Experiment Log - FAMILY_REGIONAL_WEATHER_DIVERGENCE

Hypothesis

Divergent weather patterns between Dutch production regions (Flevoland vs Zeeland vs Noord-Brabant) create internal arbitrage opportunities and regional supply imbalances that drive predictable price movements.

Data Sources (REAL DATA ONLY)

• Open-Meteo API: Multi-location weather (Flevoland, Zeeland, Brabant, Groningen)
• BRP API: Regional parcel data by province
• CBS API: Regional production statistics
• Boerderij API: Weekly prices

Variants

• Variant A: Temperature divergence focus (regional temperature variance)
• Variant B: Precipitation divergence focus (drought/flood contrasts)
• Variant C: Integrated regional imbalance (combined weather + arbitrage)

Experimental Runs

Run 1: [Pending]

Date: TBD Variant: TBD Status: Not started

Configuration: - Target horizons: 30-day, 60-day - SESOI: 15% - Spatial analysis: 4 Dutch regions - Baseline models: persistent, seasonal_naive, ar2, historical_mean

Results: - Awaiting implementation

Verdicts

Overall Assessment: PENDING

Awaiting experimental validation.

HE Notes

• Created 2025-08-18
• Extends FAMILY_YIELD_VARIANCE_PREDICTORS concept from satellite to weather station divergence
• Industry evidence: Aug 2022 Flevoland drought vs Zeeland normal → €15/ton spreads
• Academic support: Schlenker & Roberts (2009) spatial weather correlation determines price volatility
• Focus on intra-Dutch regional dynamics vs cross-border
• Uses REAL weather data from 4 Dutch locations
• NO synthetic data - all from repository interfaces

Decision Log

[To be completed after experiments]

Codex Validation — 2025-11-10

Files Reviewed

• run_experiment.py
• experiment.md
• hypothesis.yml

Findings

1. Implementation exists but never executed. run_experiment.py wires Boerderij + Open-Meteo calls and builds divergence features, yet experiment.md:24-44 shows every run as “Pending” with no metrics, MLflow artifacts, or verdicts.
2. Real-data reliance still unproven. Although the loader calls the Boerderij and Open-Meteo clients (run_experiment.py:35-99), there is no captured evidence (logs, saved datasets, manifests) that the APIs were successfully queried or that regional data covered the full time span.
3. Baseline comparison unresolved. The CV loop computes MAE differences but nothing is logged to show the models beating the price-only baselines supplied by get_standard_baselines.

Verdict

NOT VALIDATED – Until a run is executed on real Boerderij/Open-Meteo data and statistically significant gains over the mandatory baselines are recorded, the regional weather divergence hypothesis remains unvalidated.