Pump.fun Graduation Regime Windows: Survival Analysis of 832,941 Token Launches and the Social-Presence Effect

Kaplan-Meier and Cox proportional-hazards survival analysis of 832,941 Solana pump.fun token launches, observed 2026-05-08 to 2026-06-10. Pooled graduation rate 0.198% (Wilson 95% CI [0.189%, 0.208%]; steady-state 0.207%). A post-publication audit established that the collector's coverage was effectively limited to about six minutes after each launch, so this figure should be read as a fast-regime graduation rate and a lower bound on the true 24-hour rate, not an estimate of it. Comparing to Marino et al. (2026)'s 0.63% for September-October 2025 gives a ratio of 3.18x; because our figure is itself a lower bound, this ratio is an upper bound on the true decline, and the true decline is smaller by an amount this dataset cannot establish. We formalise the windowed comparison as the Graduation Regime Windows framework and release the 860,213-launch dataset under CC-BY-4.0 on Zenodo. Within the fast-regime window, Telegram-channel launches graduate at 1.485% versus 0.166% without (8.94x lift; Cox HR 5.40, 95% CI [4.73, 6.17]), likely itself a lower bound on the true 24-hour effect. Initial market capitalisation above the 30 SOL platform default is the strongest single predictor (Cox HR 4.51); the top quartile graduates at 0.634%, close to though not conclusively matching the Marino et al. pooled rate. Cox concordance 0.858, stable under bootstrap and a time-window split. A subsequent attempt to recover missed late graduations from implausibly high TIMEOUT-mint market caps was tested against pump.fun's live bonding-curve state and did not hold up (0 of 100 sampled mints showed graduation); we report and withdraw it. Zenodo concept DOI 10.5281/zenodo.20633486; corrigendum v1.4 at 10.5281/zenodo.21923106.