Drouot Scraper — auction lots, sales and auction-house contacts

drouot.com is the marketplace for the Hôtel Drouot in Paris and the partner auction houses that catalogue through it. This actor exports it three ways, and every single row carries the auction house's email, telephone and street address — so a lot list is also a lead list.

Measured on the live site on 2026-09-18, through the Apify proxy, by walking every list to its last page:

	measured
Upcoming sales on the board	809 distinct (the source's own counter said 836 at the same moment — see Honest limits)
Auction houses running them	364 distinct
Lots in those catalogues	214,576 (sum of the source's own per-sale lotCount)
Lot rows actually harvested in this measurement run	62,440 (page 1 of every sale)
Lot URLs in Drouot's own sitemap	201,820
Countries represented	27, from France (491 sales) to Uruguay (1)
Cost of the whole contact-bearing sale calendar	17 requests

---

What one row is

Pick Row type in the input.

lots — one row per lot (the default)

Lot number, title, description, estimates, live bid, buyer's premium, photos — plus the whole sale and the whole auction house on the same row. This is a real row from the verified run (fixtures/lot_detail.json, with Add the auction house registry block on):

{
  "recordType": "lot",
  "lotId": 34888742,
  "lotUrl": "https://drouot.com/en/l/34888742-a-set-of-five-porcelain-coffee-cups-and-six-saucers-with",
  "lotNumber": 780,
  "title": "A set of five porcelain coffee cups and six saucers with white bases and a blue monochrome design.",
  "description": "A set of five porcelain coffee cups and six saucers with white bases and a blue monochrome design.",
  "descriptionIsTruncated": false,
  "originalDescription": "Ensemble de cinq tasses à café et six sous tasses en porcelaine à fond blanc et décor en camaïeu bleu.",
  "categoryIds": [632],
  "categories": ["Sets of tableware"],
  "categoryPath": "Sets of tableware",
  "attributes": null,
  "lowEstimate": 10,
  "highEstimate": 20,
  "estimateRange": "10-20",
  "currentBid": null,
  "nextBid": 10,
  "startingPrice": 10,
  "hammerPrice": null,
  "tenderingAmount": null,
  "reserveNotReached": false,
  "displayStartingPrice": true,
  "biddingStartsAt": "2026-09-17T09:00:00.000Z",
  "biddingEndsAt": "2026-09-19T09:00:00.000Z",
  "forProsOnly": false,
  "transportSize": null,
  "censorship": null,
  "imageUrl": "https://cdn.drouot.com/d/image/lot?size=fullHD&path=840/183595/639065a658d4a5249eac834799b23490",
  "images": [
    "https://cdn.drouot.com/d/image/lot?size=fullHD&path=840/183595/639065a658d4a5249eac834799b23490",
    "https://cdn.drouot.com/d/image/lot?size=fullHD&path=840/183595/4e7571db5c0f22f15dae9672a41c13f5",
    "https://cdn.drouot.com/d/image/lot?size=fullHD&path=840/183595/926904b9ff7a85f0a7bd8aad6f8f7635"
  ],
  "imageCount": 3,
  "saleId": 183595,
  "saleUrl": "https://drouot.com/en/v/183595-tableware-jewelry-designer-items",
  "saleTitle": "Tableware, jewelry, designer items",
  "saleDate": "2026-09-19T09:00:00.000Z",
  "saleStatus": "CREATED",
  "saleType": "LIVE",
  "saleTimeZone": "Europe/Paris",
  "currency": "EUR",
  "buyerPremiumPercent": 24,
  "platformFeesIncluded": false,
  "registrationDeadline": "2026-09-20T09:00:00.000Z",
  "houseId": 840,
  "houseName": "Étampes Enchères",
  "houseUrl": "https://drouot.com/en/auctioneer/840/etampes-encheres",
  "houseEmail": "svvetampesencheres@yahoo.fr",
  "housePhone": "+33164940233",
  "houseLifetimeLotCount": 27204,
  "houseMemberSince": "2022-05-21T12:30:00.000Z",
  "houseLogoUrl": "https://cdn.drouot.com/d/CP/logo?size=fullHD&cpId=/840",
  "street": "Route de la Ferté Alais, carrefour D191 et D837",
  "postalCode": "91150",
  "city": "Morigny-Champigny",
  "countryId": 75,
  "country": "France",
  "atHotelDrouot": false,
  "houseRegisterNumber": "44281540300021",
  "houseAgreementNumber": "2002-292",
  "houseVatNumber": "FR84442815403",
  "houseWebsite": "https://www.etampesencheres.fr",
  "houseDrouotCertified": false,
  "scrapedAt": "2026-09-18T19:41:12.004Z"
}

sales — one row per auction (809 upcoming)

The same contact list at 1/77th the row count: sale title, date, status, type, lot count, buyer's premium, registration deadline, catalogue PDF, terms of sale, and the house's full contact block. The whole calendar costs 17 requests.

auctionHouses — one row per house (364)

The prospect list. Everything above plus the company registration number, the French agrément number (the authorisation an opérateur de ventes volontaires must hold under Code de commerce L321-4), the VAT number, the house's own website, its named auctioneers, and its published hammer prices.

---

Measured field fill

Every percentage below is a count over rows this actor actually produced on 2026-09-18, not an estimate. A field is "filled" only if it is a non-empty value: 0, "", "N/A" and "-" are not.

Sale rows — n = 809 (the whole live board)

field	fill	field	fill
saleId saleUrl saleTitle saleDate	100%	houseEmail	100% (809/809)
saleStatus saleType currency	100%	housePhone	98.9% (800/809)
buyerPremiumPercent	100%	street	99.8% (807/809)
registrationDeadline	100%	postalCode city	98.0%
saleTimeZone platformFeesIncluded	100%	countryId	100%
houseId houseName houseUrl	100%	country (resolved)	100%
coverImageUrl	99.4%	atHotelDrouot	100%
saleDescription	85.3%	lotCount	80.6%
houseShippingOffer	22.6%	venueName	24.4%
catalogPdfUrl	8.8%	externalCatalogUrl	5.7%
virtualTourUrl	0.4%	saleAnnouncement	0% on this route (10.7% on the sale page)

A second, independent census of 215 sales read from their own sale pages (not the board) gave houseEmail 100%, housePhone 99.5%, street 99.5%, saleTerms 100%, lotCount 100%, saleDescription 90.7%, saleAnnouncement 10.7%, catalogPdfUrl 14.0%.

Lot rows — n = 62,440 (page 1 of all 809 sales)

field	fill	field	fill
lotId lotUrl lotNumber title	100%	houseEmail	100% (62,440/62,440)
description	100%	housePhone	98.6%
biddingEndsAt currency	100%	street	99.7%
buyerPremiumPercent	100%	postalCode city	97.6%
saleId saleTitle saleDate saleType	100%	country	100%
imageUrl images	99.8%	registrationDeadline	100%
lowEstimate	93.8%	highEstimate	90.0%
nextBid	58.2%	attributes	24.0%
currentBid	3.4% overall — 22.0% of lots in ONLINE sales, 0% everywhere else	descriptionIsTruncated = true	5.8%
biddingStartsAt	5.4%	transportSize	4.4%
hammerPrice	0% on the forward board — see Honest limits	tenderingAmount	0.4%

Sliced by sale type, because bidding data only exists where bidding is open:

sale type / status	lots	currentBid	lowEstimate	highEstimate
LIVE / CREATED	50,031	0.0%	94.6%	90.5%
ONLINE / IN_PROGRESS	9,633	22.0%	88.9%	86.8%
CATALOGUE / CREATED	2,476	0.0%	100.0%	96.0%
LIVE / IN_PROGRESS	300	0.0%	66.0%	66.0%

Lot rows read from each lot's own page — n = 299 (fetchLotDetails)

originalDescription 100% · categoryIds 90.0% · categories 89.3% · startingPrice 93.0% · houseMemberSince 100% · houseLifetimeLotCount 97.3% · lowEstimate 93.0% · highEstimate 88.3% · images 99.3% · houseEmail 100% · housePhone 99.0% · descriptionIsTruncated 0% (this route does not truncate) · attributes 0% (this route does not send them — which is why the actor keeps the value from the listing rather than overwriting it).

Auction-house rows — n = 363 (every house on the board)

field	fill	field	fill
houseName houseUrl houseEmail	100%	houseRegisterNumber	99.2% (360/363)
housePhone	98.6%	houseVatNumber	92.0%
street postalCode city	100%	houseWebsite	89.0%
country	99.7% (see Honest limits)	houseAgreementNumber	52.1%
houseMemberSince	100%	houseLifetimeLotCount	97.0%
upcomingSales	100%	houseMemberYears	94.2%
houseProfile	53.7%	topResults (hammer prices)	31.1%
houseShippingOffer	25.6%	team (named auctioneers)	14.6%

---

What you can point it at

Leave everything empty and it walks the whole live board. Otherwise, any combination of:

• Drouot URLs pasted from the browser, in any of the six site languages. A URL the actor cannot walk is named in the log, never silently dropped. What each one means depends on the Row type:

  pasted URL	in lots	in sales	in auctionHouses
  /en/v/184215-… a sale	its whole catalogue	that one sale	the house running it
  /en/l/34888742-… a lot	that one lot	—	—
  /en/c/626/paintings a category	its lots	—	—
  /en/s?query=bernard+buffet	matching lots	—	—
  /en/auctioneer/840/… a house	every lot in its upcoming sales	its upcoming sales (1 request)	that house
  /en/auctions/future the board	every lot on the board	all 809 sales	all 364 houses

• Search term — Drouot's own full-text lot search.

• Categories — all 116 ids from Drouot's own tree (6 macro, 27 categories, 83 sub-categories), with the hierarchy shown in the picker.

• Sale IDs — the numbers out of /v/ URLs.

• Language — en fr de es it zh.

Filters (a filtered row is never delivered and never charged)

Auction-house country (the 27 measured values, each with its share of the board), sale type (LIVE / ONLINE / CATALOGUE / COURANTE), sale status, estimate band, and only rows with an email or a phone.

Depth (costs requests, never extra charges)

• Open each lot page — the untruncated description, the auctioneer's original-language text, every photo, the category ids. One extra request per lot.
• Add the auction house registry block — registration, agrément and VAT numbers plus the house's own website. One request per distinct house, cached: the 809-sale board costs 364 lookups, not 809.
• Add the full terms of sale — one request per sale, 100% fill.

What a run actually costs in requests

drouot.com tolerates about 4 requests a second and no more (see Honest limits), so run time is requests ÷ 4. These are measured, not estimated:

what you asked for	rows	requests	wall clock
25 lots off the board (the demo scope)	25	18	15 s
1,000 lots off the board (the default)	1,000	28	21 s
20 lots from a category URL	20	17	~10 s
30 sales + terms + registry	30	75	~25 s
8 auction houses	8	25	~12 s
10 lots from a search, with lot pages and registry	10	180	~60 s

The last row is the worst case and worth understanding: category and search listings do not carry the auctioneer contact, so each distinct sale behind them has to be opened once (and each lot, and each house, when you turn those on). Ten search hits spread across ten different sales run by ten different houses is the least cache-friendly shape there is. Walking a sale or the board is far cheaper per row, because one request brings 100 lots that already share one contact.

---

Pricing

$0.0018 per row delivered. No run-start fee. That is $1.80 per 1,000 rows, and it is the only charge: a filtered-out row, a sale that does not exist, a sale with no catalogue and an unreachable URL are all free, and opening lot pages or joining the registry block does not add an event.

Measured for you, so the numbers are comparable: the 1,000-row default scope took 28 requests, 3.76 MB and 20.8 seconds; the 25-row demo scope took 18 requests, 2.70 MB and 15.0 seconds.

---

Compared with the other Drouot scrapers on the store

There are exactly two, scanned 2026-09-18 across drouot, encheres, auction house, art auction and auction lots: saswave/drouot-scraper ($0.002/result FREE tier plus an actor-start fee, 2 users in 30 days, 15 of its 60 public runs FAILED) and dadhalfdev/drouot-scraper ($0.005/result FREE tier, 0 users in 30 days).

Everything they deliver, this delivers, and it is named here field by field so you can check:

their field	here		their field	here
url / sku / id	lotUrl / lotId		store_name / auctioneer_name	houseName
obj_name / title	title		store_address / address	street
description	description		store_city / city	city
category	categories + categoryPath		store_zipcode	postalCode
estimate_low / _high / _range	lowEstimate / highEstimate / estimateRange		phone / auctioneer_phone	housePhone
last_bid / current_bid	currentBid		email / auctioneer_email	houseEmail
next_bid	nextBid		auctioneer_id	houseId
result	hammerPrice		image / image_url	imageUrl
auction_status / sale_status	saleStatus		images	images
auction_type / sale_type	saleType		lot_number	lotNumber
bidding_end	biddingEndsAt		sale_id / sale_url / sale_title	saleId / saleUrl / saleTitle
bidding_start	biddingStartsAt		sale_date / timezone	saleDate / saleTimeZone
frais_vente / sale_fees_percent	buyerPremiumPercent		currency	currency
reserve_price_met	reserveNotReached (the inverse — Drouot's own field is reserveNotReached, so that is what is reported rather than a flipped copy)		lot_count	lotCount
record_type	recordType

One rival field is cut: saswave's availability, which is an empty string in the example row of their own README and which has no counterpart anywhere in Drouot's payloads. A column of empty strings is not a field.

What is here that is not there:

1. The auction house as a first-class row — 364 of them, with the registration number (99.2%), VAT number (92.0%), agrément number (52.1%), own website (89.0%), named auctioneers and published hammer prices. Neither rival has an auction-house mode at all.
2. country resolved from a bare integer. Drouot stores country as 75 / 228 with no dictionary anywhere in its data — even its own schema.org output emits "addressCountry": 228. All 27 values on the live board are resolved here; anything unresolved stays null beside its countryId rather than being guessed.
3. descriptionIsTruncated. The listing routes cut descriptions at exactly 500 characters with no ellipsis and no flag (5.8% of lots). This actor flags every one and offers to re-read them.
4. saleTerms — the full conditions of sale, 100% fill.
5. registrationDeadline — 100% fill; the date after which a bidder cannot register.
6. attributes resolved to labels (colour, dimensions, brand, movement, material) on 24.0% of lots, and categories resolved to names from the 116-id tree.
7. catalogPdfUrl and externalCatalogUrl (the house's own Interenchères listing).
8. Filters that reduce your bill — country, sale type, status and estimate band are applied before delivery, so you are not charged for rows you asked not to have.
9. A RUN_SUMMARY that reconciles what Drouot said was there against what was read, counts this run's own per-field fill, and keeps the three kinds of nothing apart.
10. No run-start fee (saswave charges one), and the lowest per-row price of the three.

---

Honest limits

1. Hammer prices barely exist on drouot.com, and this is the one place a rival's field is mostly empty. hammerPrice was 0 on all 62,440 lot rows from the forward board. Sampling 500 lot pages spread evenly across all 201,820 URLs in Drouot's own sitemap: 499 answered, 36 belonged to CLOSED sales, and only 4 carried a hammer price — 0.8% of all lots, 11.1% of closed lots. Drouot's full sold-price archive is a separate paid product (gazette-drouot.com). The field is shipped because it is real when it is there, and reported as null — never 0 — when it is not. The dense source of realised prices here is topResults on auction-house rows: 31.1% of the 363 houses publish their top lots with the price they fetched.
2. 19% of upcoming sales have no catalogue to scrape. Of the 809 sales on the board, 154 sale pages return "sale not found" — the auction is announced but its lot list is not published yet. In sales mode you still get all 809 rows with their contacts; in lots mode those 154 produce nothing, and you are not charged for them. RUN_SUMMARY lists them.
3. A closed sale withdraws its catalogue. CLOSED sales return an empty lot list. Individual lot pages survive, so the way to keep a closed catalogue is to capture it before the hammer.
4. Listing descriptions are cut at 500 characters (5.8% of lots). Turn on Open each lot page for the full text — at one extra request per lot.
5. One country id could not be resolved. Id 44 (one house in Taipei) has no English rendering anywhere the actor could read, so its rows carry country: null with countryId: 44. 362 of 363 house rows and 100% of lot and sale rows resolved.
6. drouot.com rate-limits at about 4 requests a second. Measured over its own 619-URL sale sitemap: 4.00 req/s gave 615 × HTTP 200 and zero 429s; unthrottled at ~52 req/s gave 493 × HTTP 429. The actor holds itself to 4 req/s and halves that for 30 seconds if a 429 slips through. The practical consequence: the entire ~214,000-lot corpus is about 2,200 requests, i.e. roughly 9 minutes, and there is no way to make it faster without being blocked.
7. The board moves while you read it. Walking the 17 pages took the source's own totalItems from 835 to 836 and returned 809 distinct sales. Sales are deduplicated by id and the shortfall is reported, never silently accepted as "that is all there is".
8. currentBid only exists where bidding is open — 22.0% of lots in ONLINE sales, 0% in LIVE or CATALOGUE sales, which have not opened yet. That is the source's behaviour, not a gap.
9. houseLifetimeLotCount and houseMemberSince are absent from the sale board. Drouot sends lotCount: 0 there for every house, including one with 27,204 lifetime lots. They are reported as null on those rows and filled in properly by Add the auction house registry block or by Open each lot page.
10. attributes are on listing rows and not on lot pages. Opening lot pages therefore keeps the listing's attributes rather than overwriting them with nothing.
11. Search results depend on the language. "buffet" returned 308 lots on /en and 452 on /fr — the index is per-language. Set Language to match how the catalogue is written.
12. saleAnnouncement (the press note) is on the sale's own page, not the board: 10.7% of sales publish one.

---

Source, access and legality

The data is the same JSON payload drouot.com's own pages are rendered from (…/__data.json, SvelteKit). No login, no API key, no browser automation, and nothing behind a paywall — the actor reads the public catalogue exactly as a visitor's browser does, at a rate the site tolerates without complaint.

https://drouot.com/robots.txt, read 2026-09-18, contains both of these, verbatim:

Disallow: /l/*
Disallow: /v/*

and, in the same file:

Sitemap: https://drouot.com/sitemap-en-lot.xml
Sitemap: https://drouot.com/sitemap-en-sale.xml

— sitemaps that contain nothing but /l/ and /v/ URLs, published for search engines to crawl. A second stanza of that file gives ClaudeBot, GPTBot, CCBot, anthropic-ai, Scrapy, Bytespider and 27 other named agents a blanket Disallow: /; this actor is none of them and never identifies itself as one.

Drouot publishes each house's contact because French law makes it mandatory: Code de commerce L321-4 et seq. requires every opérateur de ventes volontaires to be declared to the Conseil des maisons de vente and identified in all sale publicity. That is why houseEmail is at 100% and why the houseAgreementNumber field exists at all.

---

Reproducing the numbers in this README

Everything above comes from bytes in fixtures/, captured through the Apify proxy on 2026-09-18 and listed with URL, status and size in fixtures/CAPTURE_MANIFEST.json. Run

node offline_validate.mjs

with no network and no node_modules to re-check the parser, the pagination guard, the null handling, the dictionaries and the packaging against those bytes. SIGNOFF.md carries the full measurement log.