What separates a registry from a scraped list is not accuracy — both may be largely accurate.
It is being able to show where every value came from, and being wrong in ways that can be found and fixed.
Provenance
- Every field carries a source.
- Every field carries a last-verified date.
- Confirmed and inferred values are distinguished explicitly.
- Provenance survives export — stripping it makes downstream verification impossible.
- Data without provenance cannot be evaluated by anyone, including its own maintainers.
Correctability
- A visible correction route on the record itself.
- A stated response time, and meeting it.
- Corrections logged, never applied silently.
- Previous values retained.
- A source that cannot be corrected accumulates errors permanently.
Transparency about limits
- Coverage stated by field — what percentage of records have each value.
- Known limitations published plainly.
- Methodology available.
- Correction volumes reported.
- A source claiming no limitations is not error-free, only opaque.
Stability
- Identifiers that never change or get reused.
- Retired identifiers still resolving.
- Versioned definitions so past figures remain reproducible.
- Advance notice of breaking changes.
- Instability is as damaging as inaccuracy for anyone building on the data.
Evaluating a source before relying on it
- Can you see where a given value came from?
- Can you report an error, and does anything happen?
- Is coverage stated, or implied to be complete?
- Are identifiers persistent?
- Is there a plan for what happens if it stops operating?
Warning signs
- No indication of source for any value.
- No correction mechanism.
- Coverage presented as complete.
- Aggregate figures with no stated counting rule.
- Confident precision on fields that are inherently uncertain.
One thing worth remembering
Ask one question of any academic data source: can you see where a specific value came from?
If not, you cannot evaluate the data — and neither can its maintainers, which means errors in it are not being found by anyone.
Câu hỏi thường gặp
What separates a registry from a scraped list?
Not accuracy but provenance — being able to show where every value came from, and being wrong in ways that can be found and fixed.
What does provenance require?
Every field carrying a source and last-verified date, confirmed and inferred values distinguished, and provenance surviving export.
Why does correctability matter?
Because a source that cannot be corrected accumulates errors permanently — requiring a visible correction route, stated response times, and logged corrections.
What does transparency about limits involve?
Coverage stated by field, known limitations published plainly, methodology available, and correction volumes reported — since a source claiming no limitations is opaque rather than error-free.
What single question evaluates a data source?
Can you see where a specific value came from — if not, neither you nor its maintainers can evaluate the data.
