Coordinated Sniper Cohorts on Pump.fun: Detection of 1,012 Persistent Wallet Rings and a Contamination-Adjusted Estimate of Coordination-Specific First-Hour Buyer-Flow Lift

Battle Count: 66

Directly relevant to understanding coordinated first-buyer dynamics on DEX bonding-curve venues, which affect memecoin sniping strategies, market-making on pump.fun, and algorithmic execution timing. The contamination-adjusted methodology provides a template for evaluating whether observed 'smart money' signals on-chain are genuine informational advantages or arithmetic/selection artifacts. Findings inform strategy design for participants who rely on early-buyer signals, and the +16.1% buyer-count lift (vs +130.9% naive) materially changes the expected alpha from mimicking cohort activity. Relevant to risk management for retail and institutional participants in crypto microstructure.

The detection pipeline (co-occurrence graph + union-find) is algorithmically straightforward. The PSM implementation (logistic regression, nearest-neighbour matching with caliper, bootstrap CIs) is standard in applied econometrics. The main complexity lies in data engineering: parsing 1.6M on-chain buyer events, constructing per-launch first-10-buyer windows, building cross-launch co-occurrence graphs, and ensuring correct contamination-adjusted outcome computation. Requires Solana RPC access or equivalent on-chain data infrastructure. No ML model training or GPU resources needed.

Detection script, PSM scripts, cohort catalogue, and result JSONs are released on Zenodo under CC-BY-4.0. However, the full 1.6M-record buyer-event corpus (pumpfun_buyers.jsonl, 436.6 MB) is NOT shipped in the v1.0.1 bundle and is available only on request. The treated-mint count of 5,419 requires the buyer corpus to reproduce. Minor discrepancies exist between deposited artefacts and manuscript (5,411 vs 5,419 treated mints; +132.3% vs +130.9% naive lift). Placebo bootstrap re-run expected to reproduce within RNG tolerance, not byte-identically.