## TL;DR
Completeness failures mean required fields arrive empty more often than tolerated: measure missing rates per field and per source segment, find the segment or field that crossed the threshold, and fix its population path. A single field going empty on one source is the common pattern behind an overall completeness breach.

```text
completeness check failed error: missing values percentage failed threshold
```

## Use this when
- The check or run fails with this symptom: the completeness check reports missing values above the allowed percentage.
- You are working on a completeness check failing on missing values above the threshold and the failure repeats on the same data or configuration.
- You need the failing rows, rule, or stage isolated before changing thresholds or code.

## Not for this skill when
- A table that is missing whole rows rather than fields, which is a volume problem instead.
- The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

## Steps
1. Compute the missing percentage per required field and per source segment for the current batch. Verify: the breaching field and segment are identified
2. Sample records in the breaching segment and compare their raw form with the transform's expectations. Verify: the raw-versus-transform gap is visible
3. Check for a source format change, a field rename, or a join that stopped matching for that segment. Verify: the population break is located
4. Fix the mapping or join, reload the segment, and quarantine records that genuinely lack the data with a reason code. Verify: the segment reloads with fields populated
5. Rerun the completeness check and confirm every required field is under its threshold. Verify: the check passes field by field

## Variant phrasings
### completeness check missing values percentage threshold
The verbatim completeness phrasing.
### required field empty completeness failed segment
The segment phrasing for one source going empty.
### data completeness check failed after source change
The source-change phrasing for renames and formats.

Compatibility: Soda completeness checks, Great Expectations, and dbt tests on warehouse tables

## Why it happens
Completeness aggregates many fields into one verdict, but fields break one at a time through renames, format changes, and join misses. The aggregate breach is real; its cause is always specific to a field and a segment, which the aggregate number conceals.

## Edge cases / pitfalls
- Set thresholds per field by criticality. One blanket percentage makes critical fields as lax as optional ones.
- Track completeness per segment in the run metadata so the breaching segment is in the alert itself.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_Dl80_vp3IYpM_0fZYzSEHA