Capture
Store the source, its terms, scope and the specific claim or observation being represented.
Methodology 0.1.0
Trust is a data model, not a badge.
Terplax designs its methodology so a careful reader can reconstruct why a record exists, what it applies to and where uncertainty remains.
Claim pipeline
From source to public record
Normalize
Resolve entities, units and vocabulary without erasing original language or losing provenance.
Review
Assess source type, method, coverage, conflicts, relevance and quality limitations.
Publish
Release a versioned public record with state, confidence and inspectable evidence objects.
Reassess
Record corrections, new evidence and methodology changes instead of silently overwriting history.
Evidence grades
Confidence is disclosed, not implied.
Directly observed
Primary source with method and relevant sample context available for inspection.
Supported
Credible and attributable source with enough scope to interpret the relationship.
Provisional
Plausible but missing material context, independent review or adequate source coverage.
Disputed
Conflicting evidence, unresolved terminology or a material methodological concern.
Index policy
Global requires diversity.
The Global Flavor Attention Index is a publication framework, not a promise that one source can measure worldwide interest. A global value is withheld when independent sources, freshness, coverage or reproducibility are insufficient.
- Signal inputs are retained at source level with provider, time window and transformation notes.
- Demo fixtures remain visually and programmatically labelled until review-backed live publication.
- Scores never substitute for sales, prevalence, safety, efficacy or formulation performance.
- Uncertainty, disagreement and missing coverage remain part of the record.
AI policy
AI can synthesize; it cannot certify.
AI-generated text is allowed only as a bounded synthesis over linked source objects. It must be labelled, preserve caveats and never manufacture citations, evidence grades or a global conclusion.
- Every material AI statement needs an inspectable source link or is withheld.
- Models do not autonomously publish, alter source metadata or approve evidence.
- Human review and correction history are modeled as records, not informal notes.
Contribution protocol
Begin with a concrete source and a scoped claim.
Future contributions will require a stable source reference, the proposed entity or relationship, original wording, scope and any method context. Uncited “industry knowledge” enters a review queue and is not auto-published.
About the platform