How We Verify Data

This page exists because a reference site is only worth the method behind it. Anyone can publish a container dimensions table; what makes ours usable for load planning is that every number has a traceable origin, a date and a route to being corrected.

Every figure against the documents, with how sure it is. The outside of a box is set by ISO 668, which we read in the free official texts that adopt it. Everything else — the inside, the door, the tare, the volume — is the maker’s, so each figure is checked against the published sheets of lines, lessors and builders: confirmed when more than half of the independent sheets that print the figure, and at least two, print it within ±5 mm, ±2 % for a mass or ±1 % for a volume, a range when no figure has such a majority, one sheet when only one exists, and named as such. A sheet that copies another counts once. A sheet that prints several series agrees with a figure when at least half of its series do, so a sheet of two ratings agrees with either of them; it counts once. The maximum gross mass is compared exactly, because it is the rating on the plate. The site’s figure must be within the tolerance of the consensus, the printed figure with the most sheets behind it, or the build stops; where two figures have as many sheets each, the site keeps its own when it is one of them, and calls it a range. Every figure, its status and each document’s reading are on where every container figure comes from; the mark beside a value in a spec table opens the same.

Where that stands today. The site publishes 191 container figures: 44 set by the standard, 114 confirmed by more than half of the sheets that print them, 21 a range where no figure has such a majority and 12 on a single sheet. Behind them are 88 specification sheets from 30 lines, lessors and makers; in all we have read 308 documents — standards, laws, notices, registers and sheets — and written 5,250 readings out of them, each with its place in the document. These counts are taken from the evidence base when the site is built, so they move when a document is added.

Owner prefixes. Each of the 85 prefixes we curate was looked up on its page in the BIC register, which names the company that registered it and its country; 85 of 85 are there. 79 are registered to the owner we name, and 6 to a company of the same group: CBHU to Florens, the COSCO group’s lessor; NYKU to NYK Line and KKFU, KKTU and KLFU to K Line, two of the lines that founded ONE; EITU to Gaining Enterprise, of the Evergreen group. Where the company publishes its own codes — a fleet list, an annual report, a leasing sheet — we read that too: 15 prefixes are confirmed that way. A prefix we recognise from the register alone, with no page of its own, is labelled as such in the tracking tool.

What never counts as a source. Wikipedia, Reddit, forum posts, marketplace listings and content farms are excluded, however plausible the figure looks. A source enters our citations registry only if a figure on this site actually rests on it — the registry is a record of what we used, not a reading list.

Tested, not assumed. Claims about behaviour get tested rather than inferred. The carrier trackers in our tracker study were opened in a real browser with a real container number to see what each one accepts; where we have not run the test, the result reads "not tested" instead of a guess. The same rule governs deep links: a link format is published only after it opened a tracking result.

Dates on the data, not just the page. Every prefix record carries the date it was verified, and the registry as a whole carries a verification date shown on each page that uses it. A page’s "updated" stamp changes when its content changes — we never bulk-refresh dates to look fresh, because a false freshness signal is a form of inaccuracy.

One dataset, rendered everywhere. Every table, diagram, calculator and downloadable file reads from the same machine-readable dataset, which is why a correction lands on every page at once and why the tables cannot silently drift from the data. Running text is written by hand, and the figures in it that matter most — search snippets, prices, pallet counts, the front page — are compared with the dataset by automated tests on every build; a sentence that falls out of step fails the build. That dataset is public: container specifications and carrier and prefix data, free to reuse with attribution.

Corrections are logged, not quietly patched. When we get something wrong we fix it and record it. Send the page URL, the statement you believe is wrong and the document that supports the correction to data@contsize.com; the correction policy is on the about page and every correction made is in the correction log.

What we have not read. We did not buy the current ISO texts. ISO 668 was read in IS 6566:2001 and ГОСТ Р 53350-2009, the 1995 text with its amendments; ISO 1496-1 in the 1990 edition the US Code of Federal Regulations incorporates; ISO 6346 in IS 6928:2001 and BIC’s 2022 code table. ISO 668:2020, ISO 1496-1:2013 and ISO 6346:2022 we have not read, and no page says it did. A figure no document prints — the cost of a conversion, a rule of thumb from a yard — is not given a status, and no mark beside it says it was checked.

What we cannot verify. Prices move weekly and vary by market, so price figures carry a date and a range and should be treated as orientation, not quotes. Tare weights vary by manufacturer and production year: when this site disagrees with the weights marked on the doors of your specific container, the doors win. Saying so is part of the method.