AIS: vessel reports in and out, with the sentinels handled properly

An AIS message arrives as one or more NMEA 0183 sentences carrying an armoured binary payload. Position reports, base station reports, static and voyage data, and aid to navigation reports each have their own fixed bit layout.

The layouts are fully allocated. There is no spare field and no vendor extension block, which has consequences for what can be written back out.

AIS spells not available as an in-band value inside a field that otherwise carries a real measurement, and every one of those values is a plausible-looking number.

Maritime and subsea infrastructure protection begins with knowing which vessels have declared themselves, where, and with what stated intent.

The interesting cases are the disagreements: a vessel whose declared draught is impossible for its declared cargo, or whose reported behaviour does not match its declared voyage. Those comparisons are only possible if the unavailable values were not silently turned into zeros first.

Because AIS is self-reported, a decision layer must be able to distinguish what a vessel said, what it did not say, and what nobody knows. Those are three different facts.

Ingest turns one message, delivered as one or more sentences, into an Entity and an Event.

Every unavailable sentinel becomes an absent canonical field rather than a value: the latitude one degree past the pole, the longitude past the antimeridian, the speed value the not-available convention uses, the bearing outside the valid range, the nine-bit heading with every bit set, the rate of turn at the signed floor, and the zero draught, which is the dangerous one because it is also a plausible reading. A forwarder that nulls the other six can still state that a laden tanker draws nothing.

Which fields the source itself marked unavailable is recorded on the object, because a vessel saying it does not know its heading and an adapter having nothing to say are different facts and only one of them belongs in the data.

A position at zero latitude and zero longitude is a real position and is translated as one. A speed at the top of the reportable range means that speed or higher, so it is kept as a floored measurement with the floor recorded.

The one-bit position accuracy flag is not written into the metres-valued accuracy field. A threshold is not a measurement, so the flag is parked and the accuracy field stays unknown.

Time comes from three places depending on what the message actually says. A base station report carries a full date and time and is used directly. Position reports carry a second of the minute, and the instant chosen is the one bearing that second nearest the reception time. Static and voyage data carry no time at all, so the reception instant is used and the object says exactly that. Where the feed provides a receiver timestamp in the sentence tag block it is read; otherwise the injected clock is used, never the system wall clock.

Egress emits sentences. An Entity emits the message type it arrived as, and a Track emits one position report per sample in the track's own order.