Data Quality

Data Quality shows how ready your product catalog is for export and analysis. Each snapshot captures coverage, freshness, and attribute-completion metrics, plus a bounded sample of row-level issues. This lets you track whether data quality improves or slips over time.

The page separates two independent quality contracts, shown as tabs:

Catalog Quality. The shared standard that applies across all tenants. It covers core price, taxonomy, brand, and retailer coverage, freshness, completion of the global category attribute vocabulary, and catalog-level findings.
Feature Set Quality. Compliance with your own tenant’s configured category attributes. See Feature Sets. A category attribute that you have overridden or added yourself is scored here instead of in Catalog Quality, so the same attribute slot is never scored twice.

Use the Refresh button at the top of the page to reload the latest snapshot and history at any time.

Catalog Quality

The Catalog Quality tab shows:

Catalog Score. A 0-100 score that combines core coverage, freshness, global category attribute fill, and catalog-scoped findings.
Status. Pass, Warn, or Fail. The status moves to Warn as coverage or freshness drop below target levels, or as catalog-scoped findings accumulate. It moves to Fail when the snapshot contains no products, when any catalog-scoped finding has fail severity, or when coverage falls far enough below target. Examples of fail severity are a zero or placeholder price, a non-ISO currency code, and a price column that contains non-numeric characters.
Products. Distinct products in the latest snapshot. The underlying row count appears underneath.
Price Coverage. The share of distinct products with a positive latest price, next to the share that counts as fresh. Fresh means priced within the last week.
Global Attributes. Filled global category attribute slots out of the slots expected, given each product’s assigned category.
Direction. Whether the score and finding count improved, declined, moved in conflicting directions, or stayed flat compared to the previous run under the same configuration. The card reads Improving, Declining, Mixed, or Flat. Conflicting directions show as Mixed. The first run under a configuration shows Baseline.

Below the KPI cards, a Trend chart plots the Catalog Score across past runs. A Finding Mix chart breaks down the latest sampled findings by category. A Sample Findings table lists up to 30 example rows. Each row shows the affected product, the retailer, the check that flagged it, the column and value involved, the reason, and the severity. A Retailers table shows product count and price, taxonomy, and brand coverage for up to 10 retailers.

Feature Set Quality

The Feature Set Quality tab measures how completely your tenant-configured category attributes are filled in. If no tenant-required attribute slots apply yet, the tab shows a “not configured” state instead of a score. Configure category attributes in Feature Sets to start measuring it.

After configuration, the tab shows:

Feature Set Compliance. A 0-100 score that starts from your tenant attribute fill rate. Feature Set findings reduce it.
Status. Pass, Warn, or Fail, based on tenant attribute fill rate and Feature Set findings.
Workspace Attributes. Filled tenant-owned attribute slots out of the slots expected.
Findings. The total of warn and fail findings in this contract’s sample. The fail count is called out.
Direction. The same Improving, Declining, Mixed, Flat, or Baseline comparison as Catalog Quality, scoped to Feature Set runs.

The same Trend, Finding Mix, and Sample Findings views appear here, scoped to Feature Set data.

Finding categories

Row-level findings fall into these checks. Format covers malformed prices, non-ISO currency codes, and dimensions that lack expected units. Logic covers placeholder or outlier prices and contradictions between a product name and an attribute. Missing Inferred covers a required attribute that is blank while the value looks inferable from the product’s text. Leakage covers dimension or material details captured in the wrong attribute. Taxonomy Brand covers category path or brand normalization issues. Attribute-level checks (seating capacity, colour, upholstery and leg material contradictions, blank-but-inferable attributes, and completely empty attribute columns) belong to the contract that configures that attribute. Catalog-wide checks, including price, currency, price-column format, retailer and taxonomy/brand issues, and attribute leakage, always count against Catalog Quality, even when the affected attribute is one of your own Feature Set attributes. Each contract keeps its own sample budget, so activity in one does not crowd out the other’s sample. The deterministic text inference behind Logic and Missing Inferred reads product names and feature text in both English and German. It covers seating capacity, upholstery material, leg material, assembly, and style. It always reports the canonical English value, so a German catalog shows those findings the way an English one does. Colour contradiction detection still uses English colour words only.

Category Breakdown

Both tabs end with a Category Breakdown table. The table is an expandable category tree. Parent rows roll up the products and attribute fill of every descendant category. Leaf rows also list attributes.

Search and level filter narrow the tree to matching categories. Their parent rows stay visible for context, and matching branches expand automatically. Expand all / Collapse all control the tree. The table shows 30 rows at a time. A Show more button reveals additional matching rows. Each row also shows its taxonomy level, for example Level 1 for a root category, so you can see whether coverage problems sit high in the tree or only in specific levels.
Attribute Fill and Filled Slots show how many configured attribute slots are filled versus expected, for that category and all of its descendant categories. Search and level filter change only which rows are shown. The rollup numbers do not change.
Each tab scopes the table to its own contract. On Catalog Quality, leaf rows list global category attributes, and the fill totals count only those. On Feature Set Quality, leaf rows list your tenant-configured attributes, and the fill totals count only those.
Each leaf attribute line shows a value-type badge next to its fill rate and its filled-versus-configured counts. The badge is Categorical, Numeric, Boolean, or Text. Use the badge as a guide to the likely fix. A low fill rate on a Numeric attribute usually points at unit parsing. A low rate on a Categorical attribute points at value or pattern mappings.
A comparison dropdown lets you pick an earlier run to compare against. Matched category rows then show a point-change delta next to their fill percentage. Each leaf attribute line shows a stacked current/previous comparison bar next to its own delta.

Trends and comparability

Score history and run-over-run comparisons only compare snapshots captured under matching configuration. Each tab checks its own contract’s configuration independently. Catalog Quality compares runs that share the same effective global attribute definitions, which are the global attributes you have not overridden. Feature Set Quality compares runs that share the same tenant category attributes. Adding a tenant-only attribute does not reset the Catalog Quality trend, but overriding a global attribute affects both. If your relevant configuration changes, comparisons start fresh from that point. The page does not show the change as a quality regression.

When snapshots are captured

A new snapshot is captured automatically once a day. Data Quality is a reporting signal, not a gate. A failing status never blocks a product export.