Product & Party Master Data

There is no "the HS code."
Origin is a fact, never an inference.

One product record and one party record per tenant — with per-jurisdiction customs classifications hanging off the product, country of manufacture and country of origin kept as different facts, party identity / roles / registrations as separate axes, and change detection that flags when a supplier or spec moves in a way that should trigger re-review.

1
product record · N jurisdiction classifications
0
origins inferred from a manufacturer country
3
party axes: identity · roles · registrations

The problem

Your master data assumes one answer to questions that have several.

"We import the same part into the US, the EU, and Mexico. Our 'HS code' field is one value — those are different classifications with different review status."

One code, many jurisdictions

"Origin gets set to whatever the manufacturer's country is. That's not how origin works, and it's a Reasonable Care problem."

Origin ≠ manufacture

"A supplier changed. Nobody re-checked the parts we source from them. Is this party a 'verified supplier'? Depends who you ask."

Stale, ambiguous parties

How Qubere addresses it

Model the facts as they actually are — separate, and each with its own status.

  • One product holds what's true everywhere; a US classification, an EU classification are separate records, each with its own status, reviewer, effective window. Only APPROVED counts.
  • Country of manufacture and country of origin are different facts. Origin is never inferred from a manufacturer, supplier, seller, export, or shipping country.
  • A party has roles (supplier, importer, carrier…) and per-country registration claims, each with its own lifecycle — not one "verified" flag.
  • A name match produces candidates for a human to confirm — never an auto-merge.
  • Change detection closes the loop with screening: a supplier moves → the parties sourced from them get re-screened.

One product, many classifications

A US classification can be approved while the EU one is mid-review — on the same product.

The most important design point for a multi-region import program. A classification is a reviewable case with the GRI analysis, the rulings referenced, and the reviewer's identity — not a code in a field.

  • Each classification: own status, reviewer, effective window
  • Sourcing evidence attached and inspectable
  • The product page opens with a "Needs your attention" strip: classifications waiting for approval (or none approved yet) → Classification History tab; open re-checks → the re-check block on the page; unverified country of origin → Trade & customs tab; master data not reviewed or changed since review → Overview. With nothing outstanding it says so
  • The product page is organised into tabs: Overview, Identifiers, Master & aliases, Attributes, Composition, Parties & manufacturing, Trade & customs, Classification History, Compliance Profile, PGA requirements, Value Intelligence, FTA qualifications, Evidence & lineage, Documents and History
  • AI enrichment only proposes: each suggested attribute is marked new, changed (with the current value) or unchanged, a person picks which to keep, and an applied enrichment can be reverted. "Create canonical record" (Master & aliases) reports whether it created a record, merged into an existing one or found it already known; a merge into this product's own record can be undone with one click, a newly created record cannot
  • A product can carry Kit/BOM components — sub-parts stay individually classifiable while the kit is bought and sold as one item
How to demo
/app/products/[id] → the classifications section: US = APPROVED, EU = IN_REVIEW. Then open the classification case — the reasoning, the rulings, who approved it.
Products/BMS Control Board Rev C
AC
Alice · Acme
JurisdictionClassificationStatusReviewerEffective
🇺🇸 US HTS8537.10.9170APPROVEDD. Osei (broker)2026-04-01 →
🇪🇺 EU CN8537 10 91IN_REVIEWpending—
🇲🇽 MX—not started——
US case CASE-8841GRI 1 · CROSS N301245 cited · evidence: datasheet p.2

Product master-data review

The product record itself goes through the same reviewer gate its classifications do.

Product.reviewStatus is a separate lifecycle from any classification, origin, or trade-remedy decision attached to the product — it's the sign-off on the record itself: its name, description, identifiers. A product can carry an APPROVED US classification while its own master data has never been looked at, and the reverse is also true. Neither one is allowed to stand in for the other.

  • UNREVIEWED → IN_REVIEW → APPROVED / REJECTED / NEEDS_REVIEW, enforced the same way classification transitions are
  • Approving requires a named reviewer and the products.review.approve permission — checked from the actor, not just the route, so no caller can approve a product it isn't allowed to
  • REJECTED and NEEDS_REVIEW only ever loop back to IN_REVIEW — there's no path that lets either jump straight to APPROVED
  • Every transition writes an audit entry (product.review.*) with who did it and when
  • A product a reviewer creates by hand can be approved on creation — same permission check, same named reviewer, just folded into one step instead of two
  • Approving a page of products no longer means one click per row: select a batch from the list and approve it in one call — each product still goes through the identical single-item gate, so a batch of 40 with one ineligible row approves the other 39 and counts the one failure instead of skipping it silently (the API returns each product's outcome and error by id)
How to demo
/app/products/[id] → the review badge next to the product name. Request changes on an IN_REVIEW product, then start review again and approve it — the history shows every hop with its reviewer. Then from /app/products, select several IN_REVIEW rows and use Approve selected to approve the batch in one action.
Products/BMS Control Board Rev C
AC
Alice · Acme
Review
ActionStart review · Approve product · Request changes · Reject
Approving needsproducts.review.approve
ActionTransitionReviewerWhen
APPROVEIN_REVIEW → APPROVEDD. Osei2026-10-02
START_REVIEWNEEDS_REVIEW → IN_REVIEWD. Osei2026-10-01
REQUEST_CHANGESIN_REVIEW → NEEDS_REVIEWD. Osei2026-09-30
Bulk approve (products list, 3 selected)
Result (illustrative)Approved 2, 1 failed
Per-product outcome (API)PRD-2309 — not yet IN_REVIEW

Origin as a fact

Qubere will not fill an origin field to look complete.

Country of manufacture and country of origin are stored as different facts. If there's no approved origin determination, the system says so — and the AI assistant refuses to substitute a manufacturer or supplier country, even when its own fields mention one.

  • A deliberate Reasonable Care stance, enforced end to end
  • Origin determination is a reviewed case with its own evidence
How to demo
On a product, show manufacture country populated but origin marked "no approved determination." Ask the assistant for the origin — it declines to guess.
Products/Origin
AC
Alice · Acme
Physical facts (not origin)
Country of manufactureVietnam
SupplierHanoi Precision Assembly
Ship-from (last shipment)Vietnam
Legal country of origin
Determination statusNONE — not inferred from any of the above
ActionOpen origin determination case

Party master · identity / roles / registrations

A party can be a VERIFIED-registered supplier in one country and an unreviewed identity overall.

Those aren't the same question and Qubere doesn't pretend they are. PartyRole records what a party does; PartyRegistration tracks per-country claims through their own CLAIMED → UNDER_REVIEW → VERIFIED lifecycle — independent of the party's UNREVIEWED → IN_REVIEW → APPROVED status.

  • The party page is grouped as Overview, Identity (names, identifiers, registrations), Locations & Contacts, Roles & Relationships, and Compliance & History (evidence, documents, screening, history, continuous monitoring)
  • Each party carries a screening summary (status, last result, staleness)
  • Match → candidates for a human; never auto-merge, never auto-split
  • Duplicates are ranked by the evidence they share. A merge shows a plan of every row that would move, applies only that confirmed plan, retires the absorbed party rather than deleting it, and can be undone from the party page within an undo window (the page shows days left)
  • A review queue (Parties › Review) lists parties waiting for approval with the checks a reviewer should see, and approves a batch while skipping flagged parties unless the reviewer confirms they looked
  • Address entry now autocompletes live against HERE, so registered/mailing addresses land clean and verified — better input for screening and for network/ownership resolution
How to demo
/app/parties/[id] — a party with two roles and a country registration mid-review while the party itself is APPROVED.
Parties/Hanoi Precision Assembly
AC
Alice · Acme
Roles
Suppliersince 2024 · 41 shipments
Manufacturer2 products
Registrations
🇻🇳 VN business registrationVERIFIED
🇺🇸 US MIDUNDER_REVIEW
Screening summaryCLEAR · last 2026-08-25 · stale (identity change)

One party, every role

The same company, wherever it shows up.

A supplier that later becomes your client's importer of record doesn't start a second life as a new record. It's the same party gaining a role — its screening history, its known aliases, every document already linked to it, all carried forward. Search for the company once and see everything it does for you, not whichever module happened to create it first.

  • Registering an existing party as an importer adds a role — it never duplicates identity
  • The importer's CBP registration, POA, and bond are extensions on the role, the same way a carrier's SCAC and insurance status are
  • A brand-new importer bridges into the party master automatically on creation — but an ambiguous match is never auto-linked, it waits for a person to confirm
  • The same identity resolver is shared with apps/tms carrier onboarding — one party master across the platform, not one per app
  • Cross-entity search (⌘K / Ctrl+K) finds a product or party from anywhere in the app — an EIN or a part number both resolve straight to the record
How to demo
Search "Meridian" → one party, tagged Supplier. Run Register as importer from onboarding → search "Meridian" again → same party, now Supplier + Importer, with a full "also known as" trail.
Parties/Meridian Components GmbH
AC
Alice · Acme
Roles
Suppliersince 2025 · 3 products
Importer of recordReady to file
Also known as
Supplier on 3 products
Importer of record for Northwind Retail Inc.
Screening summaryCLEAR · carried forward from the supplier record

Change detection & CSV import

A supplier moves → the parts you source from them get re-screened → you find out.

Signals that a product's spec, sourcing, or supplier moved in a way that should re-open review — surfaced, not silent. CSV import matches against existing records rather than blindly duplicating, and flags ambiguous matches for a human.

  • Won't silently create 500 duplicates; won't silently merge two real entities
  • Change history drives RDPS re-screening (see Compliance deck)
How to demo
/app/products/import → the import wizard → the match/review step. Then a product's change history with a flagged change awaiting review.
Products/Import · match review
AC
Alice · Acme
CSV rowBest matchScoreDecision
"BMS Ctrl Bd RevC"PRD-2291 BMS Control Board Rev C0.96auto-match
"Enclosure ALU 240"PRD-2288 / PRD-23120.71 / 0.69review
"Cell 21700 5Ah"——create new
Change flagged · PRD-2291supplier region changed → re-open classification + origin review

Product Intelligence

The product master keeps checking itself after the filing is done.

Matching a line item to a product isn't the end of the story. Once a filing is accepted or released, Qubere compares what was actually filed against the product's current approved classification and verified origin — and flags a mismatch for review, even if the filing already went out the door. An unusually priced line gets the same treatment against that product's own historical pricing.

  • Exact-match resolution rate and reviewer override rate over a trailing window (30 days by default), computed from stored match records, so a drifting matcher gets caught, not just felt
  • Average and median time for a person to resolve a match proposal, plus counts of re-validation and value-anomaly exceptions raised — a slow review queue is visible, not anecdotal
  • These metrics are read through the Copilot assistant (get_product_intelligence_metrics); there is no dedicated metrics screen
  • The post-filing check runs for accounts that have product historical learning enabled
  • Never retroactively guesses at data that can't be verified — old exceptions that predate product-linking stay unlinked rather than being attributed to the wrong product
How to demo
An accepted filing whose line was filed with a classification that no longer matches the product's current APPROVED code — the resulting exception. Then ask the Copilot for the product-matching metrics (match resolution rate, override rate, time to resolve).
Copilot/Product intelligence metrics
AC
Alice · Acme
Exact-match resolution ratexx%
Proposal override ratexx%
Avg / median time to resolvexx h / xx h
Exceptions raised in the window
Re-validation requiredn
Value anomalyn

The master, extended

Everything that hangs off identity is its own managed record.

◇ Importer registration — an importer role on the party: CBP number, POA, and bond, linked to filings
◇ Power of Attorney — POA records per client
◇ Bonds — amount, expiry, sufficiency checks feed filing validation
◇ Clients — the account relationship, auto-routing, portal scope
◇ Documents — any file links to any of Shipment/Party/Product/License/Filing; the Evidence tab resolves those links to the actual document and shipment, not a raw id
◇ Identifiers — Exporter Tax ID / Importer EIN / CBP Importer No. auto-extracted from parsed documents onto an already-resolved party, never overwriting a value that's already there
◇ First-sale arrangements — a valuation registry entry can now name the Product Master record its price governs; the product's own Compliance Profile lists every arrangement that points back to it

Assistant reads all of it — get_product, get_party, get_party_history, get_product_origin_position — RBAC-gated, through the same services the screens use.

Objection handling

Every hard question, answered directly.

"We already have an ERP item master / MDM."Qubere isn't replacing the commercial item master — it's the customs master: per-jurisdiction classification, origin determination, sourcing evidence, screening summary. Import from the ERP; be the system of record for the trade attributes the ERP models badly.
"Auto-classification?"AI assists (GRI analysis, ruling references) and can auto-approve above a confidence threshold you set. Assisted, auditable, reversible — never a black box that silently sets the field.
"PEP / beneficial ownership on parties?"PEP (politically exposed person) screening is not built. Ownership and relationships are: each party has a Relationships tab, network findings show the ownership graph behind a screening hit, and a 20-character LEI can be looked up against public GLEIF data and attached as evidence without changing party facts.
"If a supplier becomes an importer, do we lose their history?"No — it's the same party gaining a role. Screening history, aliases, and every document link carry forward; nothing restarts from zero.

The close

"One product, N jurisdictions, each approved-or-not. Origin never inferred. That's the foundation everything else trusts."

Next step: send us a slice of your product catalog. We run the CSV import live — you see the match-don't-duplicate behavior and the per-jurisdiction model on your own data.

Full detail: docs/apps/customs/sales/product-and-party-master.md · docs/apps/customs/data/product-master.md · docs/apps/customs/data/party-master.md