# TT-006 · Arrival reliability — method

**Grade: OBSERVED on both sides.** We heard the claim and we saw the outcome. No published
reliability index can say that; they compare a carrier's schedule against a carrier's report.

## What it measures
The gap between a vessel's **own declared ETA** and when it **actually came alongside**, at the
eight US gateways TT-001 watches.

## What it is NOT
**This is the ship's declaration, not the carrier's booking-page schedule.** AIS static messages
carry a destination and an ETA typed in on the bridge; a carrier schedule is a commercial promise
made to a customer. Both are claims. They are different claims, and this instrument measures the
first. Reporting it as "carrier schedule reliability" would be false.

## How a declaration is matched to an arrival
A declaration counts only when all of these hold:
1. It was broadcast **before** the berthing, and no more than **21 days** before.
2. Its free-text destination **names the gateway the ship then entered** — "USLGB", "US LGB",
   "LONG BEACH" and "LOS ANGELES" all resolve to the same place.
3. It leads the arrival by at least **1 hour**. Below that it is a status note, not a forecast.
4. The error is within **14 days**. Beyond that the declaration is about a different voyage.

Filters 3 and 4 exist because of a real case: `CAPE STARR` declared an ETA of 2027-01-03 and
berthed two minutes later. That is not an arrival 3,849 hours early — it is a ship already
alongside broadcasting the ETA of its next leg. Without those filters the instrument reported a
fleet arriving 90 hours early.

## Operator attribution
AIS carries no operator field, so carriers are inferred from vessel naming conventions, held in
`carriers.json`. Graded: **high** for an exclusive fleet prefix (MAERSK, MSC, ONE, OOCL, EVER);
**medium** for a class convention others occasionally share (the "… EXPRESS" pattern used for
Hapag-Lloyd). A vessel that matches nothing is left unattributed rather than guessed at.

This does double duty. A ship named "MAERSK …" is certainly a container ship, which is a sharper
test than the 150 m length filter — AIS ship-type codes 70–79 cannot separate a container ship
from a bulker, and the unattributed merchant vessels in our archive are largely bulkers.

## The limit that matters today
The AIS ETA is **crew-entered and frequently stale**. That is not a defect of the instrument; it
is the finding. But it means a small sample is dominated by whoever happened to update their
declaration, so this series is withheld until the sample is large enough to be about the fleet
rather than about a handful of diligent bridges.

## Cadence
Hourly. Declarations and berthings both arrive continuously; the matcher is idempotent.