What 8,450 Appliance Records Actually Contain: A Self-Crawl

I wanted a real coverage number for our own appliance database, not a vibe. So I crawled it. The trigger was boring. Someone asked for a Maytag dryer fault code and landed on a page that listed twelve washing machines and mentioned, in passing, that one of them had a code list. Their model wasn't in the twelve and wasn't a washer. The page wasn't lying. It just wasn't useful, and the gap between those two things is what I wanted to measure. The crawl Our sitemap index for models held 4,800 URLs. 817 are brand × appliance index pages, and each one embeds a tally of itself in prose: "9 Kenmore clothes dryers documented. Together they carry 0 published replacement part numbers, and 0 of them have a confirmed error-code list." That's three integers per page, so a full catalogue total costs 817 requests instead of 8,000. Roughly: PAT = re.compile(r"(\d+) .{0,40}? documented\. Together they carry (\d+) " r"published replacement part numbers, and (\d+) of them " r"have a confirmed error-code list") Fetch with curl over a proxy, ten workers, parse with the pattern above. 815 of 817 matched. The two that didn't were left out — if a page doesn't print a number, inventing one turns an audit into fiction. Elapsed: about four minutes. 8 November 2026. Results Measure Value Index pages parsed 815 Brands 503 Categories 9 Model records 8,450 Replacement part numbers 3,800 Pages carrying any part number 57 (7.0%) Models on pages with zero part numbers 6,369 Pages with any confirmed error-code list 2 (0.25%) Pages with exactly one model 229 (28.1%) Median models per page 3 What the distribution looks like Part numbers are concentrated, not sparse. Refrigerators: 249 pages, 4,120 models, 1,570 part numbers. Frigidaire's refrigerator page alone publishes 448. Whirlpool adds 242, LG 267. Air purifiers: 66 pages, 196 models, 0 part numbers. Range hoods: 70 pages, 907 models, 62 part numbers. So a dishwasher owner and an air purifier owner get two different products on the same domain, and nothing on either page tells them which one they're looking at. Three findings worth keeping The error-code field is a rounding error. Two pages of 815. Both washing machines. The title tag sells error codes; 0.25% of the catalogue delivers them. Three quarters of models are on pages with no part numbers at all. 6,369 of 8,450. That isn't "coverage is thin" — it's "coverage is one category deep and empty elsewhere". The median page has three models, and 229 pages have one. The long tail is generated, not curated. One indexed model, one page around it, a category headline on top. What I did not test I didn't crawl the 3,672 model-level pages (manual, specs, parts, problems). An index dash might not reflect what a model page holds. Unmeasured. I didn't validate a single part number against a manufacturer parts catalogue. Presence and correctness are different tests and I only ran the first. I didn't investigate why 503 brands yield 8,450 models — dead brands, rebadges, regional naming. Out of scope for a four-minute crawl, and I'd be guessing. Fixes queued The parts copy claims "part numbers with the symptom each fixes". Symptom data doesn't exist. That claim is being removed. An internal note of mine said "8,450 appliances with manuals". It's 8,450 model records. Manual links weren't counted here, so that sentence overstated by an unknown amount. Disclosure I run ModelNumberFinder. The crawl, the regex and the two excluded pages are all above so the numbers can be reproduced or refuted. Nothing here was sponsored and no manufacturer reviewed it. Dryers still don't have codes. Washers are next because 44 washing-machine pages already hold 664 part numbers, which gives the code list something to attach to. Whether that ordering survives contact with the actual inbox is a separate question.