Restaurant nutrition data for Balanced Approach Fitness. Updated 2026-01-22.
| Tier | Check | Purpose |
|---|---|---|
| T1a | nutrient > cal/2 | EXTREME thermodynamic |
| T1b | nutrient*4 > cal+20 | Thermodynamic impossibility (FDA buffer) |
| T2 | sat_fat > total_fat | Compositional integrity |
| T3a | Beverage protein limits | Category coherence (expanded patterns) |
| T3b | Dessert protein limits | Category coherence |
| T3c | Condiment protein limits | Category coherence |
| T4 | High protein + zero sugar | Column shift heuristic |
| T5 | Fruit protein limits | Category coherence |
| Version | Items | Excluded | Key Learning |
|---|---|---|---|
| v15 | 12,008 | 0 | No audit gate |
| v15.1 | 11,196 | 812 | FDA rounding = false positives |
| v15.2 | 12,003 | 5 | Two-tier separates true errors |
| v15.3 | 11,911 | 92 | Category coherence catches beverage shifts |
| v15.4 | 11,297 | 706 | Atomized audit with encoding normalization |
Key insight: Audit logic complexity must be congruently atomized with source PDF realities - encoding artifacts, creative naming, and multi-section column shifts.