What public data cannot prove on its own
Structured data can make evidence easier to use. It cannot make incomplete sources complete or turn a correlation into a fact.
Sources describe a point in time
A change can occur before it appears in a source, and different sources refresh on different schedules. Police notices are preliminary and can be updated; Statistics Sweden describes weekly administrative updates to its Business Register. [2][6] Every delivery therefore needs collection dates and a refresh policy.
Coverage differs by field and jurisdiction
A public feed can be selective, a register can omit fields, and definitions differ between sources and jurisdictions. A blank value can mean unavailable, not applicable, not collected, or not found; those states should not be collapsed.
Entity matching can be wrong
Similar names, reused addresses, reorganizations, and inconsistent identifiers create false merges and missed matches. High-impact matches need stronger identifiers or manual review.
A signal is not an outcome
Predictions reflect their training data, chosen target, and observation window. A high score means the model found a defined pattern. It does not mean an event will occur or that a subject should be accepted or rejected.
Screening defined populations, standardizing records, tracking registered changes, and making source evidence repeatable.
Resolving ambiguous entities, interpreting unusual context, handling conflicting evidence, and making consequential decisions.
Good data products expose uncertainty instead of hiding it: provenance, dates, null reasons, confidence, model version, and known coverage gaps belong in the delivery.
Sources
- Statistics Sweden's Business Register: Statistics Sweden
- API över polisens händelser: Polismyndigheten