Netherlands and Belgium marketplace data, with the seller table as the cheap call

This is the reason market is a real parameter. Re-measured live for this page: a Sony MDR-ZX110 headphone was 15.00 EUR on the Dutch storefront and 13.05 EUR on the Belgian one, on the same product id, in the same minute. be and be-fr are the Dutch and French storefronts of Belgium, and a French search returns French titles and French category names for the same ids. If you are price-monitoring the Benelux you need both countries, not a translation.

All three storefronts quote EUR, so unlike a multi-country marketplace there is no conversion question here - a cross-border difference is a real price difference rather than an exchange-rate artefact. That is what makes the two-call comparison above meaningful.

A bol URL carries a slug, but only the numeric id resolves it, and every action takes the bare id or a full URL. The same holds for categories, and there it is a trap: a category URL carrying the koptelefoons slug with id 4770 serves Laptops, because 4770 IS Laptops. So the category action echoes bol's own name for what it actually resolved, which makes a wrong id visible instead of silent - three guessed ids during the build resolved to some other category rather than failing.

bol publishes its whole competing-offer table on its own page, and that page is far smaller than a product page - measured at a median of 1,224 ms and 45-48 KB against 1,550 ms and 288 KB for product/detail. It is priced accordingly, at one credit against two. If what you want is who sells it and for how much, go straight there rather than reading the product first.

A measured call on an Anker headset returned 6 sellers between 84.00 and 92.71 EUR - Anker Direct, ASD, gsmpunt, Eufy Official, bol itself and IT Distributie - each with price, condition, shipping cost, the date it arrives, whether that is within 48 hours, and a bol seller rating out of 10 (8.7, 8.4, 9.3 on those rows). bol's own row has no rating, because bol is not one of its own marketplace partners; that comes back null with an explicit is-bol flag rather than a flattering default. Across a seven-product chain the action returned 1, 1, 1, 1, 2, 3 and 6 offers.

Search page 11 returns 0 rows and a total of 0, so the limit is published on the parameter and page 11 is rejected rather than charged for; category browse pages far deeper, with one measured category reporting 114 pages. And the total itself is not always a fact: bol flags some searches as early-terminated and the number then moves between identical calls - one query measured at 10,006 and then 60,468 - while very large categories clamp at exactly 300,000. The response carries an explicit is-estimate flag so you know which kind of number you are holding.

0 of 199 measured listing rows carried an EAN, in any department. That is bol's shape rather than a parse gap - a listing card ships a summary spec block and zero spec groups, and EAN lives in a group. product/detail has it, and it was real on 7 of 7 products checked, one per department. If you are building a catalogue keyed on EAN, budget the second call.

Brand was filled on 5 of 30 rows in the books department, because a book has no brand on bol - it has an author, which bol publishes as a specification. Films filled 28 of 28 with a real distributor name, so the field itself works. Ratings behave the same way: 6 of 28 film rows had one, because most of those titles have no reviews yet. Written down rather than averaged away.

What looks like an RRP is bol's own reference price, which bol defines as the most-shown price of the last 90 days excluding promotions, and it is returned under that name on roughly the 40% of rows that have one. The discount is computed against that. So a discount here means cheaper than bol has typically been showing it, not cheaper than a manufacturer's list price - which bol does not publish anywhere in the payload.

bol models a colour or a size as its own product id rather than as a variant of one. product/detail returns those siblings with the feature that distinguishes them, so you can walk to them, but there is no size matrix to read and none is invented - and it means price and seller can differ per colour. Separately, bol translates spec VALUES per storefront, not only labels: the same product's Model reads "Ace" on the Dutch storefront and "As" on the Belgian French one. The value you get is the value that storefront publishes; inventing a canonical one would be fabricating it.

bol publishes a delivery promise, a latest-arrival date and a scarcity flag, and there is no stock number anywhere in the payload. Review text sits behind an authenticated call that answers 403 without a session, so only the average rating and the review total are returned. Seller address, chamber-of-commerce number and phone live on a separate bol page this engine does not read. All of it comes back null with the reason, rather than estimated.

135 live calls across three separate runs came back 135 successful, median around 1.8 to 2.2 seconds; per action, offers 1,224 ms, detail 1,550 ms, search 2,219 ms, category 2,694 ms. The tail is real and is stated rather than hidden: one search in 135 took 17.2 seconds. An occasional block is typed as retryable so it is retried rather than surfaced as a dead end. No storefront is meaningfully slower than another.