Institutional
AI methodology
Where models are used, what they are permitted to do, and what remains deterministic.
Indices are not machine-learned
Index computation is deterministic, versioned and reproducible. A score a reader cannot recompute from the published weights and inputs would not be defensible, so no index is produced by a learned model.
Where models are used
Machine learning is applied where pattern extraction from unstructured material is genuinely required: attribute extraction from imagery, text classification, entity matching and similarity scoring. Forecasting uses statistical and hybrid methods chosen for small-sample honesty rather than apparent sophistication.
Each deployed model has a documented purpose, input list and limitation statement.
The language interface does not answer from memory
The conversational layer resolves entities, selects from a registered set of queries, and narrates only the values those queries return. It does not generate figures. Where the platform holds no answer, it states that rather than producing one, and points to the nearest figure it does hold.
No inference of protected attributes
No model on this platform infers or stores race, ethnicity, religion, health, sexual orientation or political affiliation, for attendees, designers or consumer segments. Consumer outputs are aggregate and are suppressed below a minimum cell size.