Resources & technical notes
Engineering principles, made practical.
Notes on provenance, timeline integrity and parser evaluation that shape QUARDENCE’s development.
01 / PROVENANCETraceable AI: keep the source in sight
A useful summary should let the reviewer move from a statement to the supporting artifact and then to the source record. A generated observation should identify the records it used, preserve relevant context and make missing support visible.
- Give each parsed artifact a stable reference within its processing context.
- Keep source text and generated summaries in separate fields or views.
- Show record-level support for material statements where practical.
- Route consequential interpretations for practitioner review.
02 / TIMELINESNormalize time without losing original time
Timeline views become easier to compare when original and normalized values are shown together. The conversion method should be explicit enough to reproduce the analytical view.
- Retain the original stored value and source field.
- Record the timestamp unit and known timezone or offset.
- Show conversion logic beside normalized analytical time.
- Preserve event meaning and flag unresolved time context.
03 / VALIDATIONTest parsers with known answers and regressions
Parser changes should be checked against inputs whose expected interpretation is documented. Test results can be organized by parser version, source schema and error type.
- Use known-answer and synthetic test corpora, plus authorized anonymized material where appropriate.
- Run version-aware regression tests when schemas or parsers change.
- Track field interpretation, timestamp conversion, false-positive and false-negative errors.
- Record processing versions and coverage gaps to make results repeatable.
Public reference material
These public documents provide context for the project’s engineering priorities.
