A used Raspberry Pi case asked 31% above new while the retail anchor returned no results

The desk ranks its evidence: completed sales of the same product first, then live asking prices carrying a warning that nobody has paid them. Alongside both sits a retail anchor, the new price scaled down by a used-goods factor and independent of every resale price ^[4]. Each of those is a vote. A vote can be absent and the run still finishes. On Friday the anchor's discovery step returned nothing, because its only route to a price-comparison site was a `site:` search through a self-hosted metasearch instance whose engine had been rate-limited since breakfast ^[7]. The run logged `retail-anchor discovery unavailable: no results and these engines were unresponsive: google cse` ^[6]. The anchor, the author wrote, "said so, politely, in a log line nobody reads before a listing goes up" ^[8].

Feeding it by hand changed nothing about the output. With the shop price typed into the item's notes, the anchor produced its number. The crosscheck reported that "the price sources disagree with each other (live asks £42.29, retail anchor £12.62)", marked the price "thin", and left it at £35.39 ^[13]. The anchor was one vote among several, and the design was careful never to let a single source dictate ^[14].

So the absent anchor and the present anchor produced the same listing. £35.39 against a new price of £26.99 is 31% above retail ^[21]. The anchor's own reading, £12.62, is about 47% of the new price, which is what the category factor estimates a used one realises ^[23].

The fix sits outside the scoring. Once the new price is known, the ask is capped at that price times a condition multiplier: 85% for "good", 92% for "like new", 70% for "fair". That multiplier is explicitly not the category factor, which belongs to the anchor ^[16]. For this case the cap is £26.99 times 0.85, which is £22.94 ^[22]. That is £12.45 below the number the pipeline produced with nothing broken, a cut of 35% ^[24].

The manual override cost two parser changes. "New" is already a condition keyword, and it tells the comp filter to discard every used sale. So "new on amazon £26.99" in the notes silently reclassifies the item, unless the price phrase is read and removed before the condition words ^[11]. The second problem is which keyword owns the pound sign. In "brand new, sealed, rrp £30" the price belongs to `rrp`. The author's first pattern matched from the first keyword and swallowed "new, sealed, rrp £30" whole. The item stopped being new. A negative lookahead in the gap between keyword and `£` fixed it, with a test ^[12].

Typing prices by hand works one item at a time. The comparison site indexes electronics well and everything else badly, and the stock includes a tin of malt drink powder, a newborn bottle set and a graphing calculator, none of which it has ever anchored; Amazon sells all of it ^[18]. Before writing a scraper the author checked that Amazon's `robots.txt` does not disallow the search page or product pages for `User-agent: *`, only sub-paths the tool never touches ^[19]. The client identifies itself as `listing-desk/2.31` with a link to the source, with no browser string and no headers pretending to be Chrome ^[20]. The post says three things were measured and breaks off partway through the second ^[25].