Loading chart...
The pieces here are not a themed collection. Each measures a coordinate the trade record contains and the discipline has no routine number for. Proximity in the product space is a minimum of conditional probabilities and structurally cannot go negative, so an entire half of the co-occurrence structure is unmapped. A trade balance is the divergence of a bilateral flow field, so the divergence-free part of every imbalance is invisible to every bilateral statistic on this site. Concentration indices are cross-sectional and carry no time domain. Entry probability on the development ladder is Markov in density, so it has no term for having made the thing before.
Research answers trade questions with trade methods. Replications reproduces named papers and reports whether they hold. This section does neither, and it is the only place on TradeWeave where a null result is publishable as the finding.
Proximity is a minimum of two conditional probabilities, so it cannot take a negative value and the repulsion half of the product space has never been drawn. Of 190 exporting economies in 2024, none holds a revealed advantage in both crude petroleum and women’s blouses, against 13.2 expected by chance. In 1995 eight did.
Curveball fixed-fixed co-occurrence nullCommunity ecology
A trade balance is the divergence of a flow field, so a bilateral imbalance is two orthogonal things: a pressure difference between two countries, and a circulating current that would still run if every balance were forced to zero.
Helmholtz-Hodge decomposition at the Onsager conductanceNon-equilibrium thermodynamics
This site measures supply concentration four ways and none of them has a clock. Six suppliers who all ship in March score identically to six spread round the year.
Circular mean resultant length and harmonic phaseDirectional statistics
The noise floor of the panel every other page computes on. The largest apparent shocks in the HS6 record are the classification’s own revision boundaries, and they collapse on aggregation.
Interpolated backoff codebook, surprisal in bitsInformation theory
Every complexity tool here treats entry probability as a function of where a country stands now, with no term for what it used to make. At matched product-space density, history is worth a multiple.
Density-matched relearning contrastMemory science
This section is allowed to be wrong in public. It pays for that licence with disclosure, and the terms are the same on every page.
Every page carries the single strongest objection against it, stated at full strength in the objector’s own words, followed by either the answer and the run that produced it or the sentence This is not answered. An unanswered objection does not block publication. Hiding one does.
Where two defensible specifications disagree on a magnitude, both are published and the page quotes the range, never the more favourable end of it.
Every figure exposes the query that produced it and downloads its own rows as CSV. A number no reader can regenerate does not ship.
Each page names its own kill condition in the first screen, before any result. An experiment is retired when the run it named as decisive comes back against it, when prior literature is found that already published the headline construction, or when a data-vintage change moves a headline number outside its published sensitivity band. Retirement means the route stays live, a dated banner states what killed it, the page moves to the archive above, and the reason is written into the codebase ledger. Nothing is silently deleted, because a section that quietly removes its failures is not running experiments.