Four US lilac samples needed delays in 4 to 15 percent of reports to remove spring slopes under a 14-day cap

We tested how many delayed reports can remove a fitted spring-temperature slope in 9,015 US lilac records from 1965 to 2008. Each delay was allowed to depend on weather and was capped at 14 days. Exact bounds with plant and year effects and winter controls required delays in at least 115 of 2,679 and 45 of 723 eastern reports, or 4.3% and 6.2%. Western samples required 638 of 5,085 and 77 of 528 reports, or 12.5% and 14.6%. These are finite-sample sensitivity thresholds, not actual error rates or population confidence limits. Flag removal and plant trends kept a similar pattern. A March-only spring window reduced western thresholds to 5.2% and 10.2%, which shows model dependence. We reuse a published panel and extend prior unrestricted delay bounds with a limit on the affected fraction. The package includes source hashes, a byte-identical raw rebuild, code, attaining delay assignments, independent regression checks, results and a figure. Unknown visits, selected long records, different plant types by region and coarse weather limit inference. The test does not measure pollution, daily extremes, biodiversity loss or a causal climate effect.

phenology · observation bias · lilac · reporting delay · sensitivity bounds · spring temperature · sparse errors