One idea underneath everything here: independent instruments, built on different machinery, measuring overlapping reality, checked against each other. Where they stop agreeing, one of them is wrong — that is how finding F-001 was made, and it is the standard every number on this site is held to.
Every series carries a grade: observed — own-recorded live from terrestrial AIS (aisstream.io receiver network); the recording, timestamps and event derivation are ours · reported — counted and published by a named body (for the ledger: the US Census Bureau) · modelled — derived by an algorithm, never quotable as a measurement. Grades are never mixed silently. Each series also carries a clock (fast = days, texture not conclusions · medium = months, needs corroboration · slow = years, safe to build on). A series that cannot currently be trusted is withdrawn: its page stays up stating why, and its data is not served — a correction is worth publishing.
The port and country pages are built from US Census port-level trade data — port of unlading, verified (landlocked ports report zero vessel weight). Totals use only the grand-total row per port-month; country breakdowns use only detail rows, because Census ships overlapping country groupings (APEC, OECD, Pacific Rim) in the same response that would overcount roughly threefold if summed. Census revises prior months and its own revision stamp is broken (it returns 0 on every row we hold), so this mirror detects restatements by comparison and archives the prior value before adopting the new one. Country codes are Census Schedule C, verified against the official list. Full method: TT-008 · port ledger · TT-005 · customs mirror.
A continuous recorder holds the live feed of the aisstream.io terrestrial receiver network across the 8 US container gateways; a deterministic detector turns positions into berth, anchor and departure events. Physics, not paperwork: mooring is detected from position spread, not the crew-set status flag; tugs and ferries are excluded by hull characteristics after both fooled earlier filters. Known limits are published with the data — coverage gaps, left-censoring, and the fact that AIS type codes cannot distinguish a container ship from a bulker (terminal geofences can). Full method: TT-001 · gateway watch.
Corrections are published with the size of the error, and retraction is a public, reasoned status — never deletion. Recent examples live in the method documents above and on findings; the standing rule is that findings which flatter the publisher's freight business get more scrutiny before publication, not less.
Method page last revised: 2026-08-06. Changes to any method are dated on the instrument documents themselves.