Old Faithful intervals alternate in selected 2007-2025 records after source and interval cuts

We tested short-term memory in Old Faithful eruption intervals from a fixed GeyserTimes archive. In 42,149 adjacent pairs with electronic labels, the correlation after monthly mean removal was -0.332. A strict check used only electronic timestamps. It gave -0.322, with conditional 95% three-month block limits of -0.333 to -0.312. The lag-two correlation was +0.146. The adjacent correlation remained -0.268 when all tested intervals were 80 to 130 minutes. Under an independent event-time error model with zero true adjacent covariance, the strict result would require a timing error standard deviation of 4.91 minutes. Independent one-minute rounding gives only -0.083 square minutes of covariance, against -24.13 observed. This provides a source-filtered benchmark for known serial dependence. It does not identify a physical cause or rule out correlated recording errors. Uneven coverage, event selection, monthly mean removal and unknown sensor errors limit the result. Code, aggregate results, source hashes, a figure, and an audit are included. Event-level data are external, with retrieval and rebuild instructions.

Old Faithful · geyser · serial dependence · timing error · hydrothermal system