Name the claim
Capture the wording of the statement before attempting to verify it.
The following principles describe the documentation workflow we want to explore for specialised nutrition-related AI systems.
This is an intended design approach. It is not a claim that every feature or verification step is already deployed.
Capture the wording of the statement before attempting to verify it.
Record who produced the material and the role they played.
Separate enduring principles from information that may have changed.
Check animal species, age, study conditions and the actual subject examined.
Look for disagreement, methodological limitations and alternative explanations.
Explain when the available evidence does not justify an individual recommendation.
Research articles, industry standards, regulatory texts, manufacturer product sheets and educational resources can all be useful. Their evidential weight depends on the question being asked.
A robust interface should distinguish a claim from its supporting reference, where possible provide a link and explain whether the information was independently checked.
The intended audience affects terminology, assumptions and how findings should be applied.
No single score can represent every animal, feeding context or health consideration. Any proposed indicator must disclose its criteria.
Yes. Models can overgeneralise, omit conditions or misstate a source. Citations and human checks reduce some risks but do not remove them.
For an animal with clinical symptoms, a suspected allergy, an existing diagnosis or a proposed therapeutic diet, consult a veterinarian rather than relying on a general-purpose content tool.
Describe your intended audience, data constraints and desired output. We can discuss a realistic technical approach before any development begins.