A raw sold-listings search is not automatically a usable comp set.
Search for a phone and you may also get cases, chargers, broken screens, empty boxes, and nearby models. Search for a camera lens and you may get caps, adapters, or a different focal length. If those rows go directly into a median, the result can describe the search noise instead of the product.
I wanted a larger measurement than a single convenient example, so I ran a fixed 100-product study through CompSniper, the sold-price API I own.
The goal was not to prove that an automated classifier is always correct. The goal was narrower:
Measure what the production relevance cleaner removed and how the product-level median changed on one predeclared sample.
The protocol
I selected the products before making the first request:
- 20 smartphones and tablets
- 20 gaming and computing products
- 20 cameras and lenses
- 20 audio and music products
- 20 collectibles and luxury products
Every search used the same settings:
Marketplace: ebay.com
Sold window: 2026-06-02 through 2026-08-31
Page: 1
Requested rows: 240
Sort: ended recently
Condition: any
Relevance cleaning: enabled
Each relevance-enabled response contained the raw sample count and raw median captured before classification, followed by the cleaned rows and deterministic price summary from the same fetched page. That meant one production request per product, not separate raw and cleaned fetches.
All 100 requests succeeded with unique request IDs.
The headline results
Across the study:
- 19,220 priced raw rows were parsed
- 11,942 priced rows remained after cleaning
- 7,278 rows were classified out
- The weighted removal rate was 37.87%
- 34 of 100 product medians changed by at least 10%
- 15 of 100 changed by at least 25%
- 11 of 100 changed by at least 50%
The direction was not always upward:
- 73 medians increased
- 21 medians decreased
- 6 medians stayed unchanged
That is important. The cleaner is not instructed to raise prices. It tries to retain listings for the requested product. Removing bad matches can move the median in either direction.
The largest movements
Here were five of the largest increases in this sample:
| Product | Raw rows | Cleaned rows | Raw median | Cleaned median | Change |
|---|---|---|---|---|---|
| GoPro Hero 12 Black | 219 | 61 | $59.99 | $230.00 | +283.40% |
| Lenovo ThinkPad X1 Carbon Gen 11 | 222 | 14 | $293.50 | $602.49 | +105.28% |
| Canon RF 70-200mm f/2.8 L IS USM | 224 | 46 | $1,104.50 | $2,097.00 | +89.86% |
| NVIDIA GeForce RTX 4090 | 233 | 51 | $1,399.00 | $2,599.99 | +85.85% |
| Sony Walkman TPS-L2 | 114 | 45 | $328.55 | $573.00 | +74.40% |
The largest decrease was the Google Pixel 7 Pro 128GB search, where the raw median was $424.09 and the cleaned median was $199.99, a 52.84% decrease.
These outliers are reasons to inspect the row-level evidence. They are not proof that every removed listing was wrong.
Removal rate and median movement are different metrics
The median product-level removal rate was 28.85%, but the median absolute median-price change was only 4.2%.
That is a useful distinction. A search can contain many cheap accessories spread across the results without moving the center very much. Another search can contain fewer mismatches concentrated far away from the actual product price and move the median substantially.
At the category level, the weighted removal rates were:
| Category | Weighted removal rate | Median absolute price change |
|---|---|---|
| Collectibles and luxury | 47.05% | 8.84% |
| Smartphones and tablets | 45.99% | 4.74% |
| Audio and music | 40.60% | 7.03% |
| Cameras and lenses | 30.31% | 0.52% |
| Gaming and computing | 26.28% | 3.21% |
Best Offer remains a real limitation
The cleaned sample contained 2,184 rows marked as accepted Best Offer, or 18.29% of cleaned priced rows.
eBay indicates that an offer was accepted but does not disclose the accepted amount. The visible listing price may therefore differ from the actual transaction price. If your workflow requires exact transaction-level prices, exclude rows where bestOfferAccepted is true.
What this study does not prove
This was not a human-labeled precision and recall evaluation.
It used:
- one marketplace
- one collection run
- one query per product
- page one only
- title-based classification
- a bounded recent sold window
Different wording, dates, pages, marketplaces, and condition filters can produce different samples. A cleaned median is still historical sample evidence, not a formal appraisal or a guarantee of the next sale price.
Reproduce or inspect it
The complete report includes category charts, the 15 largest absolute changes, all 100 predeclared products, integrity checks, FAQ, and the full methodology.
- Read the complete study
- Download the aggregate CSV
- View the product list, collector, and analysis code
- Read the CompSniper methodology
The public CSV contains product-level counts and statistics. It excludes listing titles, seller data, API keys, and internal request IDs.
Disclosure
I am Marc Andrew, owner of CompSniper, and this is first-party research produced with the CompSniper API. CompSniper is an independent product and is not affiliated with or endorsed by eBay Inc.
Top comments (0)