Repère

← All guides

How to clear a whole supplier list against the REP register

Published

You have a spreadsheet of a few hundred suppliers and one question about each of them: what does the REP register say? Working through it is a different job from looking up one SIRET, and the hardest part comes before the first query. Here is how to key the list, what to record for each row, and why a cleared list starts ageing the day it is finished.

The list you have, and why a name is not a key

The list usually arrives as a spreadsheet: a few hundred rows exported from the vendor master, with a supplier name, an address, perhaps a VAT number, and a SIRET column that is filled in for some rows and blank for others. The question attached to it sounds simple enough, namely which of these suppliers appear on the REP register for the products we buy from them. In this guide, clearing the list means exactly one thing: working through it, row by row, against the register, and recording what the register stated for each row on the day you asked. It does not mean approving suppliers, certifying them or declaring any of them compliant, and nothing in the result should be filed as if it did. What you end up with is a set of dated records, one per supplier and stream, rather than a column of green ticks. That distinction shapes every step that follows, starting with how a row is matched to the register at all.

The tempting way to start is to paste each name into the search on ADEME's register and keep the first result. A name is not an identifier, though, and a match on a name is a guess. Company names are not unique; the legal name on the register often differs from the trading name on your invoices; and one group frequently registers several entities under nearly the same name, a holding company, a manufacturing company and a distribution company among them. ADEME's full-text search also ranks results by its own notion of relevance, so a name query can put an unrelated company, often a foreign one, at the top of the list, with nothing on screen to say that it is the wrong one. On a single lookup, done carefully, a person can catch that by eye. Across a few hundred rows, worked through at the end of a long afternoon, the wrong matches do not look like errors: they look like answers, and they go into the file as such.

So before anything is cleared, the list has to be keyed: every row needs a SIREN or a SIRET that identifies the supplier without ambiguity. That is the real first task, and it is a finite one. A SIREN stays with a legal entity for its whole life, so the work is done once per supplier and serves every later pass over the list. For a French supplier, the identifier is on its invoices and its official company documents, and the supplier can simply be asked for it; a number the supplier has confirmed is worth more than one inferred from a name. Rows that cannot be keyed should be marked as not keyed and set aside, not filled in with the best name match available, because a blank says honestly that a question is open while a guess hides it. Repère's lookups and its monitoring both require a SIREN or a SIRET for the same reason: a name match is a guess, and a guess repeated down a list is a guess multiplied.

Keying the list: SIREN, SIRET and establishments

The two identifiers do not name the same thing. A SIREN, nine digits, identifies the legal entity; a SIRET, fourteen digits, is that SIREN followed by five more and identifies one establishment of it, such as a factory, a warehouse or an office. The distinction matters here because registration on the REP register is made per establishment, not per company. One SIREN can therefore hold several immatriculations, each with its own streams: SIREN 384560942 held three on 11 October 2026. A list that records only a company name, or only a SIREN when the purchasing relationship is with one site, is not yet precise about what it is asking. The identifier you put in each row is, in effect, the question you are putting to the register on that supplier's behalf.

Ask by SIREN when you want to see every establishment the company has registered, for instance when the contract is with the company as a whole, or when you do not yet know which site supplies you. Ask by SIRET when you buy from one identified site and want that site's registrations. Neither question is more correct than the other; they are different questions, and they can have different answers for the same supplier. A SIREN query on a company with three immatriculations returns all three, each with its own streams, and whoever reads the file later needs to know whether you were asking about the company or about the site. Equally, a SIRET query that finds nothing does not show that the company has nothing registered: another establishment under the same SIREN may hold the registration for the stream concerned. Decide row by row which question fits the relationship, rather than letting whatever number happens to be in the spreadsheet decide for you.

Then record which question you asked. Keep the identifier type in a column of its own, keep the identifier exactly as queried, and do not overwrite a SIRET with its SIREN afterwards, or the record stops saying what it was a record of. Clean the numbers before you query them, too. Spreadsheets have a habit of storing identifiers as numbers, which strips leading zeros and can turn a fourteen-digit SIRET into scientific notation, so the column should be text, nine or fourteen digits long, with spaces and dots removed. Both identifiers end in a check digit, and a number that fails the check deserves a second look before it is queried rather than a quiet correction. None of this is glamorous, but a list keyed and cleaned once is the asset every later pass reuses, whether that pass is another clearing or a watch.

The answer is per stream, not per company

A producer does not register once on the REP register; it registers once for each stream it is concerned by. The register distinguishes 21 streams, each with its own code, from household packaging to electrical equipment and furniture, and on 11 October 2026, 20 of the known stream codes carried at least one row. A supplier can therefore be registered for one stream and absent from another, and both facts can be true on the same day for perfectly ordinary reasons. The answer for a supplier is a list, never a single yes or no: present in this stream under this immatriculation, absent from that one. A spreadsheet column headed REP with a yes or a no in it has the wrong shape for that answer, and whatever it says, it says less than the register did. Give each supplier as many result rows as it has streams of interest, or the list will flatten exactly the detail an auditor will ask about.

That makes the first decision yours, not the register's: work out which streams the products you buy fall under before you clear the list. Start from your own purchasing categories and map each one to a stream code, then attach to each supplier the streams of what you actually buy from it, which is not necessarily everything it sells. One supplier can fall under several streams, an office supplier delivering both furniture and desk lamps being the obvious case, while another falls under one only. With that mapping in place, each row of the list becomes a precise question: for this identifier, in these streams, what did the register state on this date? Without it, the clearing either asks about every stream for every supplier and buries the relevant answers among the rest, or asks about none in particular and records a company-level answer the register does not give. Borderline categories are best settled before the run, not after it.

An absent stream is not a finding against a supplier. The product you buy may fall outside that stream's scope once it is looked at closely, or the registration may be held by another legal entity, typically the one that places the product on the French market, such as a brand owner, a manufacturer or an importer, while your supplier only distributes it. The register also publishes nothing about the tonnages a producer has declared, the fees it has paid, or whether an eco-organisme contract covers the particular product you buy, so a registration that is present does not answer those questions either. Record an absence as what it is, an absence on a date, in one stream, for one identifier, and turn it into a question for the supplier rather than a conclusion about it. The answer may well be the name and SIREN of the entity that does hold the registration, which can then be keyed and cleared like any other row.

What a dated extract per supplier is worth

For each row and each stream, record what the register showed, not your reading of it: the identifier as queried, the stream code, the immatriculation, the IDU displayed, the dataset the row came from and the date of the snapshot. Key the result on the pair of stream code and immatriculation, never on the IDU. ADEME reissues IDUs, and a list keyed on the IDU reads a reissue as one registration ending and another starting, so the next pass over the list reports a loss and a gain where nothing happened to the supplier at all. Keep the IDU as a displayed value instead, the identifier the register showed on that date. Its shape is regular: on 11 October 2026 an IDU was always fifteen characters, FR, the six-digit immatriculation, an underscore and six more digits, FR-prefixed even for the 156,427 Chinese producers on the register that day. So the prefix cannot sort a list into domestic and foreign suppliers; it says nothing about where a producer is established.

Where a row came from matters as much as what it says, because ADEME publishes the register as two datasets. The active one held 308,665 rows on 11 October 2026, and the withdrawn one held 108,230. A withdrawn registration moves from the first to the second; it is not left in place with a changed status, so a clearing that reads only the active dataset sees a withdrawn supplier as simply missing. Only the withdrawn dataset carries date_fin_conformite, the date conformity ended, and motif_desattribution, the motive, and the motive needs care in a spreadsheet: it is not a clean set of codes, values combine, as in AUTRE || SIGNAL, and it has to be split and read as a list rather than filtered on one expected value. A supplier absent from both datasets is a third case, and it must be recorded as an absence, not as a withdrawal. Each stream of each row therefore ends in one of three outcomes, present in the active dataset, present in the withdrawn one with its date and motive, or found in neither, and the file should say which.

Assembled this way, the extract for each supplier is worth something specific in an onboarding file. It is a record of what the official source stated about that identifier, in that stream, on that date. It is citable, because it names the dataset and the snapshot it was read from; it is reproducible, because it records the question as well as the answer and the snapshot it was put to, so the reading can be repeated and checked; and it is independent of anything the supplier sends you, unlike a self-declaration or a certificate forwarded by email. It is not a verdict on the supplier. It does not say that the supplier is compliant, that it meets its obligations or that you may buy from it, and a summary column that turns it into a pass or a fail adds a conclusion the register never stated. The judgement belongs to your team, informed by the record, and the record is useful precisely because it leaves that judgement where it belongs.

Re-clearing the list is a recurring job

A cleared list is a statement about the day it was cleared. The register changes daily: ADEME adds producers, records withdrawals, corrects entries and reissues IDUs. Your own list changes too, as suppliers are onboarded, sites change and purchasing categories drift into new streams. Put the clearing date on the list as a whole, not only on each row, so that nobody reads a file from the spring as a description of the autumn. The onboarding record keeps its value as evidence of what was known when the supplier was taken on; what it loses, day by day, is any claim to describe the present. Treating the clearing as a one-off project is how a supplier file ends up holding a perfectly accurate answer that is a year old. The useful question is therefore not whether to clear the list again, but how the next change should reach you.

There are two ways to go about it. The first is to re-run the clearing on a schedule, every quarter or ahead of an audit, against the same keyed list, and to compare each run with the previous one on the pair of stream code and immatriculation: the differences are the work, whether a withdrawal with its date and motive, an absence, or a new registration under another establishment. Its limit is the interval, because a change made the week after a run stays unseen until the next one. The second is to put the identifiers you care about under continuous monitoring, so that the change comes to you instead. Detection is daily, because the register is published daily, and delivery of each alert is best-effort, with a few retries on a fixed schedule; the guide to monitoring covers both in detail. Either way the prerequisite is the same keyed list, because lookups and watches alike take a SIREN or a SIRET and nothing else.

The two are not exclusive, and the list itself suggests the split: clear all of it on a schedule, and watch continuously the suppliers whose streams weigh most in what you buy. Per-call pricing suits the first, since a list cleared in one pass costs the lookups that pass makes, and nothing while it sits in a drawer. How many suppliers you can watch continuously depends on your plan, and there is a free tier to try both on a handful of rows from your own list before deciding how much of it deserves daily attention. Whichever you choose, what comes back is the same kind of object: a dated statement of what the register recorded, stream by stream, for an identifier you chose. What to make of it for a given supplier is a decision that stays with your team, and Repère does not make it for you.

Figures quoted in this guide were read from ADEME's published datasets on the date above. The register changes every day, so treat them as a dated observation rather than a current count.

Other guides

Put a date on the answer

Query the register by SIRET or SIREN and keep what it said, with the day it said it. No credit card to start.