I’d love to use it in my integration, but am finding it only sporadically useful.

In my (brief) tests I found that I get failure about twice as often as success:

Failure: 4582, 1062027, 2012, 17013, 9032, 10011693, 6170, 20028, 1116
Success: 9156, 11282, 11215, 11485

For example, 17013 (leg of lamb) returns failure, which is unexpected.

I would love to see the scope of this endpoint broadened. Maybe with a "confidence’ level (as is seen in the Google Cloud Vision API when tagging images).

Alternative search methods to consider:

• Taxonomy / Ontology - if the food is categorised “meat”, any other meat products
• Macro Search - any other ingredient with a similar macro profile
• General Search - assuming a blacklist of common words (ie: leg, cut, wedge etc) this would catch other products mentioning lamb
• Fuzzy Search - not a great approach, but shouldn’t be ruled out