We connect an annual disclosed seller–buyer relationship to each company's product segments, then look for directed supply links between those products.
01Disclosed relationship
Four sources supply sales and purchase records. Annual flows, interim flows, and balances retain separate meanings.
02Candidate product path
We follow each company’s SAM product strands to find directed supply links. Each pair of strands keeps its most detailed supported connections. Endpoints may be at different levels, and separate product branches remain separate possibilities.
03Independent evidence
Same-year tender awards are held out of construction and ranking. A title mention can corroborate a product; absence cannot rule it out.
Does the signal exceed broad category similarity?
Compare the share of disclosed relationship-years with at least one product link against shuffled controls. These are coverage rates, not accuracy scores. The level comparisons below are research sensitivity checks from the original study; the explorer automatically considers every level.
Disclosed pairsBuyer-category controlSame-year random
How to interpret the ordering
Within each link type, a connection is removed only when a more detailed connection covers the same pair of product strands. Alternatives that are more detailed at different ends are retained. Results are ordered by the combined detail of both endpoints, then segment relevance. Revenue shares show importance to each company, not how much of their transaction involved this product. Missing or out-of-range shares are labelled unavailable.
Time and coverage
Latest available annual disclosures are used per filer, year, and source. Overlapping source lines are not summed. Current company resolution and current product ontology mean this is retrospective attribution, not a historically vintage-safe investment backtest. No supported path does not mean no trade.
Data lineage and build journey