The previous interval did not improve Old Faithful forecasts in selected 2011-2025 records

We tested whether the interval before an Old Faithful eruption improved forecasts of the next interval beyond duration and calendar terms. We used a fixed GeyserTimes archive and 3,134 selected records with durations from 3.5 to less than 5 minutes. Each annual fit used only outcomes that ended before that test year. Adding the previous interval changed mean squared error by 10.1% upward. Root mean squared error rose from 7.22 to 7.58 minutes. The conditional three-month block interval for error reduction was -17.0% to -0.1%. The six-month interval included zero. A check without calendar terms also had no overall gain. That check gave a gain in logger-labeled triples and a loss in webcam-duration records. Those groups cover different years, so this is not a matched source comparison. The result shows that known interval alternation did not yield an overall gain for these fitted rules. It does not rule out physical memory or other forecast methods. Selected duration reports, outcome cuts, source changes, sparse years, and retrospective data limit use. Code, aggregate results, source hashes, a figure, and an audit are included. Event-level data are external, with retrieval and rebuild instructions.

Old Faithful · hydrothermal dynamics · chronological validation · eruption duration · forecast transfer