Build1 distinct publisher3 min readUpdated
Receiving becomes mandatory for VAT-liable French firms on 1 September 2026. The older problem: supplier records carry a SIREN, VAT number and IBAN that nothing has checked against a live authority.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
France's schedule is fixed: from 1 September 2026, receiving an electronic invoice becomes mandatory for every VAT-liable company in the country, with issuance phased behind it, large and mid-size companies from the same date and SMEs and micro-enterprises from 1 September 2027, under article 91 of the Loi de finances pour 2024 [1]. Belgium's B2B mandate has been live since 1 January 2026, eight months ahead of the French receiving date [2][3].
The format work is the visible part and the smaller part. A post from Sirenic on dev.to makes the more useful point: the mandate does not create a problem so much as remove the excuse for an old one, because a supplier record leaving an ERP, a CRM or an LLM-parsed PDF carries a SIREN, a VAT number and an IBAN that nothing in the pipeline has checked against an authority [4].
Three failure modes sit in that record. A company that stopped trading in March keeps a well-formed, checksum-valid SIREN [5]. An intra-EU VAT number that was valid when someone exported the supplier master may not be valid today, and VIES is the authority whose answer moves [6]. IBAN mod-97 tells you the string is well formed, not that the bank code maps to an institution that exists [7]. Each has a public answer, and per the same post, collecting all three normally means three integrations, two accounts and a SOAP client [8].
Sirenic's answer is one GET on /v1/facturation/dossier at $0.03, returning legal identity and obligation dates, the computed intra-EU VAT number checked live against VIES, an IBAN check against official bank registries, and a verdict [9]. The verdict shape is worth copying whether or not you buy the service: a boolean, pret_a_facturer, plus raisons as a closed list of codes, four blocking (entreprise_cessee, tva_invalide_vies, tva_non_calculable, iban_invalide) and the rest informational (diffusion_partielle, tva_non_verifiable, iban_non_fourni, banque_non_identifiee, preparation_degradee) [10][11]. An agent branches on the code instead of asking a model what a sentence meant [10].
One design decision earns its own paragraph. A VIES outage returns tva_non_verifiable, informational, never a false tva_invalide_vies, on the stated grounds that an API which renders "the tax authority did not answer" as "this VAT number is invalid" will eventually block payment to a supplier who did nothing wrong [12].
A second endpoint, /v1/facture/verifier at $0.02, added on 16 August according to the post, cross-checks the identifiers printed on a received invoice against each other; because the French VAT key is deterministic, it catches a VAT number that is perfectly VIES-valid but belongs to a different company than the SIREN on the invoice, which a plain VIES lookup cannot see [13]. Both calls together cost $0.05 per counterparty [14]; a 10,000-supplier master runs $300 at the dossier price [15].
The commercial mechanics are unusual and carry their own cost. There is no account and no API key: the first call returns 402 with a base64 quote in a PAYMENT-REQUIRED header covering amount, asset, network and expiry, which the caller signs in USDC or EURC on Base and replays with a PAYMENT-SIGNATURE header [16]. Errors are quoted before input validation, so a 400 or 404 costs nothing, which has to be structural because on-chain micro-payments cannot be refunded [17]. Every paid 2xx carries a detached Ed25519 signature over the sha256 digest of the exact bytes returned [18]. In exchange for skipping onboarding, you put a funded wallet key inside the accounts-payable path and your audit trail becomes a signature you verify yourself.
Watch whether the raisons enum stays closed, since adding a code is a breaking change for every agent branching on it; whether receipt-side identifier cross-checking becomes standard before the SME issuance date of 1 September 2027 [1]; and how much of this ERP vendors absorb natively once receiving is compulsory, at which point a per-call fee for three registry lookups is a stopgap rather than a line item.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
On 1 September 2026, receiving an electronic invoice becomes mandatory for every VAT-liable company in France; issuance is phased behind it, with large and mid-size companies from that same date and SMEs and micro-enterprises from 1 September 2027 (art. 91, Loi de finances pour 2024).
Belgium's B2B e-invoicing mandate has been live since 1 January 2026.
A ceased company keeps a perfectly well-formed, checksum-valid SIREN; nothing in the string indicates it stopped trading.
VIES is the only authority on whether an intra-EU VAT number is valid today, and its answer moves over time, so a value that was valid when the supplier master was exported may not be valid now.
IBAN mod-97 validation shows only that the string is well formed; it does not show that the bank code maps to an institution that exists.
GET /v1/facturation/dossier costs $0.03 and returns the recipient's legal identity and obligation dates, the computed intra-EU VAT number checked live against VIES, an IBAN check against official bank registries, and a deterministic verdict.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One vendor post; statute dates checkable, product behaviour self-reported
The cluster has a single source, authored by the API vendor. Its regulatory core is specific and independently checkable (art. 91, Loi de finances pour 2024; Belgian mandate live date), and the technical description is unusually concrete — enumerated reason codes, signed-message construction, runnable client code. But every product assertion (live VIES lookups, IBAN checks against official registries, outage degradation, unbilled 4xx) rests on the vendor's own word, with no independent test, log, benchmark or corroborating publisher.
Release notes only, no usage evidence
The supplied source documents that two endpoints exist and dates one of them to 16 August, but discloses no users, customers, call volumes, integrations, partners or third-party deployments. Availability announcements alone cannot be scored as adoption, and nothing in the cluster permits inferring uptake.
Headline promise outruns what is verified
The framing — 'the one call your agent needs before paying any French supplier' — positions a single paid endpoint as the answer to a statutory compliance wave, while the evidenced part of the story is limited to the mandate dates and the interface specification. Cost claims are arithmetic on list prices, accuracy and reliability claims are unverified, and there is zero adoption evidence. The gap is moderate rather than severe because the regulatory driver is real and the technical detail is checkable in principle.
Vendor writing about its own paid API
The single source is published under the Sirenic account on dev.to and its call to action is Sirenic's own metered endpoints, priced per call and settled in stablecoins. The compliance deadline supplies urgency for the purchase, and no conflict-of-interest disclosure, competing option or independent assessment appears in the text.
Confident on dates, weak on product reality
Assessment confidence is limited by single-publisher, self-interested sourcing. The regulatory timeline and the shape of the documented interface can be stated with reasonable assurance; claims about live data quality, uptime behaviour, billing mechanics and market traction cannot be assessed at all from the supplied material.
invest
OpenAI and AWS just made agent-initiated payments a documented pattern, not a demo1 distinct publisher
build
AWS moves agent payments to GA: the plumbing is done, the sign-off is not1 distinct publisher
build
Your meter now runs on someone else's machine: signed receipts, fsync, and failing open1 distinct publisher
invest
MiCA never said whether two issuers can mint one stablecoin. Brussels now has to.1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
dev.to
1 article · August 17, 2026