Data Quality

Optional. Ohne Schlüssel antwortet dieser Endpunkt nach der öffentlichen Richtlinie; ein Schlüssel erhöht Ihre Limits und schaltet planabhängige Felder frei. Authorization: Bearer <key> wird ebenfalls akzeptiert.

Report on one catalog instead of all of them. Every check is a scan of a catalog table, so a scoped call runs only the checks that concern that catalog and is the cheaper request by a wide margin. Omitted, the whole report comes back.

Akzeptiert

manufacturersfirearmscalibersattachmentsammunition

The response payload.

Which catalog this report covers, or all. Echoed so a stored copy still says what it describes.

Two different times, deliberately separate. The gap report is recomputed on every request, so computedAt describes this body. The daily snapshot is what has a last run and a next run, and it is what /v1/data/gaps/history records.

When this response was computed. Not when you read it: the response is edge-cached for a minute, so a body can be up to that old.

How often the recorded snapshot is taken.

When the last snapshot was stored, or null before the job has ever run.

The UTC day that snapshot describes.

When the next snapshot is due.

When the daily source audit last judged the catalog's citations, or null before it has ever run. Its own time rather than the snapshot's: record-cites-an-authority reads a table a separate job writes, and the two run in parallel.

Keyed by entity. Each carries its total and, per tracked field, how many records lack it. A field is missing when it is null or empty, so a column full of empty strings does not read as complete.

Visual coverage, one entry per catalog that stores media - firearms today, and an empty array for a scope that stores none. A catalog with no pictures and a catalog that does not hold pictures are different answers, so absence is never reported as zero coverage. Each kind is counted twice: records is how many records have at least one asset of that kind, which is the coverage figure, and assets is how many files exist, which is larger wherever a record carries several. none is how many records have nothing to show at all. This is what field coverage cannot report. A firearm's line drawing is a column, but its photographs, renders and 3D models are rows in a media table, so "does this record have a render" is not a question a field count can answer.

Which catalog this covers.

How many records that catalog holds.

Keyed by kind: silhouette, photo, render, model, schematic. Each carries records, assets and coveragePct.

Records with no drawing, photograph, render, model or schematic.

One entry per integrity check, each naming what passing means so a failure is actionable without reading the source.

Stable id, usable at /v1/data/gaps/checks/{checkId}.

What passing means.

Production status, one of in_production, discontinued, out_of_production, in_service, limited_production, prototype. Discontinued records are kept deliberately: a rifle built in 2004 wears parts nobody sells today.

How many records failed.

Up to five failing ids. Ask the check endpoint for the rest.

The distinct values behind the failure, present only on a check about a value rather than a record. A vocabulary failure is decided once per spelling rather than once per record, so thirty-four records reporting nineteen spellings are nineteen decisions, and ids alone would mean fetching thirty-four records to recover them. Absent on every other check rather than empty, so a check with no values to report and a check that found none cannot be confused.

Reserved for whether each cited website still resolves, which a scheduled job will fill. Reports unavailable with a reason until then.