Legal · Corrections
What we got wrong, and what replaced it
Every figure on this site names the source it was read from. That promise is only worth something if the record of being wrong is public too. Below is every published number or verdict the registry has had to correct — what stood before, what replaced it, and why the earlier version was wrong. The log opens on 2026-09-12, the day the page was built; earlier corrections exist only in the rules history and in the commit log.
Last updated 2026-09-12
The log — 24 since 2026-09-12
| Date | Where | Was published | Now | Why |
|---|---|---|---|---|
| 2026-09-15 | Prices on the profiles of 686 stores — until 14 September for 58 of them, until today for 628 — and the deals, price-drop, was-price and weekly pages that read them; the 14 September entry below | UK prices in pounds, shown under the store’s own currency: $14.00 for a ColourPop lip product that sells for $17.00, $203 for 3sixteen jeans at $270, $49.00 for a DNA Hair Tools item at $64.95. On 14 September the UK numbers were put back and the stores’ real prices called wrong. | Each store’s own price as its product feed sends it, checked store by store against Shopify’s storefront API for the store’s home market. Price history for these stores starts again on 15 September; the UK numbers are set aside and no longer count toward any was-price, price-drop or weekly figure. At 29 other stores the feed still sends another market’s price, or the check got no answer; they stay as they were until their prices can be read for the home market. | Our nightly read runs from a server Shopify places in the UK. From 23 July, stores that sell to the UK in pounds answered it with their UK prices, and we stored the numbers under the store’s currency. On 14 September the feed switched to US-market prices — for an American store, its own — and we took the change for the fault. Shopify’s storefront API, which quotes a product for any market, shows the UK price in pounds equal to what we had stored and the home price equal to what the feed now sends. The nightly read still refuses a night in which a whole catalogue moves by one multiplier — now starting from the right price. |
| 2026-09-14 | Product prices on 653 store profiles, and the deals, price-drop and was-price pages that read them | Prices recorded on 14 September from Shopify’s international storefront — for example $17.00 instead of $14.00 on ColourPop lip products and $270 instead of $203 on 3sixteen jeans | Withdrawn on 15 September (entry above): the $17.00 and $270 were the stores’ own prices, and the $14.00 and $203 this entry restored were UK prices in pounds. | As written on 14 September, and wrong: “Shopify began answering our server as a UK visitor and sent its international-market prices under the store’s own currency label.” It was the reverse — the UK prices were what we had been recording since 23 July. |
| 2026-09-13 | Profiles: 127 stores (store names) | Names such as “Shop Now Direct”, “Welcome to Ackroyd's Scottish Bakery”, “Spring-Summer 25 is Here!”, “Nashville Flower Delivery” and “Sg” | The store’s domain, until its product feed gives a real name | The name was taken from the page title, which is often a slogan, a greeting or a region code rather than the store’s name. A name is now kept only when it matches the domain; the 127 were reviewed by hand, and real names that differ from the domain stayed. |
| 2026-09-13 | Registry: tarte, organifishop, shopbeam, bareperformancenutrition | Each store listed twice, under two of its domains (for example /store/tarte and /store/tartecosmetics), and counted twice in the niche hubs | One profile each; the second address redirects to it | The second domain redirects to the same shop, but both were added as separate stores. |
| 2026-09-13 | Feed, score history and the API: about 550 stores | Feed items such as “Printfresh: trust score 81 → 73” — “Trust score 81/100 on 2026-09-09, 73/100 on 2026-09-10, both computed under rules v0.8: the formula did not change between the two readings.” | Those days are filed under rules v0.9, so they are no longer compared with v0.8 readings | The policy-page rule of v0.9 ran from 10 September while the readings were still labelled v0.8, so our own rule change was presented as the stores moving. |
| 2026-09-13 | Profiles: every store (partner queue) | “Partner link issued the moment the program activates.” and “Your email, traffic type and audience size go to this store’s verified owner, who runs the queue.” | The queue is shown to the owner if they claim the profile, and StoreProfiles does not issue affiliate links. A store that runs its own program links straight to it. | No store has claimed its profile yet, and nothing issues links when a program starts. On 444 profiles the promise also sat beside the store’s own, already open affiliate program. |
| 2026-09-13 | Profiles: about 630 stores (structured data) | A product rating in the page markup, e.g. 4.8 from 132 reviews for a Fashion Nova product | Removed from the markup; the rating stays on the page with a link to where it was read | The figure was rounded up (4.75 shown as 4.8), and search engines’ rules for review markup forbid presenting ratings collected on another site as the page’s own. |
| 2026-09-13 | Profiles, audits and /weekly: about 770 stores | “The catalog went from 100 to 2,000 products … — up 1,900”, “Catalog grew from 500 to 4,000 products”, “Catalog shrank from 515 to 254 products” | Removed; a catalog change is stated only from full-crawl counts under the same rules version | The comparisons used rows we had stored, not the store’s catalog: the first snapshots held our old load limit (100, then 250), stored rows piled up between crawls, and a pagination fault before 12 September cut some crawls short. None of it was the store’s doing. Trends restart with rules v0.10 and return after 14 days of comparable readings. |
| 2026-09-13 | Audits and /research/was-prices: 74–97 stores | “N of M marked-down items quote a “was” price they did not sell at on any of the 30+ days we recorded their price” | Counted only on days the product was in the store’s feed | Until 12 September every stored product got a daily price snapshot, including products the feed no longer returned. Those days repeated a stale price and were counted as observations; for some stores they alone crossed the 30-day bar. |
| 2026-09-13 | Profiles: 3denergydrinks, sacheu and others | “Long stated delivery — up to 45 days per the shipping policy” | Removed where the figure came from international, APO/FPO, wholesale or made-to-order terms | The parser took the longest time anywhere on the policy page. For 3denergydrinks that was the APO/FPO line; for sacheu, international priority shipping, while US orders are stated at 3–5 business days. |
| 2026-09-13 | Profiles: fashionnova and about 100 others | “No public refund policy found”, “No public shipping policy found”, “Main caveat: no public refund policy found.” | Removed | A 404 at the store’s /policies/ address was recorded as “no policy”. Fashion Nova publishes a 30-day return policy at /pages/return-policy. A missing page at the address we check now means “not collected”, which is never a claim about the store. |
| 2026-09-13 | Profiles: about 1,600 stores | “No return window is stated on its policy pages, so none is claimed here.” | Removed | The sentence was printed whenever the parser found no window, including for stores whose policy states one (futurekind and olivercabell give 30 days). Not reading a figure is not the same as the store not publishing it. |
| 2026-09-13 | List /lists/free-shipping and profiles: hufworldwide, basksuncare, masterchow | “Shipping is free on every order” | Removed | “Free shipping on all orders over 75$” was read as unconditional, because the parser recognised a threshold only when the currency sign came first. |
| 2026-09-13 | Profiles and the API: every store scoring 65 or more | “Is X legit? Yes” | “Is X legit? On record” | The registry reads what a store publishes; it does not see deliveries, refunds or complaints (see /about). The “Yes” stood even beside a blocklist listing. |
| 2026-09-13 | Niche reports: beauty-skincare, supplements and others | “Fastest stated delivery: Alphagvrd, 2 business days” | Ranked by calendar days and printed in the store’s own unit | The ranking sorted the bare number and printed every figure as business days, so “at least 2 weeks” read as 2 business days and ranked first. |
| 2026-09-13 | Profiles: about 1,200 stores | “What would raise this score” without the step for policy pages | Listed, with its gain | The recount for that step left out the rule version the score requires, so the gain came out as zero and the step was dropped. |
| 2026-09-13 | Pages: /dataset, /api-docs | “niche … assigned from the store’s own product_type values rather than from a third-party category”; “We store the first 250 products per store” | Both definitions state what the data holds | 1,393 stores (46%) keep the vertical of the third-party catalog where discovery found them, because their product types do not point clearly to one vertical. And stored rows piled up past 250: 1,029 stores had more, fashionnova 9,176 with 250 seen in the last crawl. |
| 2026-09-13 | Pages: /store/{slug}/similar (3,081 in the sitemap) | “Stores like X — 12 audited alternatives”, with rows that shared no product category with X | Only stores that share at least 30% of product categories with X are listed; a page with fewer than 3 is kept out of search and the sitemap | The nightly rebuild of product-category tokens deleted and re-inserted them in one statement, so the table emptied every other night and “similar” fell back to trust score and price alone. The rebuild now runs as two steps in one transaction, and a store has to share categories to be called similar. |
| 2026-09-12 | Profiles, home page and feed: about 2,000 stores | “Main caveat: external reviews not yet collected — score may move once they are” | Removed; a gap in our coverage is no longer listed as the store’s weakness | It was a note about our own crawl printed under Weaknesses. External reviews had been collected for 1,020 of 3,010 stores, so two profiles in three carried our blind spot as a finding about the store. |
| 2026-09-12 | Profiles: 580south, 1951coffee, banditcoffee and five others | Store names “Home”, “Shop” and “Coffee Roaster” | The store’s domain label | The names came from the label a third-party catalog gave each store, which was the heading of its front page rather than its name. |
| 2026-09-12 | Profile: mudwtr | Long stated delivery — up to 42 days | Removed | The 42 days came from a footnote citing a Healthline article that happened to sit on the store’s policy page, not from anything the store promised about delivery. The parser now reads the policy text without page furniture, and the audit ignores parses made by rules it has since retired. |
| 2026-09-12 | Profile: meedenart | Short 7-day return window | Removed | “If no refund is received within 7 days, please contact us” tells a buyer when the money should arrive. It is not the window for sending an item back — the store gives 30 days for that. A number is now read as a return window only when the sentence around it says the goods go back, and says nothing about when a refund lands. |
| 2026-09-12 | Profiles: mcclothingandgifts, rosenskincare, longgroveconfectionery | Short 7-day / 10-day return window | Removed | All three were parsed before the registry recorded which version of the rules produced a fact. The profile already hid those readings; the audit still built weaknesses from them. Both now read the current parser only. |
| 2026-09-12 | Page: /deals | “147 on sale of 250” | “147 on sale of 250 sampled” | 250 is how many products the registry samples per store, not how large the catalog is. Stated as a catalog size it made a discount share look precise when it was a share of the sample. |
How a correction starts
Three ways, and the third is the one that has found the most so far.
- A reader. Every profile carries Something here wrong? Report it, and each strength and weakness has its own report link that names the point you are disputing. No account, no email required.
- The store. An owner who has verified the domain can answer on the record and ask for a figure to be re-read — see for stores.
- Our own audit. Eight checks run against the published site and the database together: a quoted number has to appear inside its own quote; a store cannot hold two facts that contradict each other; a threshold written in prose has to match the constant the code filters on; a percentage has to carry the population it is a share of; a counter beside a page has to use that page’s own filter; published totals are re-derived from the database; a lower bound is never printed as an exact figure; and a claim of freshness is compared against how old the facts actually are.
What happens to a report
It is read against the source cited on the page, not against our opinion of the store. If the source says something else, the figure is corrected or removed and the change is written here with its date. A fact we cannot stand behind is removed rather than defended.
Reports are a queue for manual review, never published automatically, and are not shown on the store’s page. An email address is optional and used only to reply about that report ([email protected] reaches the same queue).
What is not a correction
- A fact we do not have. Absence never counts against a store, so there is nothing to correct when a signal is missing.
- A store left out of a list because its evidence failed a check. That is the rule working. Each list says how many rows it held back.
- Disagreement with a verdict the store’s own policy text supports. The route there is the right of reply on the profile, not a change to the reading. How the score is computed is set out in how we score.