Energisation schedule slip in US generating units, 2019-2026

A per-project panel measuring how far generator in-service dates move between first announcement and actual energisation. Built from 90 consecutive monthly EIA-860M vintages (January 2019 to June 2026) with stable Plant ID + Generator ID keys, per-project entry as time zero, and explicit competing-risk treatment of cancellation and censoring. Measured against each unit's most recent published in-service date, 98.7% of projects arrived on time and the 90th-percentile slip was 0 months. Measured against the first date the same units ever published, 30.6% arrived on time and the 90th-percentile slip was 16 months (n = 4,156). The first-promise measure is stable across sampling densities (29.1% semiannual, 30.6% quarterly, 30.6% monthly) while the last-promise measure is not (47.4%, 67.2%, 98.7%). Denser observation catches more intermediate revisions, each of which resets the last-promise baseline closer to the outcome, so that measure degenerates toward 100% the more often you look. Scope and limitations: this measures generating capacity interconnecting to the grid. It is not large load - data centres, industrial campuses, or other interconnecting demand. Some delay mechanisms overlap (transformer and switchgear lead times, substation and transmission construction, utility crew availability, local permitting); others do not (tax-credit safe-harbour deadlines, power purchase agreement structures, the generator cluster-study process). No public series giving projected energisation dates per project across successive queue updates was found for large load, so the equivalent measurement could not be constructed for that population. The deposit includes a manifest giving SHA-256 for every source vintage, since EIA revises and re-posts files: the hashes rather than the URLs identify the exact inputs behind every figure.