Evidence Methodology
How TruthStack sources and evaluates interaction evidence.
Data Sources
TruthStack draws from four primary source categories:
Published literature. Peer-reviewed studies retrieved via PubMed and ClinicalTrials.gov. Each finding is extracted using a structured 19-field schema capturing study type, design, sample size, duration, compound form, outcome metric, and confidence indicators.
FDA FAERS (Federal Adverse Event Reporting System). Post-market pharmacovigilance data from the FDA's adverse event database. FAERS signals indicate co-reported adverse events — they are not proof of causation. Proportional Reporting Ratios (PRR) are calculated where data supports it.
FDA CAERS (CFSAN Adverse Event Reporting System). Supplement-specific adverse event reports from the FDA's Center for Food Safety and Applied Nutrition. Used for dietary supplement signals where FAERS coverage is limited.
CYP450 enzyme pathway analysis. Mechanistic interaction data derived from known cytochrome P450 enzyme induction and inhibition profiles, used to flag metabolic interactions where direct clinical evidence is limited.
Evidence Tiers
Every interaction is assigned an evidence tier based on a multi-factor scoring model:
| Strong | Multiple concordant sources, high combined confidence score, direct clinical evidence in human populations. |
|---|---|
| Moderate | Some direct evidence or multiple indirect signals. Combined confidence above threshold. May include pharmacovigilance data alongside mechanistic evidence. |
| Limited | Single-source evidence, animal or in vitro data, or mechanistic inference without direct clinical confirmation. |
| Needs Verification | Low confidence flag triggered. Evidence is sparse, contradictory, or based solely on pharmacovigilance signal without mechanistic support. |
Severity Classification
Severity levels reflect the potential clinical significance of the interaction, not the probability of occurrence:
| CRITICAL | Narrow therapeutic index medications, potentially life-altering absorption interference, or documented serious harm in published case reports. |
|---|---|
| HIGH | Significant pharmacodynamic or pharmacokinetic interaction with potential for meaningful clinical impact. Requires prescriber awareness. |
| MODERATE | Interaction flagged in literature or pharmacovigilance data. Monitor and discuss with prescriber. |
| LOW | Weak signal. Limited evidence of clinically meaningful interaction in typical use. |
| INFO | Informational flag. No interaction concern identified; entry documents absence of known interaction. |
Nightly Research Pipeline
TruthStack runs an automated research agent nightly that queries PubMed and ClinicalTrials.gov for new findings across all compounds in the database. Each finding is extracted, scored, and submitted to a multi-LLM validation review before being approved for inclusion. Approved findings update the evidence tier and severity for affected interactions.
The pipeline uses a 19-field extraction schema with a living test set benchmark. Extraction accuracy is measured against hand-labeled ground truth across study types. Regression detection alerts fire if per-field accuracy drops below threshold.
What We Do Not Do
TruthStack does not provide individualized medical advice, dosing guidance, or treatment recommendations. Severity and evidence tier reflect population-level patterns in published data — they do not account for individual health status, concurrent medications beyond the queried pair, or dose-specific risk.
Proprietary scoring fields (combined confidence, concordance, recency score, evidence gap, funding bias risk) are used internally to derive evidence tiers but are never exposed publicly. Only the derived tier and severity are surfaced.
Patent-Protected Methodology
TruthStack has filed 5 provisional patents covering the wearable-supplement correlation methodology, including matched control design for individual-level biomarker correlation with supplement ingestion events. These patents cover the methodology for generating outcome data that will supplement the published literature layer over time. Filing date: January 20, 2026.
Update Cadence
Interaction records are updated as new evidence is approved through the nightly pipeline. The last_reviewed field on each interaction page reflects the most recent data review date for that specific compound pair. Dataset versions are published quarterly at /data/glp1-index.json.