Methodology

This page exists because our teardowns ask you to trust numbers. Trust requires showing the machinery. Here is exactly how every report — public teardown or private sprint — is built, what the underlying data can and cannot see, and the rules we hold ourselves to.

Where the data comes from

Before a vessel arrives at a U.S. port, its carrier files manifests with U.S. Customs and Border Protection. For ocean freight, bill-of-lading data — shipper, consignee, notify party, ports of loading and discharge, cargo description, weight, container counts — is public record under 19 CFR §103.31, and members of the public (including the press) have inspected these manifests for decades. Several companies aggregate these records into searchable databases. The free, public tiers we use:

• ImportInfo (importinfo.com) — public bill-of-lading search with yearly counters and filterable queries (by date, shipper, consignee, commodity). Counts BOL documents (house and master bills separately). Row detail is partially blurred without a free account; CSV export is account-gated.
• ImportKey (importkey.com) — public search returning shipment-level counts, cumulative weights, and buyer/supplier aggregations. Anonymous use is rate-capped; public result sets are truncated.
• Panjiva public profiles (panjiva.com) — free company-level profiles showing shipment totals, top products, and customer counts for many suppliers. Useful as an independent cross-check on the aggregators above.
• PortExaminer (portexaminer.com) — free archive of individual bills of lading with full field detail. Coverage skews older; good for founding-era records.

For company context — revenue, funding, ownership, litigation — we cite primary or press sources directly (company sites, court dockets, reputable trade/business press) and always label those figures as company- or press-reported rather than verified by us.

How a report is built

1. Entity mapping. We identify the target's legal entities, addresses, and trade names from the bills of lading themselves, cross-checked against public filings or court captions. Brands rarely import under one clean name — mapping the entity web comes first, or everything downstream is miscounted.
2. Aggregate counters. We pull yearly/quarterly record counts from at least two independent indexes (typically ImportInfo + ImportKey) and reconcile them. Where the two disagree, we explain why (usually document-vs-shipment counting) rather than picking the number we prefer.
3. Record sampling. We read individual bills of lading — recent and oldest — to extract ports, carriers, cargo descriptions, TEU, and weights, and to confirm that keyword matches are genuinely the target's cargo rather than coincidental text matches.
4. Supplier and consignee structure. We compute shipper-of-record and consignee splits only where a verified numerator and denominator both exist in public counters. Where free tiers truncate the aggregation, we say so instead of extrapolating.
5. Cross-validation. Key suppliers are checked against an independent source (a Panjiva public profile, a government trade portal, the supplier's own site). Our HexClad teardown, for example, validated its main supplier count against Panjiva within 0.7%.
6. Labeling. Every figure in the report is one of three things: a number from a cited customs record, a number from a cited company/press source (labeled as reported), or our arithmetic (labeled as an estimate, with the method shown). There is no fourth category.

What the data can see — and what it cannot

Being precise about blind spots is not a disclaimer; it is half the product.

• Ocean only. U.S. import manifest data covers vessel-borne freight. Air, rail, and truck shipments are invisible to these sources. If a brand flies its product in, we will not see it — and our reports say so.
• Imports only, U.S. only. U.S. export manifests and other countries' import records are not covered by these free sources. (Canada-bound legs, for instance, appear only when they transit U.S. records.)
• De-identification gaps. Shippers can consign cargo to a freight forwarder or bank ("to order" bills), which hides the real buyer from keyword search. Our HexClad teardown found 32% of brand-marked volume consigned to a forwarder — and states plainly that further volume may be fully de-identified and invisible. Manifest-visible volume is a floor, not a ceiling.
• Documents, not dollars. Manifests describe cargo and logistics. They do not contain prices, contract terms, or duty amounts paid. When we estimate economics from volumes, the estimate is labeled and the method is shown.
• Free-tier truncation. Without paid accounts, aggregators cap result rows, blur record fields, and truncate supplier aggregations. A free-data report can verify who, when, and relative shares with high confidence, but not always absolute physical counts. Paid sprints close most of that gap with subscribed sources.
• Counter semantics differ by vendor. One index counts documents, another de-duplicated shipments. We state which universe each number comes from and never mix them silently.

The no-fabrication rule

Every number in a Manifest Sprints report traces to a specific record or cited source, or is explicitly labeled an estimate with its derivation shown. If a source fails mid-report — a paywall, a bot check, a rate cap — we disclose the failure in the Data Confidence section rather than route around it or fill the gap with a plausible guess. If free data can only see part of a target's flows, the report says exactly which part. We would rather publish "this is the 70% we can verify" than pretend to 100%. Any figure we cannot source, we do not print.

Each report ends with a Data Confidence section listing what is verified, what is partial, what is not covered, and which sources failed. Raw extracts with per-file provenance ship alongside paid reports so your team can re-pull every figure.

Corrections

If you believe a figure in any teardown is wrong, email corrections@manifest.jc.holdings with the figure and your counter-evidence. We verify against the underlying record and correct the public page, with a dated note, when we're wrong.