How evidence is graded.

The GRADE framework, PMID sourcing, HMAC attestation, and what every evidence card does and does not certify.

The GRADE framework

Probara uses the GRADE (Grading of Recommendations, Assessment, Development and Evaluations) methodology to classify the certainty of evidence for ingredient-outcome pairs. GRADE was developed by an international working group and is used by WHO, Cochrane, and major clinical guidelines bodies.

The grades are assigned based on the strength of the underlying study design, consistency across studies, directness of evidence, and precision of effect estimates. They are not a statement about safety, dosage, or medical efficacy.

A
High certainty
Multiple consistent RCTs or systematic reviews with low risk of bias. Further research is very unlikely to change confidence in the effect estimate.
B
Moderate certainty
Consistent RCTs with some limitations, or strong observational data. Further research may change the estimate.
C
Low certainty
Limited trials, high heterogeneity, or indirect evidence. Effect estimate may change substantially with additional research.
D
Very low certainty
Case reports, anecdote, animal studies, or conflicting results. Any estimate of effect is very uncertain.

How sources are selected

Every evidence card includes pmids: the PubMed IDs of the studies that support the grade. PubMed is the National Library of Medicine's database of peer-reviewed biomedical literature.

Source selection applies the following filters before a study is included:

Indexed in PubMed with a valid PMID
Human participants (animal-only studies are excluded from grade, cited separately as "preclinical")
Directly measures the claimed outcome (indirect surrogates are downgraded)
Published in English (non-English papers require verified translation)
Retraction status checked against PubMed retraction database before inclusion

What attestation proves

Each card carries an attestation.signature field: an HMAC-SHA256 digest computed over a canonical serialization of the card payload at the moment it was issued. Verification uses the shared secret key issued with your API credentials (HMAC-SHA256: recompute and compare).

What it proves The card was issued by Probara's signing service. The card contents have not been modified after issuance. The timestamp in attestation.timestamp is the moment of signing.
What it does not prove That the evidence is universally applicable, that the grade is appropriate for a specific product, population, or dosage, or that the underlying studies are free from methodological limitations. The grade reflects certainty of the research finding, not a medical recommendation.
How to verify Call POST /v1/verify with your attestation_token, or recompute HMAC-SHA256 over the payload field using the shared secret key issued with your API credentials. Compare against attestation.signature. Your shared key is required.

Honest limitations

This is not medical advice.
Evidence cards describe what peer-reviewed literature says about an ingredient-outcome relationship. They are not a substitute for medical consultation, and Probara does not provide dosage recommendations.
Grades can change.
New studies can upgrade or downgrade a grade. Probara monitors the literature and re-grades when material new evidence is indexed. The GET /changes endpoint surfaces updates to grades you have retrieved.
Coverage is bounded.
Not every ingredient-outcome pair is covered. If coverage is insufficient, the API returns an explicit evidence_coverage_insufficient error rather than a low-grade card.
PMID retraction lag.
Retraction checks run at card-issue time against the PubMed retraction database. A paper retracted after a card was issued will be flagged on subsequent calls. Stale cards include a stale_pmids warning field.