Controls kept a weighted spring shift estimate near 1.25 days per C in 317 North American bird species

We tested whether the bird survey spring count-curve shift survives nonlinear climate response, lagged shape controls and shared observation errors. We rebuilt the fixed BBS 2026 release and GHCN_CAMS climate for the 317 species in SCI 314c7ca6. Fits used 8,915,599 species-survey rows in 1966-2024. A free quadratic spring effect and previous and next spring date interactions kept the source variance-weighted estimate at 1.25 days per C, with a conditional 95% interval of 0.94 to 1.55. Corrected weighted same-minus-previous and same-minus-next intervals were 0.81 to 1.76 and 0.81 to 1.73. The planned equal-weight estimate was 0.73, with interval 0.03 to 1.43; its two corrected control contrasts included zero. Shared route and year errors caused little inflation for the source weighted estimator. Opposite-half seasonal curvatures correlated only 0.51. A secondary curve-noise ratio gave 1.43 days per C and a Fieller interval of 0.15 to 3.33 under zero cross-half error covariance. Early-period and June-only records did not give bounded calibration sets. The evidence strengthens the same-spring weighted shape result but does not identify detection, breeding dates or a thermal-trend correction. An exact count-model example shows that availability and seasonal abundance can give the same data. The package includes code, aggregate evidence, figures, source hashes and external-data retrieval. Selection, model shape, estimated weights and approximate uncertainty limit inference.