FAMILY_VEGETATION_PEAK_TIMING - Experiment Results

Overview

This document tracks all experimental runs for the vegetation peak timing intelligence hypothesis family. Each experiment tests whether NDVI peak timing derived from real Sentinel-2 satellite data can predict harvest season potato prices.

Experimental Status

• Status: Ready for implementation
• Created: 2025-08-20
• Data Sources: Real Zarr satellite data + BoerderijApi prices
• Priority: High - Novel real NDVI analysis after synthetic data violations

Data Validation

• ✅ Zarr store available: lake_31UFU_small.zarr (525MB)
• ✅ Price data accessible: BoerderijApi NL.157.2086
• ✅ BRP parcels: Consumption potato boundaries
• ✅ Standard baselines: All 4 standard baselines (persistent, seasonal_naive, ar2, historical_mean) ready

Experiment Queue

1. Variant A: Peak Date Predictor (early stress detection)
2. Variant B: Peak Intensity Predictor (vigor-oversupply dynamics)
3. Variant C: Peak-to-Harvest Lag (quality optimization timing)

Implementation Notes

• CRITICAL: Use ONLY real satellite data from Zarr stores
• MANDATORY: Test against all 4 standard baselines
• Method: Rolling-origin cross-validation with proper temporal ordering
• Target: >5% improvement over strongest baseline
• Validation: DM test + TOST equivalence testing

Verdict Template Ready

Each experiment will follow the standard verdict format with: - Complete baseline comparison (all 4 baselines standard baselines (persistent, seasonal_naive, ar2, historical_mean)) - Statistical significance testing - Practical improvement metrics - Data provenance documentation - MLflow run links and artifact paths

Experiments will be appended below as completed

Codex Validation — 2025-11-10

Files Reviewed

• hypothesis.yml
• hypothesis.md
• experiment.md

Findings

1. No implementation. The family contains only documentation; there is no runner, dataset builder, or script that ingests Sentinel/BRP/price feeds.
2. No evidence of real-data usage. Without code, we cannot confirm that any of the referenced data sources have been accessed.
3. No baseline comparison. Since no experiment was run, there is no proof that the proposed vegetation-peak timing features outperform a price-only model.

Verdict

NOT VALIDATED – This family is still at the planning stage. It will remain unvalidated until executable code uses real data and demonstrates statistically significant gains over the mandated baselines.