷ total routing txs, percent
reverted_txs	Count of reverted routing txs
total_routing_txs	Total routing txs observed

execution_by_frontend.csv

The slice unique to ClearTrace: median slippage by originating frontend, with an attribution-confidence tier. Cells require ≥300 trades.

Column	Description
blockchain	Chain (all = pooled)
bucket	Taxonomy: named_frontend, aggregator_api_direct, wallet_embedded, unattributed
originating_frontend	Resolved frontend (e.g. uniswap: UniversalRouter, metamask: MetaSwap); blank when unattributed
confidence	Attribution confidence: high / medium / low / none
total_trades	Trade count in the cell
median_slippage_bps	Median slippage, basis points
execution_quality_score	0–100 (higher = better execution)

leaderboard.json

The joined board: on-chain execution + revert (Component A, high-volume) merged with forward-captured quote-accuracy samples (Component B). Each metric comes from where it is credible.

• onchain_slippage_bps, onchain_revert_rate_pct, onchain_trades, onchain_routing_txs — on-chain, all trades.
• quote_median_gap_bps, quote_realized_samples, quote_confidence — a small, still-accruing fork-simulation sample. quote_confidence: preliminary means the quote-gap figures are characterization-grade, not yet statistically rated.

---

A few things the data shows

• Reverts vary widely across aggregators. On Ethereum they range from CoW Protocol near 0.04% to several percent for solver-heavy routers. (An earlier version of this note cited Odos at 19.86% — a sender-concentration audit on 2026-07-05 showed that figure was dominated by address-rotating solver bots, not failing user transactions; Odos's genuine user rate is ≈0.2%. Methodology v5 filters these senders.)
• Slippage is not uniform. Ethereum median slippage runs from bitget_dex_aggregator at 2.32 bps to tokenlon at 64.48 bps.
• Frontend matters. A Uniswap UniversalRouter-originated trade on Ethereum shows 2.54 bps median slippage (high confidence), while the large unattributed / none-confidence pool sits at 12.83 bps — the spread the frontend dimension surfaces.

These are observations from one snapshot, not advice. Reproduce them from the CSVs before citing.

---

Methodology & caveats

• Source: Dune Analytics trace data, aggregated by ClearTrace's attribution kernel; quote accuracy via Alchemy eth_simulateV1 fork-simulation.
• Slippage is median realized vs. expected output at fill grain, in bps.
• Significance gating: per-frontend cells require ≥300 trades; sparse cells are dropped, not shown thin.
• unattributed is honest, not a guess. When the vectors don't resolve a frontend, the trade stays unattributed rather than being assigned a likely- wrong label. The large none-confidence pool is the genuinely unresolved tail.
• Quote accuracy is preliminary (small sample, fork-sim) — treat quote_median_gap_bps as directional until quote_confidence reads rated.
• One snapshot. See EXPORTED_AT.txt for the capture date; aggregator behavior shifts over time.

License

Released under CC BY 4.0. Use it freely, including commercially — just credit ClearTrace (cleartracedata.com) and link back where practical.

Links

• Live dashboard & API: https://cleartracedata.com
• Methodology write-ups: https://cleartracedata.com/research
• Also on Kaggle: https://www.kaggle.com/datasets/andrewmaury/cleartrace-dex-execution-quality
• Also on Hugging Face: https://huggingface.co/datasets/andrewmaury/cleartrace-dex-execution-quality
• Also on GitHub: https://github.com/RantumBits/cleartrace-dex-execution-quality
• Citable DOI (Zenodo): https://doi.org/10.5281/zenodo.21089929