Methodology

Three layers, three rhythms

The map (Made here, the trade lanes, the Buyer Book) is computed from the world trade record, which moves once a year, HMRC's UK record, which moves every month, and the public award records of 33 countries. What moved is the weekly feed of developments read from primary sources, UK public-sector supply today. The brief joins the two for one product family, and its tender and award lines come from a record that moves every working day. Nothing on the site claims to move faster than its source.

How a gap is chosen

A line enters a country's list only when the country imports it at scale, does not already export it well, and sits close to what it does export well. Closeness is the product-space method (Hidalgo, Klinger, Barabasi and Hausmann, Science 2007) on every bilateral trade flow: two products are close when the countries that export one well tend to export the other well. Each line is scored out of 8: the import bill 0 to 3, proximity 0 to 3, supply concentration 0 to 2, every point printed beside it. Public buyers are context and never a point. Ores, fuels, precious metals, raw farm output, scrap, one-off capital goods and defence are out of scope, and entrepots get no page.

How a lane is built

A lane is one exporter to one importer, published when the lane carried at least USD 1bn in the latest year and at least fifty in-scope lines of USD 1m or more. It lists the largest lines, the fastest growing, the lines the exporter dominates, and the lines another country ships at a lower unit value. A unit value is shown only when weight is declared, three or more sources ship the line with a weight, and the figure sits within a factor of five of their median.

How a public buyer is counted

Award notices are joined to customs headings through crosswalks BuzzBaboon wrote, because no official one exists. Only notices the buyer classified as goods count, a join needs a specific code rather than a division, and identical rows are counted once. The free pages show counts; no award value is displayed and no supplier is named on a public page.

The weekly developments

Ingest, normalise, deduplicate, resolve entities, resolve geography, classify, check source quality, write the record, match to companies, deliver. Every stage is separable, and every stage can drop a candidate.

Ingest once, match many times

A company check never crawls the internet, never runs a web search and never spends a model call to answer a question that has already been answered. Developments are ingested once into a central store; a company profile is built once and cached against its canonical domain; matching is a cheap retrieval against that store.

Fact and interpretation are different fields

A record stores what happened in one field and what it may mean in another. They are rendered differently and labelled differently: the facts are quoted, the interpretation is marked as interpretation, and every link is the publisher's own. BuzzBaboon never rewrites a source into a summary and then discards the source.

Relevance is not confidence

Relevance, the Buzz Score, measures how closely a development matches your capabilities, sectors, markets and commercial triggers. It is deterministic set intersection over a controlled vocabulary, weighted 40/30/18/12 across capabilities, sectors, geography and trigger. It is not a model, and every score can be expanded to show which items matched.

Confidence measures how much BuzzBaboon trusts the underlying facts. It is reported as high, medium or low rather than a percentage, because a decimal here would imply precision that does not exist.

The labels, and how each is derived

Priority is Act now, Worth investigating, Watch or Low relevance. It is derived from the Buzz Score band and the opportunity window: a high score with an open window is Act now; a high score with a later window, or a good score with an open window, is Worth investigating; anything above the threshold is Watch. A single-source, unconfirmed development is held at Watch whatever its score, until a primary source appears.

Potential impact (High, Medium, Low) is the score band: how much of what the development needs is something you do. Timing (Now, Emerging, Longer-term) is the opportunity window in plain words. Evidence (Strong, Developing, Early) is source confidence in plain words. Every label shows the sentence that explains it.

The threshold is published

A company check shows a development only when it scores 45 per cent or higher against the profile. Below that, BuzzBaboon says nothing was strong enough rather than padding the page. When you name a goal, only developments that touch that goal are considered at all, and each still needs 45 per cent or higher: the goal narrows the question, it never lowers the bar.

What BuzzBaboon will not do

It will not invent investment amounts, company names, dates, locations, people, attendance figures, procurement opportunities, statistics or source URLs. It will not tell you that an opportunity will generate business. Where evidence is insufficient it says so rather than filling the gap. Where a source link does not exist, it says the link does not exist. Where a statement is inference, it is labelled as inference.

Limits

Coverage is selective and may be delayed or incomplete. Automated interpretation may be wrong. Sources change, disappear and contain errors. BuzzBaboon is a discovery service that tells you where to look, and it is not a substitute for verification or professional advice.