# CORRECTED Case20 protocol: Florida–Niño 3.4, 1986–2025

Status: fixed before the corrected association calculations. This is one transparent protocol correction solely to meet the Case20 requirement of one state, one index, and 20–40 complete calendar years when available. The earlier 1950–2025 analysis used 76 years and is superseded for acceptance. Its results were already known; this is not a claim that the correction predates that analysis. No alternative windows will be calculated or screened.

The latest 40 complete calendar years are fixed as **1 January 1986 through 31 December 2025**. Outcome coverage must be exactly 480 monthly observations, with 120 observations in each of DJF, MAM, JJA and SON. Preceding Niño predictor months begin in November 1985 for the two-month lag; they do not add outcome years.

The sole scientific design change is the outcome window. Calendar-month means, trend estimates, sample counts and bootstrap record length will therefore be recomputed for this window, as required by the unchanged methods.

Unchanged choices:
- State: Florida, chosen a priori because ENSO can influence the subtropical jet and cool-season storm tracks. No other state screening.
- One index: official NOAA PSL monthly Niño 3.4 anomalies, ERSST v6, 5°N–5°S and 170°W–120°W, 1981–2010 anomaly baseline, °C. This is not three-month ONI.
- Outcome: official NOAA NCEI Climate at a Glance monthly statewide precipitation, originally inches, converted with 25.4 mm/inch. Precipitation includes snow water equivalent and is a rainfall proxy.
- Sources: reuse the exact saved public NOAA table snapshots, including their original metadata and hashes, without refreshing or substituting the sources. Retain the full snapshots; select 1986–2025 outcome rows locally.
- Seasons defined by outcome calendar month: DJF, MAM, JJA, SON. These are monthly regressions, not seasonal-total regressions.
- Niño-leading lags: 0, 1, 2 calendar months. Primary comparison: DJF lag 0. Report all 12 models without selecting the strongest.
- Estimands: monthly precipitation slope in mm per °C and correlation after removing the analysis-period calendar-month mean from outcome and aligned predictor. This is equivalent to month fixed effects and one Niño slope within each season.
- Contrasts: DJF minus JJA slope at each of the three lags; lag 1 minus lag 0 and lag 2 minus lag 0 within each season (11 contrasts total). Main seasonal contrast: DJF minus JJA at lag 0.
- Uncertainty: paired circular moving-block resampling of whole calendar years, five-year blocks, 4,000 replicates. Every model uses the same sampled years; means are recomputed per replicate. Retain seed base 20260930 and actual seed base + block length.
- Pointwise 95% percentile intervals; approximate simultaneous 95% maximum-standardized-deviation intervals across the same 23-estimate family (12 slopes and 11 contrasts).
- Sensitivities: three- and seven-year bootstrap blocks; remove a common full-record linear trend from each month-centered variable, then report all 12 detrended point estimates. No new transformations, model search, confidence procedure, or window search.
- Interpretation: observational associations, not intervention effects or validated forecast skill. Adjacent lag models are marginal associations and do not identify independent delays.
- Execution: local CPU, public monthly tables, no climate simulations, extra plugins, remote computing, or .science export.

Verification:
1. Original snapshot hashes match the earlier archive.
2. Complete unique 480 outcome months and required preceding Niño months; valid units and no sentinel/missing selected values.
3. Correct lag dates (January 1986 lag 2 uses November 1985), season counts, 12 models and 11 contrasts.
4. Primary slope matches independent least squares with month indicators; residual calendar-month means are zero.
5. Clean offline CLI rerun produces identical five CSVs and validation results; ZIP integrity and member hashes pass.
6. Earlier ZIP bytes remain unchanged; new output names clearly identify this correction; no .science member.

New deliverables: corrected report, this corrected protocol, and corrected reproducibility ZIP, all named with corrected_1986_2025.

Superseded, preserved earlier artifacts (not deleted or overwritten):
- comparison_protocol.md
- florida_nino34_report.md
- florida_nino34_reproducibility.zip

The corrected report and archive will carry this supersession notice and a provenance record. Their raw input files may span more than 40 years; only the fixed 480 outcome months enter the corrected calculations.
