# Verification notes

Names and direct source identifiers are withheld. These summaries retain the original observations and interpretation. Full captures and identity mappings are retained in the private research archive.

## Coverage and recounts

The final cohort is 486 websites and 2,430 analytic pages. A separate raw-source census located the primary VIN-bearing Vehicle/Car node on all 1,458 VDPs. Direct AI-assisted review covered 132 pages from 44 websites, 95 selected field records and 70 rendering-category checks on 41 final pages, plus targeted technical, absence and dynamic checks. Supplementary September 8 mileage review covered 13 examples. This was not independent human adjudication or a certification of every field.

## Mileage

One audited used-vehicle page displayed 41,639 miles while omitting the mileage property from its primary-vehicle markup. This demonstrates that a missing structured field does not mean the shopper could not see mileage.

One vehicle advertised as new had 0 miles in the primary vehicle's markup and 4,384 miles in visible content. Both referred to the same VIN, withheld here. A fresh capture reproduced the difference. This is an observed content inconsistency, not proof of the physical odometer reading or an overall dealer error rate.

## Price interpretation

One dealer's $35,125 structured amount excluded a $499 fee, while the $35,624 visible amount included it. Another dealer's $60,730 and $61,530 values corresponded to MSRP and a price including an $800 processing fee. Neither was counted as a confirmed price error. Other price differences remained unresolved when labels or qualifications did not establish a comparable basis. Matching numbers alone do not certify every transaction term.

## Inventory coverage

One fully mapped results page displayed 15 vehicle cards. Its 12 ItemList detail links corresponded to the first 12 cards; the remaining three displayed vehicles lacked a corresponding list entry. Some cards carried sold graphics, so displayed cards are not equivalent to available stock. A larger full-inventory total was not used as the denominator. This inspected example does not establish a study-wide inventory-accuracy rate.

## Results-page model catalog

All 149 Provider B results pages included inventory detail links in ItemList. One also declared ten Car nodes describing model families inside a business offer catalog. Those model nodes lacked stock VINs and detail-page URLs; the stock links were in a separate ItemList. Thus a Vehicle/Car type count alone is not a test of stock-level inventory identity.

All 27 Provider D results pages acquired Vehicle/Car declarations after rendering. These are implementation patterns associated with the sampled groups, not evidence of provider responsibility or search performance.

## Technical findings

Direct inspection confirmed four malformed JSON-LD block hashes across four websites, eight selected pages and 11 paired-state occurrences. Valid blocks elsewhere on the same pages retained their positive presence status. A separate homepage used the nonconforming Microdata type URL `//schema.org/AutoDealer`; itemtype requires an absolute URL. These findings are limited to the checks performed. Optional omissions, multiple blocks and unfetched external references were not automatically errors.