Appearance
ADB Ingestion Model
Purpose
ADB proves value by ingesting heterogeneous data from multiple sources and scaling from 2x to 10x to 100x without redesigning the core data foundation.
Ingestion Sources
- Doctor-led assessment forms
- Doctor notes
- SLP / therapist session inputs
- Health-worker field capture
- Parent/helper contextual inputs
- Child-generated practice responses
- Questionnaire and screening forms
- Therapy goal/progress records
- Notes, labels, metadata, and tags
- File uploads
- Audio/video references
- Historical records
- Institution / special-school records
- Rural/community-capture records
- Partner / clinic imports
- Ed-tech partner inputs
- Assistive-device partner inputs
- Future APIs
- Future device-assisted inputs
- Future AI-assisted annotation inputs
Ingestion Control Layer
The ingestion control layer should capture:
- source identification;
- source role;
- source channel;
- source system;
- capture timestamp;
- observation timestamp;
- consent / permission tagging;
- data validation;
- duplicate detection;
- versioning;
- review status tracking;
- data quality flags;
- transformation lineage;
- routing to ADB zones.
Ingestion Guardrails
- Every input must have a source.
- Every input must carry metadata.
- Raw input must be preserved before transformation.
- Structured observations should be derived, not overwrite raw data.
- Review state must be explicit.
- Doctor/clinician-approved outputs must be versioned.
- Media can start as references before deep processing.
- Partner imports must be traceable.
- Future AI annotations must be separated from human observations.
- Model-ready datasets must come only from reviewed/curated data.