One identifier for each ship
Each vessel is given a single permanent identifier the first time we see it. Every renewal record, every document and every movement hangs off that identifier rather than off a name or a spreadsheet row.
It is issued once and never changes. Not when the ship is renamed, reflagged, sold, rebuilt or re-registered, and not when we discover that something we recorded about it was wrong. It is never reused for another vessel, and it is never recycled after two records turn out to be the same ship.
A vessel with no paperwork at all is issued one by exactly the same route as a new tanker with a full set of certificates. There is no separate path for difficult cases, because a separate path for difficult cases is the path that never gets properly exercised.
The identifier carries no meaning, deliberately
It would be tempting to build something readable into it, so that a person could tell the ship's age or where it was built by looking. We decided against it, and the reason is worth stating.
An identifier that encodes a fact inherits that fact's problems. If the fact is not recorded, the identifier cannot be issued. If two sources record the fact differently, one ship ends up with two identifiers. And if the fact is later corrected, the identifier itself has to change, which means every document already sent out carries an identifier that no longer matches.
So the identifier is meaningless by design, and the facts live beside it. Build year, builder, yard, hull number and country are all recorded, each one carrying which source said it and when. That is more useful there than it ever was inside an identifier, because it is what tells us whether two records are the same ship.
Deciding which ship a record belongs to
This is the harder half, and it is now a separate mechanism. When a record arrives, whether from a spreadsheet, a certificate or a data feed, something has to decide whether it belongs to a vessel we already hold.
Only three things are allowed to make that decision, in order of strength.
-
The IMO number, where there is one and its check digit is valid
-
An inland, small craft or national registry number
-
Our own record of the building yard, together with the hull number
The IMO number settles it on its own. The others do not. Yard and hull number in particular are not reliably unique, so a match on them needs a second piece of near permanent evidence, such as the delivery year or the principal dimensions, before it counts.
A ship's name, its flag, its call sign and its MMSI number are never used to decide identity. All four change during a working life, most of them more than once, and a system that treats any of them as identity will quietly merge two different ships. They remain fully searchable, along with every former name we hold. They are simply not allowed to decide.
A record that does not add up is referred, not guessed
There are three possible outcomes and no fourth.
-
Matched to a vessel we already hold
-
Recognised as a new vessel and issued an identifier
-
Referred to a person, and held until they decide
There is no probable match that quietly becomes a match. In particular, a record that agrees with one of our vessels on one point and disagrees on another is always referred. An IMO number that matches while the hull number does not is the most useful signal we get, and swallowing it in either direction is how a book acquires duplicates that nobody finds for years.
A referral says which two claims disagree and which source made each of them, so the person deciding can do so without reopening the original file.
Where two records do turn out to be one ship, one identifier survives and the other keeps resolving to it permanently, because it is printed on documents already sent and filed. Nothing is deleted and no record is rewritten.
Checked against real ships before it was settled
None of this was settled on reasoning alone. Every assumption about what can be known about a ship was checked against a large reference database of vessel particulars covering well over a hundred thousand vessels.
That check changed the design. The construction details an identifier might have encoded turned out to be missing for a substantial minority of ships, and missing most often for the newest ones, which are exactly the vessels joining a book on delivery. Building yards proved not to resolve reliably by name in either direction. And the same source recorded the country of build under two different conventions in the same column.
It also confirmed the reason for having our own identifier at all. That database has itself had to invent a private key for the ships in it with no IMO number, which is a few per cent of the total.