Falmoran builds TradeOS — a deterministic engine that works out duty, import VAT and tariff classification, shows every rule it applied, and issues a signed certificate your bank can verify without asking you for anything. It is built to refuse rather than guess, because a confident wrong number is the most expensive thing a compliance tool can produce — and it is the only engine we know of that tells you which parts of the tariff it does not hold, on every figure it gives you.
No login, no signup: twenty products a day, or five pasted lists. The claim above is the one thing on this page you can check yourself in thirty seconds — so check it before you read anything else we wrote.
Two real determinations · 19 Aug 2026
Not a diagram of one. Both commodities below were run through the engine, and both figures were checked against the authority that publishes them — not against ourselves.
| HTSUS.6105100010.GEN | 19.7% |
| Additional measures | none in force |
| Checked against | USITC 2026 |
One rule, named, and nothing sitting on top of it. India carries no Section 301 measure, so this figure is complete within the tariff data we hold — which is a different and smaller claim than “correct”, and the only one we make.
| UK published duty | compound |
| Surviving 0% rule | not applied |
| Checked against | UK Tariff |
10% + 77 GBP / 100 kg
A 0% rule existed for this origin and would have produced a clean zero on a signed certificate. The engine refuses to use it: the duty it is an exception to cannot be read, so no figure is final. The duty is real and it is owed.
1,741 UK commodities are held this way — refused, with the published text kept, rather than dropped.
Every determination carries the reasoning that produced it — the valuation basis, the rules considered, the ones that applied, and anything it could not resolve. Replay the same shipment against the same ruleset and you get the same answer, bit for bit.
Works out the tariff code from what the goods actually are — the ingredients, the processing, the form — and records the reasoning beside the answer.
Customs value by jurisdiction, incoterms without double-counting freight, official exchange rates only, duty and import VAT in the correct order.
Issues an Ed25519-signed certificate over the complete determination. Anyone can check it, forever, with no account and no contact with us.
Ask it about a product whose description does not settle the question, and it does not quietly pick the likelier answer.
Guessing Piper would be right most of the time. Right most of the time, silently is the worst available property for a customs figure: the wrong entries file themselves, look identical to the correct ones, and turn up at audit with penalties attached.
Anti-dumping and countervailing duties routinely add 30–200%. They are in no figure we produce, and every determination says so — including the ones where the screen found nothing. A clean screen is not a clearance.
No official exchange rate, no answer. No VAT rate supplied for the member state, no VAT figure — just a flagged line naming exactly what is missing.
Every determination records the exact instant its ruleset was read, and the engine is a pure function of the shipment and the rules at that instant. Nothing else is needed to re-run your entire history as of any date you name — and tell you, in money, what changes and why.
The industry answers this question today with a consultant and three weeks.
Re-run last year's entries against today's engine. Where a rule changed, or where our classification got better, the difference is a number with a named rule behind it. Reclaim windows are finite; finding them by hand is why most of them close unused.
A new tariff schedule is published for next quarter. Run your real book against it before it takes effect and get the exposure per corridor, per commodity, per supplier — while there is still time to do something about it.
A machine-learned classifier structurally cannot do this. Its output is a function of model weights, and the model that decided a 2026 entry has been retrained several times by 2029. It can tell you what it thinks today. It cannot reconstruct what it thought then, and it cannot show you the named rule that produced the number. Determinism is not a purity preference here — it is the entire feature.
The compliance officer who filed it has left. Every record survived — nothing in this system is ever deleted — but reassembling that week out of a review queue, a certificate list, a dispute tab and an audit log is a day of billed panic.
Falmoran indexes the same data by day. One date, and you get everything that happened on it: every determination anchored, every classification a broker signed, every document issued, every amendment, every record set aside and by whom. Not that the record still exists — that a human can find it.
Setting a shipment aside removes it from your totals and leaves it on a list with the reason and the person. A record that vanishes with nowhere to look is, for every practical purpose, deleted.
Hashed, versioned by engine and ruleset, immutable. A determination is not recomputed on a later view — the record you saw is the record that stays.
One ZIP: every determination with its payload and hash, the full audit trail, dossier records with each PDF's checksum, and a manifest hashing every file. Open formats. No export fee, no notice period, no conversation.
Two companies that trade with each other have never had a shared, trusted copy of the same customs determination. The exporter has one spreadsheet, the importer has another, the broker has a third, and the first anyone knows about the disagreement is when an entry is queried.
A Falmoran network is a private group of companies. A member exposes one specific shipment's determination to the others, keeps the conversation attached to that record, and revokes access the instant the relationship changes.
Six companies in one network. Nobody can see anybody’s records.
Sharing is an explicit grant on one shipment, checked on every read. It is not a relaxation of isolation: your tables stay behind your own row-level security policies, untouched.
A counterparty gets a free workspace. They can read and answer what you share and nothing else — they cannot run their own determinations, and they cannot see a second thing you did not hand them.
Access is proven on every request, not cached in a session. Withdraw a share and it is gone from their screen mid-session, not at the next login.
Every tariff engine on the market returns a number, and a number is not an answer. None of them can tell you the difference between “no preference applies to this shipment” and “I was never given preference data for this country”. Both come back as a clean result. The first is a fact. The second is a guess wearing the same clothes.
Falmoran keeps a coverage manifest — which families of customs rule it holds, per jurisdiction — and it is derived from the rules table itself, never asserted by us. Every customer can read it. It is on the screen inside the product and it is on the certificate your bank receives.
| Rule family | United Kingdom | European Union | United States |
|---|---|---|---|
| Third-country duty | heldthe UK Global Tariff; duties expressed per unit of quantity return UNDETERMINED | held | held |
| Tariff preference | not heldevery UK figure is the third-country rate, so a shipment that qualifies for an agreement is OVERSTATED | partiala sample, not the schedule — EBA and most FTAs not yet loaded | held |
| Country-wide additional duties | partialthe loaded edition, less rows whose rate the engine will not approximate | not held | partialSection 301 lists 1, 2, 3 and 4A on Chinese-origin goods. Other statutes — 232 metals, 201 safeguards, 338 — and every other origin are NOT held, so no measure is not a clean screen |
| Anti-dumping & countervailing | partialanti-dumping, countervailing, safeguards and retroactive-registration measures, plus a separately maintained exposure screen | partialplus a separately maintained exposure screen | no machine-readable source |
| Tariff-rate quotas | partialmeasures only; live balances never held | partialmeasures only; live balances never held | no machine-readable source |
| Licences & permits | not held | not held | not held |
| Import VAT | from your payload | from your payload | does not apply |
This grid is a copy, taken 16 August 2026. The live manifest is in the product and on every certificate, and it is the one that governs. If this page and a determination disagree, the determination is right — it read the rules table; this page was typed. “Partial” means we hold some of a family and cannot tell you which shipments fall inside it, so it flags exactly like a gap.
Not “preference data unavailable”. “The duty shown may be TOO HIGH if the goods clear inside an open quota.” “The duty shown may be TOO LOW, routinely by 30–200%, if an anti-dumping order applies.” Then it names the official register and links it.
Where a missing family could have reduced a figure, the affected lines go to a review queue instead of getting a green tick. Only families that could actually move that figure flag it — a queue nobody can clear is a queue nobody reads.
Anti-dumping duty is the clearest case. On a €40,000 Chinese hot-rolled steel line the published third-country rate is 0%. There is also an anti-dumping duty, set per exporter, through a four-character code that appears nowhere on a commercial invoice. An engine that quotes 0% is not wrong about the tariff — it is silent about the part that costs money.
Falmoran holds those measures for the UK and the EU. It records the exposure, names which exporters carry an order, and refuses to state a rate it cannot derive — because that rate depends on which exporter shipped the goods, and no invoice says. On one Chinese steel line the engine loaded nine rules and refused thirty, each refusal naming why. Thirty silent approximations would have looked better and been worth nothing.
A flagged line is assigned to a customs broker you choose. They get their own sign-in and see only the lines you assign them — never your dashboard, your shipments or your figures.
A stamp records the fingerprint of the exact determination shown to them. If the shipment is re-determined afterwards, the stamp is refused. You cannot sign a document that changed after you read it.
“I need the mill certificate before I can classify this” is recorded, sent back, and the line stays flagged. The interface does not make signing the easy button and refusing the awkward one.
A broker who spots their own error can withdraw the stamp with a reason. The original stays on the record. Every profession that signs opinions corrects them with a new signed document, not an erasure.
Said plainly, because it matters: cross-company isolation is enforced by the database — row-level security, forced, under a non-superuser role. A broker sits inside one workspace, so their boundary is role gating and disciplined query paths. Both are real. They are not the same mechanism, and we will not tell an auditor that they are.
Your firm gets its own workspace: your licence and its expiry, your people, and one list of every line waiting for you across the clients who invited you — shown only where each client has said yes, and gone the moment they take it back. A public page with your licence and your own record, and a listing in the broker directory. Free for brokerages. Ask us to set it up.
Anyone can claim accuracy. These are the numbers with their denominators, and every one is reproducible from the repository.
Compared against the rates US Customs states in its own published rulings, across 17 chapters. The six differences are all pre-2004 tariff reductions, visible as a decline in date order.
Every loaded TARIC expression read identically by two independently written parsers. The 100× confusion between per-kg and per-100kg does not occur anywhere in the corpus.
Including 30 cases where every expected figure was worked out by hand from the rule text before the engine ever ran. You can check any of them on paper.
Sixty-five products taken from published CBP rulings and put through the engine with only the facts the letter states. Eleven came back with the officer's own six digits; the rest were refused with a question rather than answered with a guess. None came back wrong. Where a chapter has thin evidence the coverage page inside the product says so, rather than scoring it zero and calling that a clean sheet.
Grading the classifier against published customs rulings found a real misclassification in our own engine — chilli returned as black pepper, at high confidence, unflagged. It was fixed, and the test that caught it runs on every change. That benchmark stays in the repository precisely because it can embarrass us.
In September 2026 we asked the U.S. International Trade Commission a question about Chapter 99. A nomenclature analyst there replied: “I would suggest reading the entire subdivision carefully. For example, if you read subdivision 40(a) it applies to headings 9903.04.60–9903.04.69. This range would include 9903.04.66.” She was describing a hole in our reader without knowing it. Heading 9903.04.66 cites subdivisions (c) and (h)(iii); it never names (a), the subdivision that establishes the duty treatment for the whole range. We were following citations forward and had never read a subdivision’s own declared scope. It is fixed, and the fix refuses rather than guesses when a range cannot be read. The same analyst noted that interpreting the scope of a provision is CBP’s call, not the Commission’s — which is exactly where this engine stops and a licensed broker starts.
Chapters 61 and 62 carry the highest rates in the tariff — routinely 16%, up to 32% — against an average near 3% across all goods. The engine resolves them from four facts, in this order.
Chapter 61 is knitted, Chapter 62 is everything else. This changes the heading, not a digit. If the description does not say, the engine refuses and names the one question to ask — it never assumes the commoner case.
Section XI Note 2. Fibres are summed per tariff family, so 40% polyester plus 20% nylon is 60% synthetic even though neither reaches half. At exactly 50% there is no chief fibre and a person decides.
Chapter 61 Note 9 and Chapter 62 Note 8: a garment not identifiable as men's is classified as women's. A real legal default with a duty consequence — applied, and recorded as applied.
The engine stops at the international subheading. The US 8- and 10-digit statistical suffixes turn on details a product description does not carry, and a broker completes them.
The same shirt, one line apart on the spec sheet.
| Goods | Code | US duty |
|---|---|---|
| T-shirt, cotton | 6109.10.00 |
16.5% |
| T-shirt, man-made fibres | 6109.90.10 |
32% |
| Men's knit shirt, cotton | 6105.10 |
19.7% |
| Men's knit shirt, man-made | 6105.20 |
32% |
Fifteen and a half points on one word. On a $200,000 container that is $31,000 — and the only thing that changed was a percentage nobody checked.
A supplier who volunteers the gap is worth more than one who claims everything works. This list is kept current, and it is the first thing we show a customer.
In August 2026 we handed ten determinations — the full reasoning, the exact rule rows the engine used, and the published tariff text — to an independent adversarial review, with instructions to assume defects existed and to name them. It found four. Here they are, and what we did about each.
This is not a claim that the engine is validated. An AI review is not a licensed broker's signed opinion and we will not present it as one. What it is: evidence that we look, that we publish what we find, and that the corrections are in the product rather than in a backlog.
One more, for completeness: the review's first report named a missing rule record as its headline finding. It was a bug in our own audit tooling, not in the engine — a step name containing a slash was being split into a rule that did not exist. We fixed it, wrote a test per failure mode, and logged it in the same place as everything else. A vendor who only publishes the findings that make them look good is publishing marketing.
The full log — every finding, its category, what we did, and the ones we rejected with reasons — ships with the product and is handed to every customer. Ask for it before you buy, not after.
Everyone in this category writes “contact us for pricing”, which means “we will work out what you can afford”. Here are the numbers, and you can compute your own bill before you ever talk to us.
A monthly product call for the founding period, where you tell us what is wrong with it — and, if you are willing, let us look at the lines the engine got stuck on, anonymised. A real edge case from your catalogue is worth more to us than the discount costs.
A conversation about a reference or case study — afterwards. If it has worked for you, and subject to your approval of anything we write. We are not asking you to endorse software you have not used yet.
This is a limit on how many companies get the offer, not on how many people you put in — every plan has unlimited people.
You are always charged the cheapest of the four — the cheapest plan for the month you actually had. The plans are not walls you fall over: go one shipment past 200 and you pay $753.50, not $2,400.
The plans are not walls you fall over. Go one shipment past 200 and you stay on Essential and pay $3.50 for that one shipment — $753.50, not $2,400. Go one past 750 and you pay $3 for it: $2,403, not $3,600. We move you up only at the point where the bigger plan is genuinely cheaper for you: Starter wins at 672 shipments, Growth at 1,151 and Scale at 2,593, and at each of those the two prices are identical to the cent. Across every month from one shipment to five thousand, one extra shipment never costs more than $3.50.
| Your month | Plan you are billed on | Bill | Per shipment |
|---|---|---|---|
| 200 | Essential | $750 | $3.75 |
| 201 | Essential | $753.50 | $3.75 |
| 500 | Essential | $1,800 | $3.60 |
| 671 | Essential | $2,398.50 | $3.57 |
| 672 | Starter | $2,400 | $3.57 |
| 750 | Starter | $2,400 | $3.20 |
| 751 | Starter | $2,403 | $3.20 |
| 1,150 | Starter | $3,600 | $3.13 |
| 1,151 | Growth | $3,600 | $3.13 |
| 1,751 | Growth | $3,602 | $2.06 |
| 2,592 | Growth | $5,200 | $2.01 |
| 2,593 | Scale | $5,200 | $2.01 |
Onboarding: $4,000, once — $2,000 on signing, $2,000 when the validation report is delivered. Bounded, and here is the boundary — so neither of us discovers it afterwards.
| Included in the $4,000 | Limit |
|---|---|
| Historical validation — your own past shipments re-run through the engine, with a written line-by-line report of where our answer differs from what you filed | up to 3 months or 5,000 shipment lines, whichever comes first |
| Your own file layout mapped to the importer, however your columns are named | — |
| One administrator account created for your company | — |
| Two training sessions | — |
| Thirty days of priority support | — |
| Historical validation beyond the cap | separately scoped and quoted before any work starts |
Your team is yours to run. We create one administrator
account for your company and hand it over. From there your
administrator adds their own people, invites their own customs brokers
and assigns each person the lanes they are responsible for —
without asking us, waiting for us, or paying us to do it. Unlimited
people on every plan, so there is nothing to ration.
You are not charged to load your own products. Bulk Import takes
thousands of rows in one pass and your administrator runs it. We map your
column names to it once, and after that adding products is something you
do, not something you buy. There is no per-product fee and no catalogue
limit.
And we do not charge per lane. Your admin decides who covers
which lanes, in a screen, in seconds. On our side there is nothing to
set up at all: the engine determines against whatever rule coverage we
hold, and Rule Coverage is a page you can open and read before you
sign. A per-corridor fee would be a charge for work nobody does.
What the fee is not. It does not buy chapters or corridors being
built for you. We do not build coverage for one customer: anything we add,
every customer gets, on the same day, at no charge. So the fee is for work
that is genuinely yours — your history, your file layout, your
people — and never for work we would have done anyway.
That is also why the fee does not shrink as our coverage grows. The
eleventh customer pays the same as the first and gets more, because the
onboarding work is the same size and the product underneath it is bigger.
Licences and permits are not part of this. We hold no import
licence, permit or agency-filing data for any jurisdiction, the manifest reports
it as not held on every determination, and loading it is not included. Saying
so before you pay is cheaper for both of us than you discovering it afterwards.
Most customers connect nothing at all — their people work in Falmoran, or they drop a spreadsheet in. That path is included and costs nothing extra. The four options below only matter if you want us wired into CargoWise, Descartes or an ERP.
Connecting your own systems is priced separately, because it is the one thing that is not the same size for everybody. Onboarding is: the eleventh customer's history takes about as long as the first's. Connecting to a CargoWise instance is not, so folding it into one fee would mean either charging a small forwarder for work they will never need or under-quoting a month of engineering. Four shapes, and most customers are the first one.
| How your data reaches us | What it costs |
|---|---|
| Your people work in Falmoran. Bulk import, Products, the broker portal, certificates. Nothing to connect. | Included |
| Your system calls our API. Your developers build against it. Keys, documentation and a sandbox are on every plan — we do not charge for access to an API that already exists. | Included, with 8 hours of our engineering time supporting your build |
| We build the connection. You drop a file on SFTP or send it by email; we pick it up, determine it and put the results back. Nothing is built on your side. | $3,200 once |
| We build into your trade software. CargoWise, Descartes, an ERP. Field mapping, duplicate handling, retries, malformed data, outages, secret rotation, monitoring. | $3,000 scoping, credited in full against the build — then a fixed price, plus $500 / month while it runs |
What the $3,200 buys is up to five weeks of testing, not three hundred
lines of code. Writing the connection is the easy part. Finding out
that your commodity code column is called tariff_no, that
05/06/2026 means June in your system and May in ours, that
weights arrive in pounds, that the last row of every export is empty,
and that the same shipment occasionally arrives twice — that is
the work, and the only way to find it is to run your real files rather
than a tidy sample. We keep going until the surprises stop.
The five weeks start when your first real file arrives, not when
you sign. Almost every integration that runs late runs late because it
is waiting on data, and a clock that starts before there is anything to
work on measures the wrong thing for both of us.
Why the last one is scoped before it is priced. We have not seen
your instance, and a build at that depth brings in specialist engineers
under contract. Quoting it sight-unseen means one of us is wrong about
the number, and that is the same quote-first-discover-later problem this
page exists to avoid. The scoping week produces a fixed price you can
accept or decline, and if you go ahead you do not pay for it twice.
And why it carries a monthly line. A built integration is not
finished when it ships. Your schema changes, your vendor upgrades, a
field is deprecated, a credential rotates — and keeping the
connection working through all of that is ongoing work, so it is charged
as ongoing work rather than quietly absorbed until it stops being
maintained.
The honest boundary. We load rules for any chapter — the engine determines duty from a declared code on any commodity. What stays limited is classification: deriving a code from a product description works for a defined set of chapters, and everywhere else you declare the code and we do the rest. We tell you exactly which ones cover your goods during onboarding, in writing, rather than publishing a number here that is out of date by the time you read it. We would rather tell you that now than have you find it in week three.
Above 4,000 the rate steps down, and each block is charged at its own rate. Reaching a lower band does not re-price the shipments below it.
| Shipments in the month | Rate for those ones |
|---|---|
| 1 – 4,000 | covered by the $5,200 plan |
| 4,001 – 10,000 | $1.20 each |
| 10,001 – 40,000 | $1.05 each |
| 40,001 and above | $0.90 each |
A month of 50,000 shipments, worked through: $5,200 covers the first 4,000; shipments 4,001–10,000 cost $1.20 each ($7,200); shipments 10,001–40,000 cost $1.05 each ($31,500); shipments 40,001–50,000 cost $0.90 each ($9,000). Total $52,900. Published, so you can check your own invoice before you ever speak to us.
Two different numbers, and it is worth being clear which is which. $0.90 is the rate on the last block — what your eighty-thousandth shipment costs. Your average across the whole month is higher, because the earlier blocks were charged at their own rates: at 80,000 shipments the bill is $79,900, which averages $1.00. The average can never fall as low as the cheapest block, and any supplier who implies otherwise is quoting you a rate you will not actually pay.
The average only ever falls. $3.20 each at 750 shipments, $2.06 at 1,750, $1.30 at 4,000, $1.24 at 10,000, $1.10 at 40,000, $1.00 at 80,000, $0.95 at 150,000. And there is no cliff at any boundary: the 751st shipment costs $3.00, the 10,001st costs $1.05 — never a jump to a new tier. A threshold you can sit just under is a threshold people manage instead of crossing, and we would rather you used the software.
Month to month, no annual commitment. Annual prepayment is available and is two months free, but it is never a condition. The first five customers keep their opening price permanently.
Were you sent a Falmoran certificate? Check it yourself. No account, no sign-up, and we do not record who you are — asking a verifier to register would defeat the entire purpose of issuing a proof.
The signature is standard Ed25519 over the JSON-canonical form of the payload, and the public keys are published. A twenty-line script does the same check offline, without contacting us at all.
Forwarders, brokers and importers — tell us the corridor and commodity you work in. We will tell you plainly whether the engine covers it.