364 Successful page fetches
87.1% Rows with complete live signals
42 Pages returning 404
12 Unresolved fetches
54 Rows needing refresh

Of the 418 rows in the dataset, 364 returned a successful HTTP 200 response during the live extraction. Those same 364 rows exposed structured data and a canonical tag. The remaining 54 rows were not quiet technical misses. Forty-two returned 404 and twelve could not be resolved by the collector.

That alignment is useful because it separates a content observation from a collection failure. A blank schema field on a page that never loaded does not prove the page lacked structured data. It proves the extractor did not receive the page.

The quality flag belongs beside the feature

The dataset retains fetch_status and fetch_error beside the content fields. That prevents a failed request from silently turning into zero headings, zero links, no canonical, and no structured data. Without the fetch fields, those zeros would look like page characteristics instead of missing observations.

What should happen on refresh

The 42 historical 404 rows and twelve unresolved rows should be rechecked before model training or reporting. If a page moved, the canonical URL and source page should be reconciled. If a host remains unavailable, the row can stay in the raw dataset with a missing-data flag, but it should not be treated as a successfully measured page.

Live extraction status

Fetch resultRowsShare of 418Structured data observedCanonical observed
HTTP 20036487.1%364364
HTTP 4044210.0%00
Unresolved or status 0122.9%00
Total418100.0%364364

A missing live signal on a failed fetch is unavailable data, not proof that the page lacked the feature.

Methodology

  • The collector requested each page recorded in the warehouse export and stored the returned status.
  • Structured data and canonical fields were recorded only when the page HTML could be inspected.
  • Counts were recomputed from the public CSV generated on August 20, 2026.
  • The report preserves failed fetches instead of deleting them from the denominator.

Limits

  • The audit is a point-in-time extraction. A later request can produce a different status.
  • The report measures whether structured data was present, not whether every node was complete or semantically correct.
  • A canonical tag being present does not establish that the selected canonical was the best choice.
  • Host resolution failures occurred only on two portfolio sources in this snapshot.