Coverage dataset
Every question the Arcnorm registry could not answer, tallied by jurisdiction and reason. It is published because the demand behind a gap should be checkable rather than asserted — it is the same tally that ranks what the registry covers next.
What this dataset is
- Publisher
- MB Arcnorm
- Reviewed answer corpus
- Lithuania, from E-TAR
- Corpus review date
- Tally as of
- Licence
- RSL 1.0, attribution to arcnorm.com required (Terms)
A row is a count, not a question. The registry stores the jurisdiction, the reason and the count — never the query, the user, the device or the email — so the tally can say that N asks went unanswered, and cannot say what was asked or by whom.
The tally
| Jurisdiction | Reason | Asks | Last recorded |
|---|---|---|---|
| Lithuania LT | Search unavailable retrieval_unavailable | 64 | |
| Latvia LV | No matching source no_results | 36 | |
| Estonia EE | Search unavailable retrieval_unavailable | 26 | |
| Latvia LV | Search unavailable retrieval_unavailable | 11 | |
| Lithuania LT | No matching source no_results | 4 | |
| Lithuania LT | Could not ground an answer chat_abstained | 2 |
Reason codes
- No matching source no_results
- The registry was searched and holds no source that answers the question.
- No valid source retrieval_invalid
- A source was retrieved but was not valid context for the question.
- Search unavailable retrieval_unavailable
- The retrieval service could not be reached, so no answer was attempted.
- Could not ground an answer chat_abstained
- Context was retrieved but the assistant declined to ground an answer in it.
Method
Each time the registry cannot answer a question, the jurisdiction and one of four reason codes are incremented in an aggregate counter. The query text, the user, the device and the email are never stored.
Counts are cumulative from the table’s creation and the timestamps are the last time a row was incremented, not the start of a window: there is no per-day breakdown in this dataset. The columns are exactly the four published above.
What it does not say
- A jurisdiction with no rows is not a jurisdiction with no gaps. A gap is recorded only when a question is actually asked and actually fails, so the tally measures recorded demand, not coverage completeness.
- retrieval_unavailable counts an outage rather than a content gap. It is published anyway: dropping it would make this dataset describe a more reliable service than the registry is.
- Counts include repeated and automated asks. A count is not a number of people, and it is not a number of distinct questions.
- No machine-readable download or bulk export is published. This page is the published form of the tally, and the registry API remains a partner surface rather than an advertised one.
Looking for an answer that is already covered? Open the registry