fix(qa): Resolve 9 NutriRange calorie back-calculation outliers — adjust EU FIC tolerance

[ericsocrat]

Problem Statement

QA Suite 11 (QA__nutrition_ranges.sql) has 9 persistent calorie back-calculation outliers that have been flagged as known failures across multiple sessions. These are products where the sum of macronutrient calories (fat×9 + carbs×4 + protein×4 + fiber×2) deviates significantly from the declared calorie value.

Root cause: Open Food Facts source data contains nutrition values exactly as printed on European food labels. Label rounding rules (EU FIC Regulation 1169/2011) allow manufacturers to round nutrition values, which creates back-calculation mismatches. This is not a data entry error — it's inherent to how food labeling works in the EU.

Current state:

• 9 products consistently fail the calorie back-calculation check
• All other 11+ nutrition range checks pass
• These failures are documented as pre-existing/non-blocking in CURRENT_STATE.md

User impact: None directly — scoring uses the declared calorie value (which is correct per the label). However, the persistent QA failures create noise that masks real regressions and erodes confidence in the QA suite.

---

Architectural Evaluation

Approach	Verdict	Rationale
A. Override OFF calorie values with back-calculated values	❌ Rejected	Violates data integrity — declared values on the label are the legal reference.
B. Widen tolerance threshold for back-calculation check	✅ Chosen	EU rounding rules allow ±10-20% deviation. Current check may use too tight a tolerance. Widen to match EU regulatory tolerance (±20% for vitamins/minerals, ±8kcal for energy).
C. Remove the back-calculation check entirely	❌ Rejected	The check catches genuine data quality issues. Should remain but with appropriate tolerance.
D. Exclude known outliers by product_id	⚠️ Last resort	Fragile — hardcodes product IDs. Only if tolerance adjustment doesn't resolve.

---

Implementation Plan

Phase 1 — Analyze the 9 Outliers

-- Identify the 9 failing products and their deviation %
SELECT p.product_name, p.brand, p.category, p.country,
       nf.calories_kcal AS declared,
       ROUND((nf.total_fat_g * 9 + nf.carbs_g * 4 + nf.protein_g * 4 + COALESCE(nf.fiber_g, 0) * 2)::numeric, 0) AS calculated,
       ROUND(ABS(nf.calories_kcal - (nf.total_fat_g * 9 + nf.carbs_g * 4 + nf.protein_g * 4 + COALESCE(nf.fiber_g, 0) * 2)) / NULLIF(nf.calories_kcal, 0) * 100, 1) AS deviation_pct
FROM products p
JOIN nutrition_facts nf ON p.product_id = nf.product_id
WHERE p.is_deprecated = false
ORDER BY deviation_pct DESC
LIMIT 20;

Phase 2 — Adjust Tolerance

Update the QA check in QA__nutrition_ranges.sql to use EU FIC-compliant tolerance:

• Energy value tolerance: ±20% (per EU FIC Regulation 1169/2011 guidance)
• Current tolerance is likely ±10% or absolute deviation

Phase 3 — Validate

.\RUN_QA.ps1  # Verify 9 outliers now pass with adjusted tolerance

---

Test Requirements

• QA__nutrition_ranges.sql — all 20 checks pass (including previously-failing 9)
• No legitimate data quality issues are masked by the wider tolerance
• Tolerance value documented in the QA check comment

---

Verification Checklist

• Analysis of 9 outliers completed (deviation % documented)
• Tolerance adjusted with EU FIC regulatory justification
• All 20 nutrition range checks pass
• CURRENT_STATE.md updated (remove from Known Issues)
• copilot-instructions.md updated if QA check count changes
• CHANGELOG.md updated under [Unreleased]