RED-PUMP-2026-v1: A public benchmark of 860,213 pump.fun token launches with survival outcomes (May–June 2026 regime)

v1.4 (2026-08-13). Corrigendum: retraction of the v1.3 ceiling crossing recovery analysis. The v1.3 corrigendum proposed a "ceiling crossing" correction that reclassified 1,126 TIMEOUT mints as probable late graduations, raising the pooled rate to a claimed 0.333%. This correction is withdrawn. A direct check against pump.fun's live bonding curve API on a 100 mint sample found zero of 100 showing evidence of graduation. The field used to identify these mints, final_market_cap_sol, reflects the bonding curve's virtual reserves, which move with speculative price action and are not capped at the graduation threshold, rather than the real reserves that actually determine graduation. The 9.5x to 10x "corrected" Telegram lift range and the Cox concordance stress test result (0.858 dropping to 0.794) are withdrawn along with it. The v1.3 coverage window finding itself is not affected by this withdrawal: the published rate remains a confirmed lower bound on the true 24 hour graduation rate, measured within an observer visibility window of roughly six minutes, and the true 24 hour rate remains unknown from this dataset. Full detail is in CORRIGENDUM_v1.4_2026-08-13.md and .pdf, the updated KNOWN_LIMITATIONS.md, and CEILING_VERIFICATION_2026-08-13.zip, which contains the full reproducible verification evidence. v1.3 (2026-08-11). Corrigendum: coverage window relabeling. An independent external audit received 2026-08-11 found that the collector polled a top 50 newest REST endpoint on pump.fun's public API. At the observed launch rate of about 18.08 new mints per minute during the observation window, the top 50 buffer gave each mint an effective visibility of only about 2.77 minutes before newer mints displaced it. The v1 headline "24 hour pooled graduation rate = 0.198%" is therefore a fast regime rate measured within roughly six minutes of launch, and a lower bound on the true 24 hour rate, not an estimate of it. Raw data files (red_pump_2026_v1_launches.jsonl.gz, red_pump_2026_v1_outcomes.csv.gz) are unchanged in v1.3. Every graduation, timeout, and timestamp in the deposit reflects what the observer actually saw. See CORRIGENDUM_v1.3_2026-08-11.pdf and KNOWN_LIMITATIONS.md for full detail. A v2 dataset re collected using a Helius bonding curve program subscription, which removes the visibility window entirely, is planned within 4 to 6 weeks of the v1.3 corrigendum. Dataset. RED-PUMP-2026-v1 is a public benchmark of 860,213 Solana pump.fun token launches observed continuously between 2026-05-08 and 2026-06-10, a 34 day window. Of these, 832,941 unique mints reached a terminal outcome within the observer's tracking window: 1,651 graduated to PumpSwap, or to Raydium for tokens preceding the PumpSwap migration, and 831,290 timed out. The pooled rate is 0.198% (Wilson 95% CI 0.189% to 0.208%). The steady state rate, excluding a four day tracker warm up window and the partial last day, is 0.207% (Wilson 95% CI 0.198% to 0.218%). Both rates apply the v1.3 coverage window relabeling described above. The v1.3 ceiling crossing rate estimate of 0.333% is withdrawn as of v1.4 and should not be cited. Fields. Each launch record contains the Solana mint address, the observer detection timestamp, the on chain creation timestamp, the initial market capitalisation in SOL, three boolean indicators for advertised social channels (Twitter/X, website, Telegram), and the description character length. Each terminal outcome record contains the outcome class, minutes to outcome, and the detection lag. Companion paper. Released alongside a companion preprint titled "Pump.fun Graduation Regime Windows: Survival Analysis of 832,941 Token Launches and the Social Presence Effect" (SSRN 6915560, arXiv 2607.02823). The companion paper reports Kaplan Meier and Cox proportional hazards results stratified by social channel presence and initial market capitalisation quartile. Key findings include an 8.94x graduation rate lift for launches advertising a Telegram channel, likely a lower bound on the true 24 hour effect given the coverage window limitation described above, and a multivariate Cox concordance of 0.858, stable under bootstrap resampling and under a time window split. An earlier stress test of this concordance figure, reported in the v1.3 corrigendum, relied on the withdrawn ceiling crossing correction and is itself withdrawn as of v1.4. Schema and reproducibility. The schema is frozen as red-pump-1.0. Future monthly versions will preserve all v1 fields. Reproducible analysis code, built on Python 3.12 and lifelines 0.30.3, is included in the deposit as km_full.py and km_figures.py. Version history.v1.0 (2026-06-11): permanent original raw data release. Version DOI 10.5281/zenodo.20633487.v1.1: corrected the placeholder DOI in the README.md citation block, now referencing concept DOI 10.5281/zenodo.20633486 and v1.0 DOI 10.5281/zenodo.20633487. SHA256SUMS regenerated. All other files identical to v1.0 by SHA256. Version DOI 10.5281/zenodo.20633603.v1.2: metadata revision. Version DOI 10.5281/zenodo.21383616.v1.3 (2026-08-11): coverage window corrigendum. Added CORRIGENDUM_v1.3_2026-08-11.pdf, KNOWN_LIMITATIONS.md, and README_v1.3.md alongside the original README.md, and SCHEMA_v1.3.md alongside the original SCHEMA.md. Replaced CITATION.cff with v1.3 metadata. Raw data files unchanged. Version DOI 10.5281/zenodo.21908375.v1.4 (2026-08-13): retraction of the v1.3 ceiling crossing recovery analysis following direct on chain verification. Added CORRIGENDUM_v1.4_2026-08-13.md and .pdf, and CEILING_VERIFICATION_2026-08-13.zip. Updated KNOWN_LIMITATIONS.md and CITATION.cff, and added README_v1.4.md alongside the earlier README versions. Raw data files unchanged. Version DOI assigned on publish. Patent and commercial licensing disclosure. This dataset is released under CC-BY-4.0. Use of the dataset itself for research, analysis, or product development is permitted under CC-BY-4.0 terms, with attribution required. The author holds pending USPTO provisional patent applications in the field of algorithmic decentralized exchange trading and blockchain market microstructure, including UDAY P1 #64/022,461, filed 2026-03-30, Micro Entity status, and UDAY P2 #64/099,108, filed 2026-06-25, Micro Entity status. Release of this dataset under CC-BY-4.0 does not affect patent priority claims under 35 USC 102(b)(1)(A), which provides a 12 month grace period for inventor authored publications following the priority date. Commercial re-implementation of specific methodologies covered by pending patent claims may require patent licensing. Academic re-implementation is unaffected. For commercial licensing, extended window data, consulting, or reproduction support, contact [e-mail].