Skip to main content
Vestiaire Collective is the second marketplace rega watches, and almost nothing in it works the way Vinted works. Listings are born here at about 24 a minute, against roughly 150 a second on Vinted — and not one of them is born at a moment anybody can predict, because a human being at the marketplace decides when it appears. That single fact rewrites the whole method, and this page is what it rewrites.
A wide ring of green light lying flat on an almost black background, like a wave sweeping across

A sweep passing over a window of identifiers, again and again, with nothing between two passes. That is the shape of the Vestiaire watch.

​
Two marketplaces, two different jobs

Every figure below is measured. The language line was measured on both marketplaces on the same day, 16 August, each of them from a single exit address held still for the whole reading — otherwise a comparison of that kind means nothing.
Two facts explain everything else on this page. The first: nothing exists publicly on Vestiaire before a human has validated it, so there is no starting gun to listen for. The second: because the marketplace applies none of your criteria, the same sweep can serve every client at once — which is why your filters cost the watch nothing, however many of them you write.

​
Every listing is validated by hand, and that is the whole story

A seller submits an item. It goes into verification and photo retouching at Vestiaire, and then a member of staff validates it or refuses it. Generally a few minutes to a few hours — and sometimes several days. That range is the marketplace’s own, not a delay we time, and nothing on our side shortens it by a second.
A line of matte tiles aligned in shadow, a single one lit in green

A queue of items waiting on a human decision, and only one of them lit: the moment a member of staff validates a listing is the moment it starts to exist.

Three consequences follow, and all three are load-bearing.
1

Nothing exists publicly before the validation

The listing is at no address at all. Any search for a “fresher door” placed before the validation is looking for something that does not exist. There is no early view of Vestiaire to be bought, borrowed or negotiated.
2

The public birth is a gesture of the staff, not of the seller

Which means it is predictable by nobody: not the seller, not an insider, not a competitor. Everyone finds out at the same moment — and from that moment on, it is a race about seconds.
3

A listing's number tells you nothing about its freshness

The number is assigned at submission; the validation lands later. On one real case we measured 66.7 minutes between the two. A listing validated one second ago therefore carries an old number, scattered somewhere in the moderation backlog.
There is no leading edge, and that is measured. A radar that watched only the 300 numbers just above the frontier — the intuitive design, the one that would work on Vinted — made 18 discoveries out of 1,412. Freshness cannot be read from the height of a number. It has to be swept for.

​
What we do instead: a window, swept without pause

Two doors exist onto the marketplace, and they do not have the same properties. The radar uses both, for two different reasons.
One request per listing. It sees 77 out of 77 of the listings we checked, including the very freshest. It is the only door that gives freshness, and it is the one the radar spends nearly all of its budget on.Its ceiling is comfortable: the marketplace accepts at least 16 requests a second on it, and we run at less than half of that.
We never poll the site’s “New in” page. It is refreshed in batches of about 10 minutes, which means it lags 5 to 10 minutes — and it is exactly what every other buyer is looking at. Using it would be giving away the entire advantage. It has one job here: at start-up, it tells the radar where to look. After that, the window follows the discoveries.

​
The size of the window, and why it stops there

The radar sweeps from 2,000 numbers below the frontier to 240 above it — about 2,240 numbers, over and over. That figure is not a guess; it is where a measured curve flattens out. The 0.7 % that stays out of reach is not a defect to be fixed later: those are old items put back on sale, sitting some 700,000 numbers below the frontier. No window, however wide, catches them by enumeration.
The pass is the pace. There is no wait between two sweeps — the next one starts as the previous one ends. So the only way to see a listing sooner is to make a pass shorter, and a pass gets shorter in exactly one way: more exit addresses, working in parallel. The gain is linear in the number of addresses, and it is the only lever there is.
A listing already filed is never filed twice. Once the sweep has read a number, it steps over it on every later pass without asking about it again — so a listing that stays inside the window costs the radar nothing, and you will never watch the same line climb back to the top of your detections every few seconds. A number that answered nothing, on the other hand, is tried again at every pass: it may be a listing still waiting on its human decision.

​
The pace we measured

Each of these rows is a real run, with its own date and its own sample size. None of them is derived from a setpoint. At four addresses, the full distribution over 101 detections: 34.7 % seen in 15 seconds or less, 51.5 % in 30 seconds or less, 97.0 % in a minute or less, and 100 % in two minutes or less.
The model has one term in it, and only one: the usual delay is about half the length of a pass. It was checked at both regimes — a pass of 31 s predicted 15.5 s and measured 18 s; a pass of 11 s predicted 5.5 s and measured 6 s. There is nothing else in it, because the marketplace adds nothing: the shortest delay we ever recorded is about zero seconds, over 54 detections. Vestiaire serves the listing at the very instant it validates it. There is no propagation floor to buy back.
Two defects found by measuring, both fixed, both worth knowing because they say what the pace depends on:
  • Sweeping the addresses one after another instead of together turned a pass into 110 s and a delay into 100 s, on the very same four addresses that give 36–40 s and 26–42 s in parallel.
  • Filing detections at the end of a pass instead of as they are found cost exactly half a pass: 42 s usual and 60 s at the ninth decile, against 18 s and 25 s on the same four addresses.
Do not judge the speed in the first minute. On a cold start, the radar has the whole already-validated backlog to work through, and we measured a usual delay of 169 seconds during the first minute against 3 seconds once it settles. The screen says so itself, in italics, for the first five rounds: “starting up: the first listings are the backlog already online”.
What is honestly not measured. Four addresses have held about 15 cumulative minutes without a single refusal from the marketplace, on the bench and then in the product. A round-the-clock regime is an extrapolation, not a measurement — and the refusal counter on the screen is the thread that would break first. Watch it: at zero, all is well.

​
One sweep, for every client at once

This is the founding rule of rega, carried over to a second marketplace: the watch is enumerated once, and shared. Sweeping once per client would make the cost of the watch grow with the number of clients, which is the one thing this product cannot afford. It is possible here for a precise reason: the sweep carries none of your criteria. The request the radar sends carries a list of numbers and nothing else. Brands, categories, colours, materials, sizes, prices, countries, keywords — all of it is decided here, on our side, against what came back.

Your filters cost the watch nothing

Ten filters or a hundred, the radar makes exactly the same requests. What changes is only how much of what it brings back is kept for you.

The radar holds no account

Every request it makes is anonymous. Your Vestiaire accounts are for offering, favouriting and buying — never for watching. Nothing about the watch can put them at risk.

Having an active filter is the arming

You do not have to click anything: creating an active Vestiaire filter arms the watch for your workspace. To stop detecting, deactivate the filter — disarming by hand does not survive a restart while an active filter remains.

The first pass never alerts you

It is an inventory. Without that rule, arming would pour the couple of thousand listings already online into your screen as if they had just been born.
Where the marketplace does apply your criteria: the preview, and nowhere else. The “On Vestiaire right now” counter in the filter form is a real question asked of Vestiaire, and asking it does send your brands, categories, universes, conditions, colours, materials, sizes, countries and your two price bounds to the marketplace. Your words are the one thing that never goes: what to find and what to rule out is read here, on our side, and nowhere else. Detections are a different path entirely — which is worth knowing, because a fault in one shows up on the counter and not on your detections, or the reverse, and that makes the two easy to believe in agreement when they are not.
There is a free gift in that arrangement. When the form asks Vestiaire how many listings match, the marketplace answers with the count for every value at once, in the same reply, at no extra cost. The equivalent on Vinted costs about fourteen catalogue requests on a shared address. It is why the Vestiaire filter form can show you what a criterion would gain you before you tick it — see Writing a Vestiaire filter.

​
Language is a setting here, an address at Vinted

Measured on 16 August, from production, on the same items, through the same exit address, asking first in French and then in English: The same day, from one single address, www.vinted.fr answered “petit” and www.vinted.co.uk answered “small” — and asking the French site for English was simply ignored. On Vinted, the address decides. On Vestiaire, a setting does.
Why the radar still sweeps in French, once. Because the sweep is shared by every client. Sweeping it in each client’s language would multiply its cost by the number of languages, and that is the one thing the shared sweep exists to prevent. So the radar reads once, in French — and the screen translates afterwards, which it can do because the identifiers do not move. A condition, a colour, a material, a category: the word changes, the identifier behind it does not.
What is left after that is the title and the description, which are free text and cannot be translated from a table. Those are fetched on demand, in batches of up to 60 listings per request — so a whole screen of cards costs one anonymous request — kept for an hour, and capped at 4,000 entries. If your interface is in French, this costs nothing at all: there is nothing to fetch. The lists in the filter form are not translated either: they are asked for in your language. Measured on 14 August, the marketplace returned a different label under an identical identifier for 14 of 14 top-level categories, 5 of 5 conditions, 3 of 3 universes, 20 of 24 colours, 49 of 56 materials and 86 of 99 second-level categories. Reading the same reference lists on Vinted needs a French exit address and a throwaway account; here it is one anonymous request.
The one thing this does not repair. The condition label stored inside a detection card is the one from the radar’s pass, so it is in French for everybody. What travels beside it is the identifier, and that is what your screen translates. If you ever see a French condition where you expected English, that is where it comes from, and nothing about the detection itself is wrong.

​
From validation to purchase, end to end

Everything between the validation and the detection line belongs to us and is measured in seconds. Everything after it belongs to you: no gesture on Vestiaire commits your money without a click, and the last one of all belongs to your bank.

​
What changes once you decide to buy

The rest of the marketplace is as different as the watch is. Each of the four below has a page of its own; here is the shortest honest version of each, so that nothing surprises you on the way.

​
The price on screen is not the price you pay

And the gap is neither constant nor proportional — it is not a percentage you can learn. Four real baskets, measured: Taking the third one apart, since it is the one we measured line by line:
The buyer-service fee is inside the displayed price, not on top of it. On that basket it was €8.16 — which means an item shown at €34 is paid €34 for the item, of which €8.16 goes to the marketplace and €25.84 to the seller. What is genuinely added above the shown price is delivery and duties. This is why every purchase in rega reads the basket total and never the price of the listing, and why a price criterion in a filter does not bound what you will pay. Delivery alone can more than double: we measured €4.90 at a pick-up point against €10.90 to a home address on the same order.

​
The basket belongs to the account, not to the item

On Vinted, two purchases can run at once on one account. Here, the basket is global to the account: preparing a purchase empties whatever was in it, and one account can only carry one purchase at a time. It is also why the pick-up point map needs a basket with something in it before it can be drawn at all.

​
Your bank is always the last word

We took the payment chain to the end and measured the bank’s step: four times out of four, the card we used was sent to a full challenge — 18 seconds and three form submissions on the capture. That is a policy of the card issuer, not a limit of the marketplace, and another card may well be exempted. What it means for you is simple and it does not change: rega prepares, you confirm with your bank, and it is your answer that moves the money. Nothing is charged before it. Automatic buying exists here, and it stops exactly there. On a filter you have armed for it — one tick, filter by filter, never in bulk — rega reserves an account, fills the basket, sets the delivery and opens the payment, then tells you and hands back. You have fifteen minutes to answer. If you do nothing at all, the purchase is simply dropped, the account goes back into service, and nothing has been charged. Not one line of that path can charge a card.

​
A sale is announced, not deduced

On Vinted a sold item vanishes, and we only conclude after two consecutive absences. Here, the page stays up and says “sold”, with the hour of the sale, so one reading is enough. A listing that was withdrawn answers something else again, and it is never counted as a sale — the two are told apart, which on Vinted they cannot be. And because both timestamps come from the marketplace itself, the time an item took to sell is genuinely measured, not bracketed by our own check interval. Two real ones, to show the range: 398 seconds for a Dolce & Gabbana coat, and 3,886,847 seconds — 44 days, 23 hours and 40 minutes — for a Dior dress.
The two marketplaces number in the same space. A Vestiaire item carries a number around 70 million, and Vinted handed out numbers in that range around 2016. So a listing number, on its own, does not say which marketplace it belongs to — everything in rega that follows an item carries its marketplace alongside its number. You will see it as a small marketplace tag on sales, on the watchlist and in the feed. It is not decoration.

​
What surprises people

The radar runs on our servers. Closing the tab, closing the browser, shutting the machine down: none of it changes anything, and none of it is needed for a detection to be filed.What starts it is not a button but a fact: an active Vestiaire filter is the arming request. The consequence is worth reading twice — disarming by hand is immediate, but it does not survive a restart while an active filter remains. To genuinely stop detecting, deactivate the filter.
Each address carries a single thread, and pulls at about 6 to 7 requests a second — the same at four addresses as at twelve. So addresses do not share a budget: they add up. That is why the pace scales linearly, and why the refusal counter stayed at zero across every regime we measured, including a run of some 22,000 requests over five minutes of continuous sweeping.The one ceiling that is genuinely narrow belongs to the other door: the batch search accepts about one request a second per address, against a daily budget. It is why that door is used sparingly, and only for completeness.
The shortest delay we ever recorded is about zero seconds — and, once, a second below zero, which is clock drift and nothing more. Vestiaire serves the listing at the very instant it validates it: there is no propagation delay hiding between the two. When a pass happens to land on that number right then, the delay really is nil.
The marketplace itself slows down. We measured 24.2 births a minute during the day — with a spread from 13.5 to 46.0 — and 7.2 a minute between 22:30 and 00:30 UTC. That is roughly a third of the daytime rate. A filter that produced ten lines an hour in the afternoon and one overnight has not changed; the marketplace has.
Past that, Vestiaire stops counting. The filter form therefore writes {n}+ rather than a bare “10,000”, because printing the round number would pass a limit off as a measurement. If you see the +, your filter is wide — much wider than the figure suggests.
Kept listings appear on the Vestiaire Collective screen and join the main feed, beside your Vinted finds, carrying a tag that says which marketplace they came from. You do not have to choose a screen to hunt from.
When the watch starts, the two thousand or so listings already online are all, technically, new to it. Filing them as detections would be true and useless. So the first pass is an inventory: it fills the window, and it alerts nobody. The screen marks the first five rounds in italics — “starting up: the first listings are the backlog already online” — for exactly the same reason.

​
Where to go next

Writing a Vestiaire filter

Every criterion, what it does, what it rules out, and what happens when two of them contradict each other.

Your Vestiaire accounts

Connecting one, what we keep of it, and the connection that is valid for one thing and refused for another.

Making an offer

The floor, where it comes from, and why it rises to your own amount and never comes back down.

Buying on Vestiaire

The two steps, the delivery choice, the full cost of an item, and what happens if you walk away halfway.

Following what sells

What a Vestiaire sale tells you that a Vinted sale cannot, and the watchlist across two marketplaces.

The Vestiaire screen

The visit: the arming button, the six tiles, the three tabs, and where every gesture lives.
If you are new to rega altogether, the first ten minutes is the page to read before this one. And Money is the page that says, for the whole product, who authorises a spend and who bounds it.

​
What this marketplace does not let us do yet

We would rather write it here than let you find out halfway through.

Some listings are out of reach

The 0.7 % the window cannot cover are old items put back on sale, sitting hundreds of thousands of numbers below the frontier. Widening the window does not help — it was measured, and past 2,000 there is no gain.

Round-the-clock is not proven

Four addresses have held about 15 cumulative minutes with zero refusals. A day and a night without a refusal is an expectation, not a measurement. The refusal counter on the screen is the one to watch.

Detections are wiped after 48 hours

The clean-up runs on its own and nothing warns you first. Anything you want to keep — a favourite, a tab, a place on the watchlist — take it out the same day.

Item models cannot be chosen

The marketplace publishes no list of models, and the field is missing from 79 % of listings. A filter that already carries models keeps them and lets you remove them; you cannot add one.

No purchase history

rega’s Orders screen follows Vinted only: a purchase made here never appears on it. Your Vestiaire orders live on Vestiaire, and that is where you follow them.

“Requested” is the last word we ever get

Once your bank has answered and the confirmation has gone, the marketplace says nothing more, and no page anywhere re-reads the state of a payment. rega will not claim a purchase went through when it cannot know.

No automatic offer

An offer is always a click, and always yours. This is a decision, not a gap: an offer here is irreversible, and nothing should send one on your behalf.

A favourite cannot be removed from rega

You can set one from here; you take it off on Vestiaire. We would rather not offer a button that would fail silently, on a marketplace where a wrong gesture costs an account.

Pick-up points need a full basket

Measured: zero points on an empty basket, twelve on a full one. There is no way to draw that map without an item, so rega borrows one for the length of the reading and empties the basket afterwards.

Addresses are only located in France

Outside France, the map says it does not know where the address is rather than dropping a pin in the wrong place.

Filter groups and import stay on Vinted

The Vestiaire filter form has no group selector, and the filter import and export still speak the Vinted format only.

Duties are in the total, not on their own line

The marketplace does send a duties figure — €23.12 on the basket above — but rega does not pull it out yet. It is inside the total you see, and only there.
One last thing that is a limit of the marketplace rather than of rega: the moderation delay itself. A listing can wait minutes or days on a human decision, and no watch, however fast, changes that. What we can promise is the part that is ours — the seconds between the validation and your screen — and that part is measured on this page, run by run.