How it works?

What a distance here is

One transport distance between two countries, in kilometres, made of three legs: the origin country road leg, from the origin country's capital to its relevant port; the sea leg, between the two ports; and the destination country road leg, from that port to the destination's capital. On a narrow screen the same three are labelled 1st road leg, sea leg and 2nd road leg: the shorter name for the same leg, never a different one. The total is those three added up. The sea leg on its own is not a transport distance.

A relevant port

The port each country actually ships through for that particular counterpart, chosen in the source data rather than by this app. Most countries have one, so their road leg is the same whoever they trade with. Six of them (Canada, Russia, the United States, Azerbaijan, Kazakhstan and Turkmenistan) have two, and their road leg changes with the other end of the pair. That is why the same country can show two different road legs on two rows.

Why some cells are empty

A cell reads n.a., not available, for one of two reasons, and never because a number was too hard to find. Either the source has no road distance from that country's capital to a port (true for 53 countries, mostly small islands and a few states whose road network data was missing when the database was built), in which case the three legs cannot be added up into a total; or the source has no row for the pair at all.

Nothing is estimated, interpolated or averaged to fill a gap. An empty cell is a fact about the data, which is why it is grey text and not a red error.

Where only the road leg is missing, a single lookup leads with at least the legs that do exist, added up. That is a floor under the real distance, not a guess at it: every kilometre in it is a leg the source states, and the true total is higher by the leg that is missing.

A sea distance of zero is real

Two landlocked neighbours can route through the same port. Austria and Hungary both go through Slovenia, and then the sea leg is genuinely 0 km while the two road legs carry the whole distance. Zero is a value, not a missing number.

Road distance is secondary

The road figure is capital to capital and exists only between landlocked countries, and only where the source found the road route worth recording: 1,562 of the 51,302 pairs it covers, so the Road (km) column is empty on most rows. It is shown as a separate, clearly-labelled column and is never substituted for the sea route.

On 214 pairs the source records a road route shorter than the sea route's two road legs added together, so the sea route is not the one anyone would take. Those rows say Road shorter than sea route in their own Source flag cell, in the table, in the export and under the road row of a single lookup. Nothing is substituted and nothing is estimated: the three legs are still the sea route's.

A name the app cannot read

Country names are matched to ISO3 codes against a list of official names, ISO codes and common aliases. A spelling the list does not hold is neverguessed at: the row stays in the table, the cell keeps the spelling your file used, and the Origin ISO3 or Destination ISO3 column says Not matched.

You can place it yourself instead of editing the spreadsheet. The control above the table names every spelling it could not read; choose a country once and every row that used that spelling gets it. Where a close spelling exists it is offered as a suggestion, and nothing is applied until you choose it. Your file is not changed and the table still shows what the file said: a placed name shows up as the code beside it.

Direction

Origin → destination is looked up as written. The source stores both directions and they differ slightly, so the app never flips a pair around to find a number.

What the AI does, and what it does not

• It reads a client's text or PDF and says which places ship to which, quoting the sentence each lane came from. A lane whose sentence is not in the text is thrown away.
• It suggests the country of a place it reads. Nothing is calculated until you accept, and you can change any reading first.
• It remembers the names you accepted, for your account alone, so the next document from the same client goes faster.
• It never writes a distance. Every distance, every leg and every n.a. comes from the CERDI dataset, through the same code as before.

Source

Bertoli, S., Goujon, M. & Santoni, O. (2016). The CERDI-seadistance database. Études et Documents n°7, CERDI.

The database ships inside this app, so looking up a distance sends your data nowhere. A project you save is stored in your account, sent over an encrypted connection and visible only to you, and text you ask the AI to read goes to it as the part above says.