Sift — Accuracy Disclaimer

Purpose of this disclaimer

Sift provides estimated nutrition information, ingredient guesses, and allergen flags for restaurant menu items. This page explains the specific limitations and sources of error in these estimates so you can use the App with appropriate expectations.

OCR (optical character recognition) limitations

• OCR converts a photograph of a menu into machine-readable text. This process is inherently imperfect.
• Poor lighting, camera angle, glare, creases, unusual fonts, handwriting, and low-resolution images all reduce accuracy.
• OCR may misread characters, skip items, merge separate lines, or split single items across multiple entries.
• Non-English text, special characters, and decorative formatting may not be recognized correctly.
• Sift attempts to parse dish names, descriptions, and prices from OCR output, but this parsing may introduce additional errors when menu layouts are unusual.

Nutrition estimation limitations

• Calorie, protein, carbohydrate, and fat estimates are generated by heuristic models based on dish names and descriptions. They are not sourced from the restaurant unless specifically noted.
• Portion sizes, cooking methods, oils, sauces, and side dishes can significantly affect actual nutrition. Sift cannot account for these variables.
• Estimates for complex dishes (mixed plates, chef specials, off-menu items) are less reliable than estimates for well-known standard dishes.
• Published nutrition data from chain restaurants (when available) is more accurate, but may still differ from what is actually served due to portion variation, regional recipes, and preparation differences.
• Nutrition estimates should be treated as rough approximations, not precise measurements.

Allergen and ingredient detection limitations

• Allergen flags are based on heuristic ingredient guesses derived from dish names and descriptions. They are not based on verified recipe data from the restaurant.
• Sift cannot detect cross-contamination risks, shared cooking equipment, trace ingredients, or allergens introduced during preparation.
• Hidden ingredients (e.g., butter in a sauce, flour as a thickener, nuts in a garnish) may not be identified from the menu description alone.
• Allergen detection may both miss real allergens (false negatives) and flag allergens that are not present (false positives).
• If you have a food allergy, intolerance, or any condition where consuming a specific ingredient could harm your health, you must verify with the restaurant directly. Do not rely on Sift for allergy safety.

Scoring and ranking limitations

• Dish scores are computed from estimated nutrition and ingredient data, which may themselves be inaccurate. A high score does not guarantee that a dish is healthy, safe, or suitable for your needs.
• The scoring model reflects the preferences you configure. It cannot account for factors it does not know about (e.g., specific medical conditions, prescription drug interactions with foods).
• Scores are relative rankings within a single menu scan, not absolute nutritional assessments.

Menu data freshness

• Restaurant menus change frequently. Prices, items, recipes, and available dishes may differ from what Sift has in its database.
• Sift's restaurant menu datasets are updated periodically but may not reflect the current menu at any given location.
• For the most accurate information, always refer to the restaurant's current menu and speak with staff.

Summary

Sift is designed to be a helpful starting point for exploring menu options, not a definitive source of nutritional or allergen information. All estimates are approximate. All allergen flags are best-effort guesses. Always verify information that affects your health or safety with the restaurant directly.

Contact

If you have questions about accuracy or want to report an issue, contact us at:

me@srivastsh.com