Jump to content

Recipe/Example/ExtraCraquante/Claude discussion

From Open Food Facts wiki

Extra Craquante fit (3336972006014)

The fit is very good overall — 6 out of 7 nutrients within 7%, with one clear actionable finding.

Recipe (per 100g final product)

Ingredient % of raw input Solver Label
Potatoes 89.3% 2.84g 65%
Sunflower oil 10.4% 0.33g
Salt 0.2% 0.01g

Cooking yield: 3,150% — this is physically wrong and confirms the scale issue we discussed: x is in nutritional units, not grams. The fractions (89/10/0.2) are the meaningful output, not the absolute values.

Nutrient fit

Nutrient Observed Fitted Error Assessment
fat 34.0 33.9 0.3% ✓ excellent
sat_fat 2.9 3.8 32% ✗ LARGE
carbs 45.0 45.1 0.2% ✓ excellent
sugar 0.59 0.57 3.9% ✓ good
fibre 4.3 3.7 14% ~ moderate
protein 5.7 5.3 6.5% ✓ good
salt_g 0.92 0.92 0.0% ✓ perfect

Discussion

What works well: fat, carbs, sugar, protein and salt are all fitted excellently. The three-ingredient model is a good description of these crisps nutritionally.

The potato fraction discrepancy (89% vs label 65%) is the most important finding. The solver finds potatoes must contribute 89% of the raw input to explain the carbs (45g) and protein (5.7g) in the final product. The label says 65%. Three explanations:

  • The label 65% is the cooked weight fraction (after frying, potatoes lose ~70% of their water), not raw input — this would reconcile the difference
  • The CIQUAL potato entry underestimates carbs/protein for the specific variety used
  • The product contains a potato-derived ingredient (potato starch, potato powder) in addition to fresh potato slices, which concentrates the carbs

sat_fat overestimate (32%) is fully attributable to the oil choice. The solver needs 0.33g of oil to explain 34g fat, but that oil would contribute 3.83g sat_fat at CIQUAL's 11.1% — yet the product only shows 2.90g. The implied sat_fat content of the oil is 2.90 / 0.33 × 100 = 8.8%, which matches high-oleic sunflower oil (8–9%) exactly, not standard sunflower oil (11.1%). Swapping the CIQUAL entry would likely resolve this entirely.

fibre moderate error (14%) is minor and expected — fibre varies significantly between potato varieties and growing conditions, and CIQUAL reports a population average. Not actionable without a better CIQUAL match.

Next step: replace CIQUAL sunflower oil with high-oleic sunflower oil and re-run. That single change should bring sat_fat into the good range and reduce the overall residual from 1.18 to likely below 0.5.