RED-COHORT-2026-v1: A Catalogue of 1,012 Persistent Wallet Cohorts Detected on the Solana Pump.fun Bonding-Curve Marketplace (June 12-26, 2026)

RED-COHORT-2026-v1 is a public, reproducibility-grade dataset of 1,012 persistentsniper-cohort detections on the Solana pump.fun bonding-curve marketplace, derivedfrom 1,578,333 buyer events across 166,098 token launches observed between2026-06-12 and 2026-06-26 (15 days continuous). The release accompanies the working paper: Kamat, A. U. (2026). "Coordinated Sniper Cohorts on Pump.fun: Detection of 1,012 Persistent Wallet Rings and the Limits of Naive Causal Inference for First-Hour Buyer Flow." Working paper (arXiv + SSRN, in submission). Contents (18 files): - sniper_cohorts.jsonl the 1,012-cohort catalogue, JSONL, one cohort per line - sniper_cohorts_intra.jsonl.gz raw intra-launch first-buyer-window observations, gzipped JSONL - sniper_cohorts_report.md human-readable detection summary - analyze_sniper_cohorts.py two-stage detection script (Python 3.10+, MIT license) - gen_p7_artifacts.py generator for the Section 6 causal buyer-flow analysis - table1_top10_cohorts.csv top 10 cohorts by score - table2_size_distribution.csv cohort size distribution - table3_descriptive_stats.csv headline descriptive statistics - appendix_a_ablations.csv detection edge-weight cutoff ablation (cutoffs 2 / 3 / 5) - appendix_b_placebo.txt Design 1 uniform-random placebo robustness check - appendix_b_v2_placebo_cis.txt Design 2 activity-matched placebo + 1,000-iteration bootstrap 95% CIs (canonical preferred null) - appendix_b_tier_stratification.txt by-tier lift breakdown - appendix_b_top3_exclusion.txt top-3-cohort exclusion robustness - fig1_size_distribution.svg size histogram - fig2_lorenz_curve.svg cohort-activity concentration - fig3_score_vs_launches.svg score vs launches-hit scatter - README.md dataset documentation + reproduction instructions - LICENSE CC-BY-4.0 license text Methodology overview: The detection pipeline runs in two stages. Stage 1 extracts, for each tokenlaunch, the ordered list of the first ten buyers within the bonding-curve window.Stage 2 builds a co-occurrence graph across launches, retains edges with weight>= 3, runs union-find to surface connected components, scores each component by10 x (launches hit) + 5 / max(mean first-buyer rank, 1) + sqrt(total SOL committed),and retains components with score above threshold. The full catalogue contains1,012 cohorts comprising 2,965 unique wallets, of which 153 are high-tier (>= 10launches hit or score >= 100) and 22 are premium (>= 20 launches hit). The paper that accompanies this release reports a +132.3% lift in first-30-minutebuyer count on cohort-touched launches (95% CI [+127.0%, +137.4%]) under a 3:1random-matched design, but documents that the lift does not survive anactivity-matched placebo check (placebo lift +216.3%, 95% CI [+183.8%, +255.2%],no CI overlap with the real cohort estimate). The empirical association isinterpreted as a launch-quality selection effect rather than a coordination-specific causal effect. The dataset is released so that other researchers canreproduce the descriptive results, apply alternative causal-identificationstrategies (e.g., propensity-score matching), or use the cohort catalogue as afeature in downstream graduation-prediction or surveillance models. Reproducibility: The detection script analyze_sniper_cohorts.py accepts a JSONL of buyer events(schema: mint, wallet, slot, blockTime, sol_in, tx_sig, buyer_rank) and producesa cohort catalogue identical in schema to sniper_cohorts.jsonl. The script isPython 3.10+ with no external dependencies beyond the standard library. Runningthe script against a fresh observation window of pump.fun buyer events willproduce a comparable catalogue for that window. Ethics: All data are public Solana base-58 addresses. No personally identifyinginformation is included. One wallet address in the source corpus included anethnically offensive vanity prefix; in the public release, that prefix has beenredacted to [REDACTED] while the remainder of the address is preserved fortraceability against the underlying on-chain data. Patent disclosure: The detection methodology is the subject of pending US Provisional PatentApplication No. 64/099,108 (filed 2026-06-25, Micro Entity status). The releaseof this dataset under CC-BY-4.0 does not affect the patent priority claim under35 USC 102(b)(1)(A), which provides a 12-month grace period for inventor-authored publications following the priority date. Re-implementation of thedetection methodology for academic research is unaffected; commercial use issubject to the patent rights. For commercial patent licensing inquiries: [e-mail]