Multi-operator accommodation comparator for a geographic area against the user's stay parameters — dates, guest count, optional filters. Returns a ranked list of properties together with the booking sources that offer each one and, when dates are passed, their live availability and per-operator price for the requested window.

Natural-language date references — "tonight", "this weekend", "next weekend", "the weekend of July 4", "Memorial Day weekend", "long weekend in May" — translate to concrete check_in / check_out values at the call site; concrete ISO dates also work.

user_country, currency, and language carry the user's locale, not the destination's. IMPORTANT — currency: prices are returned in currency if you set it, otherwise in the currency derived from user_country (US→USD, CA→CAD, GB→GBP, euro-area→EUR); if you set NEITHER, prices default to USD, which may not be the user's currency. So whenever you know where the user is (or what currency they want), pass user_country and/or currency — do not rely on the default. Prices are never converted client-side; each offer is quoted by the operator in that currency. user_country and language also localize the booking link (web_url). The user's own residence/billing country is the right user_country (not the destination's), and their interface language the right language.

Each result is shaped for downstream presentation without extra calls:

• location.lat and location.lon carry per-property coordinates, suitable for plotting all results on a single map so the user can compare spatial alternatives at a glance. The map widget reads these fields directly from this response — no separate lookup needed for visualization.

• thumbnail_url carries the property's first photo URL when available (null when no image is on file); useful for embedding inline or showing on the map alongside the pin.

• images on search results is capped to the first photo to keep the comparison payload compact; each item has a url field, and thumbnail_url mirrors images[0].url. Call get_property_details for a single property to retrieve its full photo gallery.

• web_url is a ready-to-open booking link for the property, already encoded with the user's check-in/check-out, language, currency, and guest count. Pass it to the user verbatim when they ask for a booking link — never reconstruct the URL from individual parameters, the query-string format is not guaranteed to match generic booking-URL conventions.

• Price is live and date-specific only. There is no date-agnostic "from" figure: a meaningful price only exists for a concrete query (property + dates + occupancy).

  • price and offers[] — the live quote for the requested dates, populated only when dates were passed and the comparator confirmed availability. offers[0] is the curated best; each offer carries amount (total stay), amount_per_night (per-night), currency, breakfast_included, refundable, rooms_left, and deeplink_url. price mirrors offers[0].

  • With no dates (or when nothing is available) price is null and offers is empty — surface the property without a price rather than inventing a starting figure.

• availability_status per result encodes the live state:

  • available — bookable rooms confirmed at the operator level. offers and price carry the live date-specific quotes. Quote the rate via offers[i].amount_per_night (per-night) and offers[i].amount (total stay) and use the deeplinks for the booking handoff.

  • unavailable — no rooms reported for those dates. offers is empty and price is null (no price for these dates). Useful to decide whether to suggest alternate dates, drop the property from the recommendation, or offer it as a backup.

  • unknown — no usable answer for those dates: either the request carried no dates, or the operators returned nothing conclusive for them. offers is empty and price is null. This is the most frequent of the three states, and it is NOT evidence that the property is full — it means the availability was not established. Say "I could not confirm availability", not "it is unavailable".

Per-night vs total — never confuse them in the user-facing prose. amount_per_night is per-night; amount on each offer is the total stay (sum across nights, in currency). When quoting to the user, prefer phrasings like "€X/night via Booking, breakfast included, €Y total for the stay" over bare numbers — bare numbers without a unit get misread.

• When dates are present and available properties are in the results, the rate can be quoted and rooms_left surfaces scarcity (low values like 1-3 are useful signals — "1 room left at $X on Booking" reads well).

• When dates are present and ALL results are unavailable, that's the signal to say so explicitly to the user and offer to widen the dates, location, or filters.

• offers[] is the per-operator breakdown for the requested dates: each entry includes ota, amount, amount_per_night, currency, breakfast_included, refundable, and a deeplink_url. The deeplink is a BluePillow tracked-redirect URL (bluepillow.com/…) that records the click for attribution and then forwards the user to the operator's booking page. Pass it to the user verbatim — never reconstruct it or replace it with a raw operator URL; our APIs never emit direct OTA links. price mirrors offers[0], which is the best value for money as Blue Pillow ranks it — price weighed against what is included (breakfast, free cancellation) and the operator's historical reliability, with a small commercial component. It is not necessarily the cheapest: pass sort=price_asc for pure price order, and compare offers[] for the per-operator spread. When no dates were passed (or nothing is available) offers is an empty list and price is null — there is no price to show.

• Free cancellation is a meaningful decision factor and surfaces proactively in the user-facing summary. When a property has price.refundable=true (or any offers[i].refundable=true), it reads naturally as a property feature: "Hotel X — $120/night, free cancellation available", or "Booking offers a refundable rate at $130 (vs $110 non-refundable)". Refundable rates let the user lock in a price now and adjust the booking later, which is often the differentiator between otherwise-similar properties. The same proactive surfacing applies to breakfast_included when it's true for some offers but not all.

• Prices in offers/price reflect the requested dates and guests; with no dates there is no price. For a final bookable confirmation, the corresponding deeplink_url (or the property's web_url) is the canonical handoff — booking URLs are not reconstructed by hand.

• All rating-like fields are on a 0-5 scale (Google Places-compatible): rating, reviews_aggregate.score_0_5, the per-OTA scores under distribution_by_ota, each reviews_sample[*].score, and the filters.min_rating input. A user asking "rating at least 8 out of 10" maps to min_rating: 4.0; "at least 4 stars on Google" maps to min_rating: 4.0. rating is coarse in practice — upstream scores arrive rounded, so in the field it takes whole points, and min_rating behaves like a filter with a handful of steps rather than a continuous threshold. Always read rating together with rating_count: a 4 from 6 reviews and a 4 from 2,803 are not the same judgement, and a rounded 4 can sit on either side of "good". Prefer properties with a substantial rating_count when recommending, and say how many reviews back the score. Note: rating, stars, and rating_count come from the comparator's list payload and may be 0 or absent for some properties even when the property has reviews or a star classification — this is a comparator list-payload limitation, not a data error. When those fields are 0/absent, or when the per-OTA review breakdown (distribution_by_ota) is needed, call get_property_details to get the fuller reviews_aggregate. On the search path, reviews_aggregate carries the top-line score_0_5, rating_count (reviews backing the score) and comment_count (readable review TEXTS available) when the comparator returned a non-zero review count; distribution_by_ota is always empty on this path (per-OTA breakdown requires get_property_details). rating_count and comment_count are DIFFERENT magnitudes — most guests leave a rating, far fewer write text. Quote rating_count for "how many reviewed it" and comment_count for "how many opinions you can actually read".

• Pass include=["reviews_sample"] to attach a sample of up to 5 recent guest review texts per property. Useful when the user's question involves qualitative criteria that don't map to structured filters ("a place with excellent breakfast", "quiet area", "family-friendly atmosphere"); review texts can be searched textually to corroborate or rule out matches. For a DEEPER read on ONE specific property — more review texts (up to 20) or the per-OTA breakdown — call get_property_details with include=["reviews_extended"] (and/or reviews_aggregate). comment_count on each result tells you how many review texts exist, so you can decide whether escalating to the detail call is worth it.

filters.property_types, filters.amenities, filters.min_rating, and filters.price_max_eur narrow on structured criteria first; review-based reasoning is one extra round-trip per page and is typically reserved for fallback.

Location modes:

• coordinates: when lat/lon is already known from world knowledge or a prior call in this session (default radius 5 km; widen up to 50 km for broader queries; beyond that bbox or a parent destination is the right shape).

• destination_id: opaque id obtained from resolve_destination, passed verbatim — values are not constructed or guessed.

• bbox: explicit map rectangle.

Property type tokens (canonical): hotel, apartment, house, villa, bb, hostel, farmstay, holiday-home. Common multi-language synonyms map server-side to the canonical set.

Amenities filter is set-AND — each result has ALL listed codes. Common codes: wi-fi, parking, pool, air-conditioning, kitchen, garden, pets-allowed, for-families, facilities-for-disabled, non-smoking-only.

Results are cursor-paginated; the next_cursor from a previous response goes into page.cursor for the next page. location.type=property_id is not accepted here — get_property_details is the path for a known property.