Reading a Competitor’s Container: How Customs Data Reveals the Factory Behind a Brand

November 19, 2026Industry Reports0 views34 min read

A bill of lading proves a shipment moved, not who made the goods. How to read customs records, test attribution and run a mirror check on trade data.

Key Takeaways

  • Weight is the one measure that does not change when a shipment crosses a currency or a valuation basis, which is why every serious comparison in this method is run in kilograms rather than dollars 5.
  • The nomenclature underneath all of this has more than 5,000 commodity groups at six digits, is used by more than 200 economies, covers over 98% of merchandise in international trade and is revised every five to six years 13.
  • Where two reporters agree within about 14%, the route is being described consistently and can be treated as real.
  • China declared USD 181.9 million of HS 230910 exports on 26.1 million kilograms to the United States in 2025, USD 217.5 million on 33.3 million kilograms to Germany, and USD 10.4 million on 1.6 million kilograms to France 20.
  • The United States reported USD 121.8 million on 20.8 million kilograms 21.
  • Germany reported USD 263.7 million on 36.1 million kilograms 22.
Reading a Competitor’s Container: How Customs Data Reveals the Factory Behind a Brand

The United States entry summary, the document a customs broker files to get imported goods out of bond, requires a manufacturer identification code. Not a shipper code alone: a manufacturer code, derived under instructions CBP publishes separately and read together with the rules of origin at 19 CFR Part 102 12. The American customs service wants to know who made the goods, because duty rates, trade-remedy orders and origin marking all attach to that answer.

The bill of lading, which is the document the commercial trade databases actually resell, does not carry that field. It carries the shipper, the consignee, the ports, a description of the cargo, a weight and a date 3. Four parties plausibly appear in a single pet-product consignment: the plant that extruded or sewed or moulded the goods, the export trading company that invoiced them, the brand owner that specified them, and the importer of record that cleared them. The public record names at most two of the four, and it names them in fields whose legal definitions were written for carriage, not for due diligence.

That asymmetry is the whole subject here. A customs record is strong evidence about a shipment and weak evidence about a supplier. A record establishes that a described quantity of goods moved, under a stated description, from a stated origin, on a stated date, consigned to a stated party. It does not establish who manufactured the goods, who owned them at any point, or whether the party named as shipper has ever owned a factory. Between those two statements sits every legitimate use of the method and every misuse of it.

The misuse is not exotic. It is the ordinary reading. A buyer searches a competitor’s brand, sees a Chinese shipper name on a hundred and forty containers, writes the name into a sourcing plan, and flies to a city to meet a company that turns out to be an export agent with four staff and a WeChat account. The record was accurate. The inference was not.

What follows treats the customs route as an instrument with a measurable evidentiary grade, in the same way that Where Buyers Find Pet-Supply Manufacturers Once They Leave Alibaba and Made-in-China treated the twelve discovery channels as instruments with different grades. The purpose is not to sell a subscription. It is to establish, field by field, what a record proves, where the proof stops, and which tests let a buyer tell a trading company from a plant before spending money on a flight.

The anatomy of the record: five documents, one of which reaches a database

A pet-product consignment generates a document set, and the set is worth separating because the documents answer different questions and only one of them becomes the commodity that trade-data vendors sell.

The commercial invoice, which is a contract artefact

The invoice names the seller, the buyer, the goods, the unit price, the Incoterm and the currency. It is the document a customs authority uses to establish value for duty, which is why the declared value on an entry is a customs value rather than a price a buyer can assume. Two features matter for a buyer reading customs data later. The seller on the invoice is the party that sold, which is frequently not the party that produced. And the invoice price is the price between the contracting parties, so where the shipper and the manufacturer are different legal entities in the same group, the intra-group price carries a transfer-pricing logic that has nothing to do with factory cost 4.

The packing list, which is the only quantity document

Cartons, units per carton, net and gross weight, dimensions, marks. The packing list is where the physical description of a consignment is fixed, and net weight is the field that survives into the statistical record. This matters more than it appears. Weight is the one measure that does not change when a shipment crosses a currency or a valuation basis, which is why every serious comparison in this method is run in kilograms rather than dollars 5.

The bill of lading, which is a receipt and a contract of carriage

A bill of lading performs three functions: it evidences receipt of goods for shipment, it evidences the contract of carriage, and, when negotiable, it functions as a document of title 6. The statutory definitions are narrow and worth quoting precisely. A consignee is “the person named in a bill of lading as the person to whom the goods are to be delivered”. A consignor is “the person named in a bill of lading as the person from whom the goods have been received for shipment” 7. Received from, not produced by. The definition of the party buyers most want to identify is a definition about physical handover, and it was drafted that way for a reason that has nothing to do with sourcing.

The manifest behind the bill of lading is also a regulated document, not a private one. In the United States the inward foreign manifest, its contents, its form and the advance filing of the cargo declaration are governed by 19 CFR §4.7, with the required manifest information listed at §4.7a 8. A carrier cannot decide to omit the shipper name to protect a customer.

The entry summary, which is the only document with a duty liability behind it

The entry summary is where the importer declares classification, value and origin and deposits duty. CBP describes it as the documentation necessary to determine duties, collect statistics and establish whether other legal requirements have been met, and it is due within 10 days of the cargo’s release, with entry documents due within 15 calendar days of arrival 9. This document carries the manufacturer identification code. It is also the document that is not public, which is the hinge on which the entire method turns.

Where a classification or origin question has already been litigated, the answer is published. CBP issues binding advance rulings on classification, appraised value and origin, maintains them in a searchable database and treats them as the agency’s stated position on a prospective transaction 10. For a buyer reading a competitor’s range, rulings are an underused instrument: a ruling on whether a particular pet chew is a prepared feed or a finished consumer product changes the duty rate, the import regime and the HS line the shipment will appear under, and therefore changes which search will find it.

What survives into a resold database, field by field

The commercial datasets are built from the manifest and bill-of-lading layer plus, in some jurisdictions, customs declarations purchased or obtained from the authority. The table separates the fields a buyer actually queries from the evidence grade behind each.

Field in the recordDocument of originVerified by an authority, or self-declaredPresent in resold US bill-of-lading data
Shipper name and addressBill of lading / manifestSelf-declared by the filerYes
Consignee name and addressBill of lading / manifestSelf-declared by the filerYes
Port of lading, port of unladingManifestCarrier-reported, checkedYes
Vessel, voyage, container countManifestCarrier-reported, checkedYes
Goods descriptionBill of ladingFree text, self-declaredYes, as free text
HS codeManifest (often added)Self-declared; inconsistentYes, frequently
Gross and net weightPacking list / manifestCarrier-reportedYes
Declared valueEntry summaryDeclared, duty-bearingFrequently absent
Country of originEntry summaryDeclared, legally consequentialSometimes, as a separate field
Manufacturer identificationEntry summaryDeclared, regulatedNo

The last two rows are the ones that matter. Origin is a legal determination, not a description of where a box was loaded, and the WCO treats it as its own technical domain precisely because origin and consignment are different concepts 11. Manufacturer identity exists in the entry file and is absent from the public layer. So the record a buyer can afford to buy is a record about movement, and the record that could answer the factory question is a record about declaration.

What the HS code does to the picture before any query is run

The vocabulary the record is written in is standardised even where the entries are not. The Census Bureau publishes the definitions the American series rests on, including the distinction between general imports and imports for consumption, and that distinction alone changes what a monthly figure means 12. The nomenclature underneath all of this has more than 5,000 commodity groups at six digits, is used by more than 200 economies, covers over 98% of merchandise in international trade and is revised every five to six years 13. Two consequences follow for pet products specifically.

The first is that six digits is coarse. HS 230910 is “dog or cat food, put up for retail sale”, which is one line for an entire category whose price range spans a value grocery bag and a freeze-dried premium pouch. Every competitor inside that line appears in the same statistical bucket, and their relative weight tells you almost nothing about their relative value.

The second is that pet products have no code. A plush toy classifies in Chapter 95, a nylon lead in Chapter 42, a ceramic bowl in Chapter 69, a cat tree in Chapter 94, a shampoo in Chapter 33. A buyer trying to map a competitor’s non-food range through customs data is not querying one line, they are querying a dozen, and the description field rather than the code becomes the only handle. The fact base records this as a structural limit and declines to construct a pet-products aggregate, which is the right call 5.

Three tools and the evidentiary grade of each

Three commercial products dominate this route, and a fourth source, the intergovernmental one, is not a competitor to them because it answers a different question at a different resolution.

Panjiva: relationship resolution at subscription price

Panjiva, now inside S&P Global Market Intelligence, is the reference product. Its own dataset page describes over 2 billion shipment records from 21 countries and 13 million company-to-company relationships, with history beginning in 2007 and a data source described as “publicly available customs declarations and trade bureau statistics” 14. What a buyer is actually paying for is the entity resolution: the machine-learning and natural-language layer that collapses variant romanisations, group subsidiaries and trading arms into a parent company. That is the hardest part of the method and the part a free tool does least well.

The same page carries three metadata fields that deserve more attention than the record count. Reporting frequency is listed as “variable”. Dataset latency is listed as “variable”. Point-in-time is listed as “No” 14. The third is the important one. A dataset that is not point-in-time returns the current state of the record, so a query run in November 2026 will not reproduce the result a buyer obtained in March 2026, because records are added, corrected and restated underneath. That is a problem for anyone who needs to defend a sourcing decision in writing a year later, and it is not a problem any amount of subscription spend solves.

Coverage is country-specific and narrower than a 21-country count suggests. The Wharton academic note on the dataset records China coverage in a 2011–2018 window 15. A buyer asking whether a Chinese factory shipped to Germany last quarter is therefore not querying the same corpus as a buyer reconstructing 2016. Panjiva’s list price is not published on the dataset page; it is sold through sales contact, which is itself a datum about the product’s price band.

ImportYeti: free, US-anchored, and honest about being a counter

ImportYeti is the entry point most buyers use first, because the search requires no login and no card. Its homepage displays a live counter that read 878,185,818 customs shipment records, updated daily, at the time of this reporting 16. The documented coverage is US import and export bills of lading plus Mexican import and export customs declarations, with bill-of-lading fields listed as shipper, consignee, ports, product description and cargo, and with “supplier” defined as a company outside the United States 3.

Two properties make it useful and one makes it dangerous. Useful: it is fast, it is free, and its field list is explicit about what it does and does not return, which is more than most vendors offer. Dangerous: the number on the homepage is a counter rather than a statistic. It increments as records load, so any figure a buyer quotes from it is a timestamp. The fact base recorded 715,151,312 on the same date that this reporting read 878,185,818, and the difference between those two numbers is not an error in either 516. It is what a live counter does.

ImportGenius: à la carte country datasets and a published floor

ImportGenius sits between the two: a US-anchored dataset with additional country files sold individually, a published entry price of USD 199 per month, and US history back to 2006 on the import side and 2014 on the export side 17. The à la carte structure is the practical difference. A buyer who needs Kazakhstan as well as the United States can buy Kazakhstan without buying a global licence.

Its published claims are also the clearest illustration of why vendor counts should never be cited as data. The same page states “3.3B+ records from 25+ countries” and “7.5B+ shipment records”, and separately shows a US dataset tile of 258,473,434 shipments next to a headline of “315M+ U.S. Import Shipments” 17. Four totals, two of them mutually inconsistent, none independently audited. The company also states that its data is admissible in federal court, which is a claim about provenance rather than a guarantee about any individual row.

What none of them is: a government statistical series

UN Comtrade is the fourth source and it is a different kind of object. The platform publishes annual and monthly imports and exports by product and trading partner, reported by about 200 countries and areas, classified the same way so that they can be compared, and, in its own words, sourced from national authorities rather than estimates, standardised by the UN Statistics Division, and revised when countries send corrections 18. The compilation standard behind it is the international merchandise trade statistics framework, which is what makes cross-country comparison legitimate at all 19.

The difference in resolution is total. A commercial database gives a named company, a container, a date and a port. Comtrade gives a country, a partner, a six-digit code, a year and a total. A buyer cannot find a factory in Comtrade. A buyer can find out whether a claimed trade route exists, at what unit value, and whether the two ends of it agree. The table sets the four side by side on the terms that matter to a decision.

SourceUnit of observationCoverageAccess and costWhat it adds that the others do not
PanjivaCompany-to-company shipment21 countries, from 2007, variable latencySubscription, price on applicationEntity resolution across name variants and group structures
ImportYetiShipment, via B/LUS import and export, plus MexicoFree search; paid APIA no-cost first pass with an explicit field list
ImportGeniusShipment, via B/L25+ countries, sold à la carteFrom USD 199 per monthSingle-country files without a global licence
UN ComtradeCountry-partner annual totalAbout 200 reporting economiesFree public API, higher limits on subscriptionThe mirror: what the other end of the route declared

The reconstruction exercise: from a shelf to a shipping record

The method buyers actually want is a reverse lookup. A product is on a shelf, or in a competitor’s catalogue, or in a marketplace listing. The question is who made it. Seven steps, in this order, produce a defensible answer or an honest statement that the record does not support one.

Step one: fix the product before the brand

The instinct is to search the brand. That is usually the wrong first move, because a brand with forty SKUs across six categories will produce a shipment history that mixes a Chinese filler of wet food, a Vietnamese supplier of toys and a domestic co-packer, and the average of those tells you nothing. Fix the specific product first: the pack format, the net weight, the claim structure, the material. Then search at the level of that product. In the non-food half of the range, where no HS code isolates the category, the description field is the only handle available, and a description search on “cat tree” returns a different and usually better set than a code search on 9403 5.

Step two: establish the import channel before running any query

Ask which market the product was bought in, then whether that market’s customs records are in the dataset being used. This is where most wasted effort occurs. ImportYeti observes US bills of lading plus Mexican declarations 3. Panjiva’s non-US coverage varies by country and by year 15. A competitor sourcing through a European distributor, or a UK brand importing through an Irish entity, may leave no record in either. The absence of a hit in a US-anchored dataset is not evidence that a brand does not import. It is evidence that the buyer searched the wrong book.

Step three: build the name set before the query

The named party on a Chinese export record can be the plant, the plant’s export arm, a Hong Kong holding company, a freight forwarder acting as shipper of record, or a consolidator. The same plant may appear under two romanisations and a group may export several plants under one trading name. A single-name query will therefore undercount. The workable approach is to assemble five to ten candidate strings first, from the brand’s own disclosure, from certificate registers and from the supplier’s website, and then query each against the dataset. Name matching is the analyst’s actual labour in this method, and it is the reason two buyers running the same query can reach opposite conclusions.

Step four: query consignee first, description second, code third

The consignee field is the most reliable thing in the record, because the consignee is a party with a duty liability and an address in the importing market. Querying a competitor’s legal name as consignee returns its import geography, which is rarely wrong. Only then invert the query and search the product description, and only then use the HS code, because codes in bill-of-lading data are often self-declared and inconsistently applied, so keyword and code searches return different sets and neither is complete 3.

Step five: read cadence, not total

A shipper with six containers a year to one buyer is a different counterparty from a shipper with monthly containers to twelve buyers, and the difference is commercially decisive: the first is a fill-in supplier, the second is a programme supplier with capacity reserved. Cadence also exposes the seasonality of a category, which for pet products is worth knowing before a buyer plans a launch. Total volume, by contrast, is the least informative statistic in the record, because it cannot distinguish a large low-value order from a small premium one.

Step six: cross the shipper against a site document

This is the step that converts an inference into evidence, and the one most buyers skip. Take the shipper name from the record and search it against a certification register or a facility registration whose scope names a physical site. If the shipper name matches the certificate holder for a plant in the right product scope, the trade record and the site documentation corroborate each other. If the shipper name resolves to a company with no site, no certificate and no registration, the record has told you about a trader. The instruments for that cross-check, and the difference between a social audit and a product certification, are set out in LinkedIn and Vendor Background Checks: What a Buyer Verifies Before the First Order and in the pack-level forensics of Working Back from the Shelf: Packaging, Barcodes and Certification Numbers as Factory Fingerprints. Social audit schemes and product certification schemes are separate instruments answering separate questions and are treated in Global Supermarket Pet Aisles: The Full Private Label Entry Path from Tender to Audit; none of them is a substitute for the attribution test above.

Step seven: state the finding at the grade it supports

A reconstruction ends in one of four sentences, and a buyer should know which one they are entitled to write. “This brand bought this product from this region in this period” is supportable from a consignee field alone. “This shipper is this brand’s supplier of this product” requires the shipper name to match a disclosed or registerable entity. “This plant made these goods” requires the shipper to match a certificated site with the right scope. “This lot came off that line” is not supportable from public data at all, and a provider who implies otherwise is selling an inference as a fact.

Two sibling pieces carry adjacent ground rather than repeating it. The layer that sits between a factory and a shelf in most markets, and which the name field usually conceals, is Importers, Agents and Distributors: Why Buyers Add a Layer Before They Reach a Factory. The routes that leave no public trace at all are covered in Referral: The Route Buyers Trust Most and Suppliers Reach Least, and the question of what a brand actually outsources, once the factory is identified, is What Local Brands Outsource: The ODM and OEM Work Pet Companies Hand to Factories in China.

The attribution problem: four parties, one name field

The single most consequential error in this method is a category error, and it is worth naming in the same words every time it appears. When a supplier ships under a trading company’s name, the customs record names the trading company. The record is not wrong. The reader has substituted one legal person for another.

Four roles that the record collapses into two fields

A consignment has a manufacturer, which owns or operates the plant that transformed the goods. It has a seller, which contracted to supply. It has an exporter of record, which filed the export declaration in the country of departure. It has an importer of record, which filed the entry and owes the duty in the country of arrival. And it has a brand owner, which may be none of the above, since a brand can own a specification and a trademark while owning no inventory at any point in the chain.

The transport document carries two of these roles, and which two depends on who booked the freight. A factory with its own export licence often appears as shipper, in which case shipper and manufacturer coincide and the record is unusually informative. A factory selling through an intermediary appears under the intermediary’s name, in which case the record is accurate and useless for the question asked. A consolidated container appears under a forwarder’s name and describes three factories’ output in one free-text line.

Why the trading-company convention is rational

The convention is not a device for evading due diligence. It exists because Chinese domestic trading companies hold export licences, value-added tax rebate handling and foreign-exchange settlement that a small plant may not, and because a buyer in a third country may prefer to contract with a Hong Kong entity for currency and legal reasons. The result is a legitimate structure that happens to break the naive reading of customs data. The same structure appears in Italy, Turkey, the United States and the United Kingdom, where an export management company or a group trading arm performs the same function.

The consequence for the reader is a rule. Before attributing a shipment to a manufacturer, establish that the shipper is the manufacturer or is disclosed as its affiliate. Note that a supplier who tells a buyer the name of its exporter of record is not thereby weakening its position; it is giving the buyer the one piece of information that makes verification possible, and a supplier who will not is telling the buyer something else.

Four tests for a trading-company shipper

None of these is conclusive alone. Run together, they separate a plant from an intermediary in most cases.

  1. Site evidence. Does the shipper name appear as certificate holder, facility registrant or establishment-number holder for a physical site with the right product scope? A trader has a registered office, not a production line.
  2. Breadth of the description field. A plant’s shipments cluster in one product family. A trader’s shipments span unrelated categories, because it fills containers with whatever its clients make.
  3. Consignee concentration. A plant usually serves a handful of buyers deeply. A trader serves many buyers shallowly, and its records show many consignees at low frequency.
  4. Name structure. Names containing the words for trade, import and export, industry and trade, or international in the original language are candidates for the trader category. This is a screening signal, not proof, and it should never be the only test applied.

What the American entry summary does that the manifest does not

The clearest demonstration of the gap is domestic to the United States. The entry summary requires an appraisement, a classification and an origin determination, and CBP publishes separate instructions for deriving a manufacturer/shipper identification code, read with the origin rules at 19 CFR Part 102 12. The manufacturer is therefore a declared field in the American customs file. It is not a field in the manifest-derived bill-of-lading data that the commercial platforms resell, which is documented in their own field lists: shipper, consignee, ports, product description, cargo 3.

So the record that answers the factory question exists, is filed with a government, and is not on the market at the price a sourcing manager can pay. That is a structural feature of the American system rather than an oversight by the vendors, and it explains why the same three products can be sold at three price points without any of them being able to name a plant with certainty. A buyer who needs manufacturer-level attribution from a customs record should be looking at a different instrument entirely, and should not pay a premium for a promise the data cannot keep.

The mirror test, and the two worked examples

The most useful thing public trade data does is not identify a supplier. It is to check whether a claimed trade route is described the same way at both ends. The method is cheap, runs on free data, and produces findings that no commercial dataset can produce, because commercial datasets observe one end of a flow.

The method, in four moves

Move one: take a claimed corridor, meaning an origin, a destination and a commodity. Move two: ask the destination what it received from the origin, using the destination’s own imports from that partner. Move three: ask the origin what it shipped to the destination, using the origin’s own exports to that partner. Move four: compare in kilograms rather than dollars, because weight survives the difference between a free-on-board export valuation and a cost-insurance-freight import valuation while value does not.

The reconciliation logic that follows is the analytical core. Freight and insurance should make the importer’s figure modestly higher, in the range of a few percent. Anything larger than that has a different cause: origin recorded against consignment, goods routed through an entrepôt and attributed by the exporter to the trader rather than the destination, transshipment and re-invoicing, or a genuine declaration error on one side. Where two reporters agree within about 14%, the route is being described consistently and can be treated as real. Where they disagree by an order of magnitude, the disagreement is itself the finding.

One commodity, three destinations, and one gap that valuation cannot explain

The same commodity line in the same year, read three ways, shows how quickly a valuation explanation runs out. China declared USD 181.9 million of HS 230910 exports on 26.1 million kilograms to the United States in 2025, USD 217.5 million on 33.3 million kilograms to Germany, and USD 10.4 million on 1.6 million kilograms to France 20. The importing declarations describe a different pattern. The United States reported USD 121.8 million on 20.8 million kilograms 21. Germany reported USD 263.7 million on 36.1 million kilograms 22. France reported USD 28.0 million on 5.0 million kilograms 23.

Read the value gaps first. Germany’s mirror is 21% above China’s declaration, which is the direction and roughly the size that cost-insurance-freight accounting should produce against a free-on-board export value. France’s mirror is 169% above, which no freight charge explains.

Then read the weight gaps, because they strip the valuation explanation out. Germany’s weight mirror is 8.4% above China’s declaration, so most of the German value gap is freight and the physical flow broadly agrees. France’s weight mirror is 216% above, so the goods themselves are unaccounted for at the origin end. The United States runs the other way on both measures, with the weight mirror 20% below China’s declaration, and a cost-insurance-freight adjustment cannot make goods disappear. A single mechanism fails all three tests, which is why the method requires the weight comparison rather than the value one.

The Gulf: 2,091,870 kilograms out and 173 kilograms in

The United Arab Emirates declared 2,091,870 kilograms and USD 5,977,735 of re-exports of HS 230910 to Saudi Arabia in 2023. Saudi Arabia declared 173 kilograms and USD 521 of imports from the UAE in the same year. The ratio is roughly twelve thousand to one 24.

This is not a rounding difference and it is not a data error in the ordinary sense. The explanation sits inside the Saudi record. Saudi Arabia declared 8,160,860 kilograms of Thai-origin imports in 2023, while Thailand declared 3,789,352 kilograms of exports to Saudi Arabia, so the Saudi book contains 4.37 million kilograms more Thai-origin product than Thailand says it shipped 24. Goods passing through a free zone without processing keep their original origin, and Saudi customs reports origin while Dubai’s re-export declaration reports destination. The same pallet is counted once in each system under a different attribute, and the two systems cannot be netted. Re-export volumes fell into that gap: for the same product and market pair, the Gulf re-export route is real, and neither customs service can size it alone.

The lesson transfers directly to a buyer. If a competitor’s product arrives in a Gulf market with European origin and no European shipment to that market can be found, the product probably moved through Dubai and lost nothing but its routing in the record. That is a purchasing insight: the brand is buying from a distributor, not from the plant.

Kazakhstan and Kyrgyzstan: seventeen to one, pointing away from the expected story

Central Asia is routinely described as the overland corridor by which Chinese goods reach Russia. For retail pet food, the China-to-Central-Asia mirrors agree to within 14% in both directions, which means the Chinese route is described consistently and is small: the five Central Asian states together took USD 2.17 million of Chinese retail pet food in 2024, or 0.15% of China’s exports in the commodity. The corridor that breaks is Kazakhstan into Kyrgyzstan, at seventeen to one, and the break points away from China rather than toward Russia 25.

The Kyrgyz export record explains the direction. In the one year Kyrgyzstan filed one, it declared USD 528,960 of exports on 207,480 kilograms, going to Mongolia, Uzbekistan, Turkmenistan, Turkey and France, with no Russia line at all 25. A country can be a documented transshipment economy for container goods and report almost nothing in a heavy, dated, language-specific, border-regulated food category. The corridor narrative is documented for machine tools and communications technology; it transfers to pet food only by association, and the mirror test is precisely the instrument that shows the transfer does not hold.

What a broken mirror licenses a buyer to say

A broken mirror licenses three statements and no more. The route exists at a scale one end reports and the other does not. The attribution of that route is not resolvable from public data. Any market size built from one end alone will be wrong by an order of magnitude, in a direction that depends on which end was used. That is enough to stop a buyer from accepting a distributor’s claimed regional volume, and it is not enough to name a factory.

What the public data can and cannot do

The public record earns its place because it is the only trade evidence that is not for sale. Its limits are equally structural, and they are the reason the method has to be described as comparative rather than absolute.

Strength: it is the mirror of record for national statistics

Comtrade’s authority comes from being a compilation of what national authorities declared, standardised into a common classification and common currency, and revised when those authorities send corrections 18. Eurostat publishes the European counterpart from the same national declarations 26. This is not a vendor’s scrape. It is the number a statistical office will defend, which is why it can be used to contradict a company’s own claim about its export markets.

Limit: annual, six-digit, and reporter-based

Annual data lags. Six digits merges a value bag with a premium pouch. Reporter-based compilation means the record describes the reporter’s own trade as the reporter chose to declare it, which is why the mirror is a test rather than a total. The fact base’s own decision to withhold a blended unit value across its tables, on the grounds that dividing value by weight mixes tiers, contract terms and valuation bases, is the correct discipline and is worth copying 5. A unit value computed inside a single partner line over a single year is a different and defensible object.

Limit: the record can be withdrawn, and withdrawal is not a market event

Russia filed 31 partner rows for HS 230910 imports in 2021 27. It filed none for 2022, 2023, 2024 or 2025. The queries return zero rows, not zero trade, and the difference between those two statements is the difference between a competent reading and an artefact. The same silence appears elsewhere: the UAE filed no import row for the commodity in 2025, while its 2023 export record returned 32 partner rows 28.

This is the risk the article wants to state plainly. Where a country’s data disappears, the trade route it was carrying does not disappear with it. It continues, and it becomes unobservable. A buyer or an analyst who reads the absence as an absence of trade will understate a live flow, and will do so in exactly the markets where enforcement attention is highest and the routing is least transparent. The Russian pattern is documented in this series’ own treatment of that market 29, and the methodological point generalises: a gap in the record is a fact about the record.

Limit: the record cannot see below the reporting threshold

Statistics are compiled from declarations, and declarations have thresholds. The US statistical programme excludes merchandise in transit through the United States from one foreign country to another, excludes certain low-valued transactions from full reporting, and treats goods in Foreign Trade Zones under separate rules for general imports and imports for consumption 4. A small-format competitor selling cross-border in low-value parcels may be genuinely absent from the record rather than badly searched. For a category where marketplace listings increasingly ship in small parcels, that gap is growing rather than shrinking, and no amount of query refinement closes it.

Why comparative readings carry the weight

Put together, the limits explain the finding this article would ask a reader to keep. Single-shipment attribution carries less weight than buyers assume, because the shipper field cannot distinguish a plant from a trader and the manufacturer field is not public. Product-mix and unit-value comparison across partner lines carries more, because those are stable aggregates that many declarations contribute to. The mirror test carries most, because it compares two independent declarations of the same physical flow.

An example of the middle category shows why. In the Kazakh import book for 2024, Russian product lands at USD 1.16 per kilogram, German at USD 2.72 and Polish at USD 3.04, on reported value divided by reported weight within each partner line 25. Those are not three prices for one product. They are three product tiers serving three buyer types, and the ranking is robust in a way that any single container’s declared value is not. A buyer who wants to know where a market’s value tier is sourced reads that table. A buyer who wants to know which factory filled one container is reading the wrong instrument.

Where the figures stop

Six caveats govern how far any of the above can be pushed, and each has bitten a real sourcing decision.

Coverage windows are not disclosed per query. Panjiva’s China coverage is documented in a 2011–2018 window in the academic note on the dataset 15, while its marketing describes 21 countries without publishing the list. A buyer cannot tell from a result set which years of which country were searched. The only defensible practice is to record the query, the access date and the corpus as described at that date, and to treat the result as a search of a corpus rather than a search of the world.

Reproducibility is not offered. Panjiva’s metadata lists point-in-time as “No” 14. ImportYeti’s counter moves daily 16. A buyer who needs to defend a decision should capture the underlying records rather than a screenshot, and should expect that the same query next quarter returns a different number for the same historical period.

Absence is ambiguous. A shipper that sells through a trader, ships to a market outside the corpus, or moves goods in low-value parcels leaves no trace. Absence of records is therefore not evidence of a false claim by a supplier, and treating it as such is the second most common error in this method after the attribution error.

Value is not price. Declared value on an entry is a customs value, frequently absent from the manifest layer, and never a factory cost. Dividing a declared value by a unit count produces an artefact that looks like a margin and is not one.

Vendor claims are claims. Four inconsistent totals appear on one vendor’s homepage 17, and two different counters appeared for the same free tool on the same day 16. None of this makes the underlying records false. It makes the marketing unusable as evidence.

These are not the compliance instruments. A customs record says nothing about how a factory treats its workers or whether its process controls are sound. Social audits and product certifications answer those questions and are a separate instrument set: a social audit tells a buyer about labour conditions, a product certification about process control, and neither certifies that goods may lawfully be sold in a given market. That distinction, and the reason the two are so often conflated in supplier decks, is set out in the series’ treatment of the private-label entry path and is not re-argued here 24.

The method’s honest summary is narrow. A customs record can tell a buyer that a named party moved a described quantity of goods between two places on a date. Read comparatively, across partner lines and years, the same data can tell a buyer where a competitor’s category is sourced, at what tier, and through which route. Read absolutely, from a single row, it will tell a buyer a name, and the name will sometimes be a trading company’s.

Method and evidence

This article was reported from primary documents and free public data between 1 and 15 November 2026, with a data cut-off of November 2026. The documentary anatomy was built from the statutory and regulatory texts themselves rather than from secondary descriptions: the WCO’s own description of the Harmonized System for the nomenclature’s scale and revision cycle, 19 CFR Part 4 for the inward manifest, 19 CFR Part 102 for origin and manufacturer identification, 49 USC §80101 for the definitions of consignor and consignee, and the Census Bureau’s guide to the trade statistical programme for the unit of record, the exclusions and the low-value treatment. The CBP pages on the entry summary and Form 7501 supplied the filing deadlines and the existence of the manufacturer/shipper identification code instruction.

Trade figures were taken from the UN Comtrade public preview API, queried directly during reporting at HS 230910, annual, total trade. Query construction and the endpoint behaviour, including the rate limits that return HTTP 429 and the retry discipline those impose, follow the developer documentation published alongside the database 30. Reporter codes used were China (156), the United States (842), Germany (276), France (251), Russia (643) and the UAE (784). Row counts returned by each query are recorded in the evidence ledger, because in four of these cases the informative result was the count rather than the value. Several queries returned HTTP 429 responses on first attempt and were resolved on retry with backoff; those figures are complete answers rather than partial pages.

The two worked examples are carried from sibling articles in this series rather than re-derived here: the UAE-to-Saudi mirror break from the Gulf channel piece 24 and the Kazakhstan-to-Kyrgyzstan corridor from the Central Asia piece 25, each with its own arithmetic caveat intact. Where a sibling’s figure is quoted, the year and the scope travel with it.

Three vendor documents were used and all three are treated as claims rather than measurements: the S&P Global dataset page for Panjiva, the ImportGenius homepage and the ImportYeti homepage and API documentation. Where a vendor publishes contradictory totals, both are reported and neither is averaged. No subscription was purchased and no records were obtained from a paid tier, so the article makes no claim about the accuracy of any individual record in any commercial dataset.

Excluded deliberately: any estimate of how often shippers are trading companies rather than producers, because no published figure exists and a percentage invented here would be a statistic of convenience; any pet-products aggregate built from multiple HS chapters, because the fact base declines to construct one and the refusal is methodologically correct; and any forward-looking statement about coverage or pricing at any of the four sources, all of which change without notice.

Social audits and product certifications are referenced only as distinct instruments. They are covered in their own right elsewhere in the series and are not re-explained here.

Featured Companies

Companies from the GlobalPetIndex database active in this area of the pet industry.

Browse the full company directory →

Related reading

GlobalPetIndex Editorial
GlobalPetIndex Editorial Editorial lead at GlobalPetIndex, covering pet industry intelligence, market trends and company research.

Editorial lead at GlobalPetIndex, covering pet industry intelligence, market trends and company research.

Share this article:

Putting this into practice?

Low-MOQ, private label and compliance checks for small pet retailers

Small-pet-business sourcing hub →

Dataset version 2026-11-19 · 50,000 indexable profiles · 123,610 published

GPI

GlobalPetIndex

Open the app for a smoother experience

Open Privacy