Scrape Booking.com hotel details from hotel URLs: room types with amenities, facilities by group, photos, address, coordinates, review score, stars and the cheapest offer for your dates (total, taxes, free cancellation, breakfast). One flat row per hotel, no personal data.
History of booking-hotel-details-scraper (folder actors/booking-hotel-details). Versions are the Apify build numbers.
Unreleased (clean stop before the run's time limit, 2026-11-07)
Brief docs/cloud-tasks/2026-11-07_studio-core-deadline.md (cause n°4 of docs/research/2026-11-07_audit-echecs-utilisateurs.md: nothing watched the run's time limit; the public tasks set timeoutSecs: 300, agents often less). .actor/actor.json stays at 0.1 (Apify versions are MAJOR.MINOR, the patch is the build number). Output columns, prices, defaults and the daily-check input are unchanged.
Clean stop before the time limit: the shared runner reads the run's time limit (ACTOR_TIMEOUT_AT) and stops starting new work min(60 s, 15 %) of the run's time before it, at least 15 s (@scraper-studio/core/deadline, 45 s for the 300 s of the public tasks): no hotel is started after it; the hotels in flight finish (page, offer and their second calls), the hotels left are named in a warning, and such a run is never a legitimate empty run (not every hotel was run). The requests in flight finish; RUN_STATS is written (with stoppedBeforeTimeLimit and timeoutAt); the status says "Stopped before the run's time limit: results; run again or raise the timeout for more", followed by the notes of status.ts; the run ends in success when it saved rows, instead of being killed as TIMED-OUT without status nor statistics. A run without a time limit behaves as before.
Proxy country refused at the start: when Apify Proxy refuses the proxy country of proxyConfiguration (the SDK's access check), the run goes on without the country, with a warning; any other refusal fails at once with a clear message in the log and in the status (openProxyConfiguration).
hotelUrls is required in the input schema (required, allowed by the shared Store checks when every required field has a valid prefill): an API or agent call without it is refused by the platform before the start, with no failed run; its prefill fills it for Apify's daily test of the Actor.
Unreleased (robust to real users' inputs, 2026-11-07)
Brief docs/cloud-tasks/2026-11-07_booking-details-repair.md, after the audit of the real users' failed runs (docs/research/2026-11-07_audit-echecs-utilisateurs.md: 83 % public success rate, 3 failures out of 18 runs). Written offline on the committed fixtures and simulated responses.
GraphQL ValidationError on one hotel: BookingDetailsSchemaError (the query rejected outright, a schema change) is now raised only for errors in a response without data, for the page and for the offer; next to data, the page is read with its errors logged, the offer is read when priced and otherwise counts as a failed offer, a page without the hotel is retried once on a new session then given up. Any exception of one hotel (rejected query, network, storage) is counted in errors, logged ("hotel skipped (…)") and the other hotels go on. A rejected query stops the run only while it has saved nothing, at the second hotel rejected (REJECTIONS_BEFORE_STOP; the first for a one-hotel run), so that a schema change still fails fast and loudly; after a saved row it only fails its hotel. New RUN_STATS.extra counters: hotels, referencesSkipped, queryRejections.
Dates and ages corrected instead of refused (warning in the log, note in the status message): a check-in alone stays 3 nights (check-out = check-in + 3, across months and years), a check-out alone is ignored (stay in 30 days, 3 nights), dates in the wrong order are swapped, a check-out on the check-in day stays 3 nights, children is raised to the number of ages in childrenAges. A date that does not exist (2026-02-30, 2027-02-29) is refused at the start with the date and the expected format (real calendar check).
Entries that are not hotel pages only skip themselves, with the reason: a share link (booking.com/Share-…, its redirection is not resolved: open it and paste the hotel page URL), a search results page (use booking-hotels-scraper), another Booking.com page, another site. The run fails before the crawl only when no entry is left, with the number of entries skipped. A missing hotelUrls fails with "hotelUrls is required: …" and the names of the fields received (an agent that wrote urls or startUrls).
Empty runs (CLAUDE.md § 8, exception of 2026-11-07): a run that saves nothing succeeds only when Booking.com answered "not found" for every hotel asked, each page read cleanly (every hotel run, no blocked request, no error, no entry skipped); it then ends with the status message "0 results: Booking.com found nothing for these inputs; nothing was charged ( hotel pages not found: removed from Booking.com or unknown)" and a warning. Every other empty run still fails. The closing status message (status.ts, set from the exit listener after the runner's) also lists the input corrections, the entries skipped, the pages not found and the hotels that failed.
Input schema: hotelUrls has minItems 1 and maxItems 1,000, and its item pattern accepts only hotel pages (/hotel/<country>/<name>… on any Booking.com domain, m.booking.com, language suffixes, query parameters) or <country>/<name>: search pages and share links are refused before the start; a test binds the pattern to parseHotelUrl. required stays empty: the shared runStoreChecks of packages/core asserts it, lifted by docs/cloud-tasks/2026-11-07_studio-core-deadline.md, which then adds "required": ["hotelUrls"]. Descriptions of hotelUrls, checkIn, checkOut, children and childrenAges follow the new rules.
README: the forms accepted and refused, the date and age corrections, the empty-run rule, one failed hotel never stops the others. Unchanged: output columns and schema, prices, title, categories, the prefilled form and the daily check input.
Unreleased (agent-facing polish: field order, first screen)
Brief docs/cloud-tasks/2026-10-24_studio-agent-facing-polish.md (2026-10-24): what an AI agent reads first. No change to the code, the defaults, the prices or the output.
Input field order (the Apify MCP search card lists the first 20 fields): unchanged: every field is among the first 20 and the proxy and crawl settings already close the form. Nothing renamed or removed.
README first screen: under the title, "Verified daily by an automated run on Apify with the default input", the Free-plan and Business-plan prices per 1,000 results of the pricing file in force and the $0.00005 start fee per run (Apify's minimum); a one-line review request before the legal note. A test binds these figures to the pricing file and forbids "to confirm" and "no start fee" in the input schema and the README.
Unreleased (all-in price: taxesIncluded and priceTotalAllIn)
Brief docs/cloud-tasks/2026-10-23_travel-prices-all-in.md (2026-10-23): rows are travel-listing/1.6 (the detail schema on top of the search schema travel-listing/1.5), two columns after taxesAndFees, nothing else changes (priceTotal keeps its meaning: the cheapest offer before the charges Booking.com shows as not included).
taxesIncluded (read by parseOfferResponse, new OfferPrice.taxesIncluded): false when the offer shows an excluded amount above 0 (both priced offers of the committed captures), true for an excluded amount of 0, null without that amount, with one in another currency, or without an offer.
priceTotalAllIn = priceTotal + taxesAndFees when taxesIncluded is false, priceTotal when it is true, null otherwise (allInPriceTotal of @scraper-studio/core/schema): €289.81 + €41.39 = €331.20 for the README example, $484.40 + $85.05 = $569.45 for the aparthotel.
Dataset schema: both columns documented; Overview shows priceTotalAllIn next to priceTotal, Prices shows both after taxesAndFees. README: Which price to compare paragraph, sample output with priceTotalAllIn (instead of checkOut), FAQ Which price is in the row?, the strengths worth knowing when switching (hard maxResults cap never charged beyond, unknown pages not charged), More scrapers pointing at priceTotalAllIn.
Unreleased (quality score: input documented for AI agents)
Brief docs/cloud-tasks/2026-10-20_studio-quality-score-insights.md: the input form passes the shared Store checks (runInputQualityChecks, runSearchTextChecks of packages/core/src/testing/store-check.ts).
Descriptions rewritten as a tool documentation (what the field does, format with an example, default and its reason, bounds): checkIn, checkOut, adults, children, childrenAges, rooms, includeOffer, maxPhotos, currency, language, maxResults, proxyConfiguration, maxRequests, delayMs, debugStoreResponses.
maxResults prefilled with 50 (default still 100): Apify's daily test of the public Actor runs the prefilled form, a 3-hotel run (the three prefilled URLs); 50 rather than 3 so that a form user who pastes their own list is not cut at three hotels.
inputs/daily-check.json = the prefilled form: the same 3 Lisbon hotel URLs and maxResults 50 (was 3: same three rows; the row count of the daily check is not checked by scripts/release-actor.mjs); maxPhotos 10 removed (the form's 45 applies: same requests, longer photo lists), keys equal to the defaults removed (adults 2, includeOffer true, currency EUR, language en-us). store.json › dailyCheck.what updated; the live daily-check task follows when the pilot re-sends it.
Store texts carry "Booking.com hotel details": actor.json description ("Scrape Booking.com hotel details from hotel URLs: …") and store.json SEO description ("Scrape Booking.com hotel details by URL: …"); title and README line 1 unchanged. store.json › searchSynonyms: Booking.com scraper, room prices, room types. README input table: maxResults, maxRequests rows.
Example tasks: the 8 SEO titles now carry the long-tail search "… – Booking.com hotel details" instead of "… – Booking.com details" (e.g. "Lisbon hotel prices next month – Booking.com hotel details"); published with scripts/publish-tasks.mjs at the publication of the Actor.
No code change: rows, logs and charges are identical.
Unreleased (README visibility)
README first screen and links (brief docs/cloud-tasks/2026-10-20_studio-readme-visibility-new-actors.md, 2026-10-21), same layout as the public Actors (PR #50). The Store search indexes only the first ~3,000 README characters, so they now hold: an opening sentence with the phrases agents search ("Booking.com hotel details", "room prices", competitor benchmarking, room price tracking with scheduled runs), a "No login, no cookies, no personal data" line, the Free-plan price in figures in How much ($4.21 per 1,000 hotels; Apify's $5 monthly credit covers about 1,187), computed from pricing/2026-10-20_pay-per-event.autopilot.json (Free tier of the primary event, one apify-actor-start event deducted), and a real output snippet (12 key columns of the README example row, the cheapest room in highlights). Further down: Use it from your AI assistant (MCP URL https://mcp.apify.com?tools=myth84/booking-hotel-details-scraper and a ready-to-copy prompt with a minimal input that never expires), Alternatives and switching (official API, home-made scraper, input-name mapping table, no competitor named), More scrapers by Flatrow Data Studio (Store links to the travel Actors already public, the Booking.com search first) with an invitation to rate the Actor (no counterpart), and the non-affiliation line at the very bottom. The cost section and the MCP cost bullet quote the same figures; the line 1 heading is unchanged. Checked by runStoreChecks (readmeVisibility). No code change: rows, logs and charges are identical.
Unreleased
Output schema travel-listing/1.3 (additive, follows docs/cloud-tasks/2026-10-20_travel-extend-monitor.md, PR #34): the search schema of the travel studio gained the price-monitoring block (changeType, previousPrice, priceDelta, priceDeltaPercent) and took the tag travel-listing/1.2; the detail schema this Actor shares follows as 1.3 so that one tag means one column set. The four columns sit after scrapedAt, before the detail columns, and are always null for this Actor (no monitoring mode). Dataset schema regenerated, README example and column table, output schema description and tests updated; no other change in the rows.
Release fix (brief docs/cloud-tasks/2026-10-20_booking-fix-release.md): the first release stopped at the control step (run ZavxrNOc583Z1E8OD: 28 rows, 29 expected) because inputs/control-scale.json listed pestana-palace-lisboa, a hotel page guessed in the Booking audit (probe input booking-run3.json, label hotel-page-guess, never answered) and not a page name returned by Booking.com; an unknown page gives no row by design (HotelPageNotFound: not charged, not an error). The control input now lists the 28 pages returned by Booking.com in the captures (maxResults 28); the five example tasks that carried the same guess use a confirmed page instead (almirante, vip-executive-suites-do-marques, family-refuge-lisbon) or one page less (upscale hotels: 7, facilities audit: 28). No code change.
Test (src/tasks.test.ts): known pages come only from Booking.com responses (city landing page, search results), no longer from probe inputs; every file of inputs/ lists known pages only, without duplicates (except the deliberately unknown page of validation.json), and its maxResults equals their count — the exact row count scripts/release-actor.mjs checks.
0.1 — 2026-10-20
First version, written offline on the fixtures captured by the probe on 2026-10-20 (audit docs/audits/booking-details.md, brief docs/cloud-tasks/2026-10-20_booking-details-build.md): two calls per hotel to Booking.com's internal GraphQL API (POST /dml/graphql, query texts sent as is, no cookie, no token) — PropertyPage (hotelPageByPageName) for the hotel page, then a one-row FullSearch on the hotel itself (destType: HOTEL) for the cheapest offer of the dates and guests, asked once more after a cold answer (no block, no price, or an irene error), as measured by the probe.
Output schema travel-listing/1.2 with ratingMax: 10: description, facilities by group with their payment notes (amenityGroups), facility titles (amenities), one line per room type with its amenities (sleepingArrangements), gallery photos (maxPhotos, 45 by default), short and full address (neighborhood, neighborhoodDescription), district (region), coordinates, review score and count, stars, accessibility, Booking.com's generated summary (locationSummary), DSA trader status (listedBy); offer: priceTotal, pricePerNight, priceBeforeDiscount, taxesAndFees, cancellationPolicy (free_cancellation_until:<date>), isAvailable, badges (meal_plan_included, third_party_offer, deal badges). Spoken languages, location flags, programmes, certifications and the room type of the cheapest offer go to highlights.
Decisions: one sticky datacenter session for the whole run (four hotels at a time), replaced once on an unusable response, then the hotel is given up; HotelPageNotFound gives no row, is not charged and is not an error of the run; a GraphQL validation error (schema change) fails the run at once instead of rotating through every hotel; an offer still empty after the second call gives isAvailable: false, an offer that could not be read (request failed or errors twice) gives isAvailable: null with empty prices (unknown is not "unavailable"); propertyTypeCode is Booking's accommodation type (HOTEL, APARTHOTEL), roomType is entire_place for apartments and aparthotels, hotel_room for hotels; facilities whose group is not described on the page go under "Other"; dates default to a check-in 30 days from the run date for 3 nights so that schedules and example tasks never expire.
Not included in v1 (audit § 3, documented in the README): the per-room rate table, house rules and check-in / check-out times, category scores, availability calendar, reviews and anything about hosts.
Pay-per-event result per hotel pushed; explicit failure when no row is produced; RUN_STATS with hotel pages, offers, empty and failed offers, second offer calls and unknown pages.