About this experiment

Normalized offshore fields such as IMO, name, type, flag, build year, class, dimensions, tonnage, power, bollard pull, DP, FiFi, deck space, crane capacity, and accommodation.

Where available, records also show former names, class-notation capabilities, a vessel photo, and source links.

AI helps find source material, extract fields, compare conflicts, and prepare records for review.

AI output is not a source. The public site shows reviewed source categories, not model confidence.

• Public site: Next.js App Router, React Server Components, TypeScript, Tailwind CSS, Node.js, npm, and Turbopack.
• Public data: static JSON snapshots for the vessel index, per-vessel records, site statistics, source labels, and structured data.
• PDF pipeline: Python tools for document download, text extraction, field extraction, source audit, and public-safe export.
• Database build: PostgreSQL-backed internal review tables, canonical vessel records, conflict queues, normalization scripts, and export checks before publication.
• AI support: OpenAI Codex, Claude Code, and Llama-family models for coding, extraction, comparison, and review workflows.
• Operations: Railway hosting, optional Sentry monitoring, optional Cloudflare Web Analytics, ESLint, Vitest, smoke tests, and browser-based responsive QA.

Every published value shows its source. People review public registers, owner/operator materials, class records, maritime sources and commercial reference databases (not individually named) before publication.

Identity and class are reviewed against public registers including IACS, classification-society registers, and national registries where available.

Owner-published documents add review depth. The reviewed owner-document corpus is 4,400+. Public source links appear on 8,275 vessel records.

The database is incomplete and uneven. Vessel names change, documents age, labels vary, and sources disagree.

We leave fields blank when no source supports a value.

Source labels identify the source category behind a field. They do not mean endorsement, approval, or operational certification. The Sources cited list writes origins out in full where a public URL exists.

Each value is tagged with its source tier: primary source, independent source, reference database or derived. The tiers and rules are set out in the next section.

Source categories and tiers also appear in vessel-page structured data for search engines and AI systems.

Every value on a vessel page carries a tag that says what kind of source it comes from. Sources are ranked in four tiers. The owner, the shipyard and the class register come first, and a lower tier is not meant to override a higher one.

1. 1. Primary source

   The vessel's owner or operator (its own documents and website), the shipyard or designer, the classification society's register, and IACS. A published copy of one of these documents counts as the document itself.

   A primary source is enough on its own, and a lower tier is not meant to override it.

2. 2. Independent source

   Named third-party publishers: vessel catalogues, P&I club registers, trade press and other third-party documents. Each one is named in the vessel's sources, with a link where one exists.

   Used where no primary source states the value. A lower tier is not meant to override it.

3. 3. Reference database

   Commercial reference databases. Where a database can be linked it is named; others are not individually named. Each database counts as its own source, but agreement among reference databases alone does not make a value trusted.

   Fills gaps. A primary or independent source is meant to replace a reference-database value; until a disagreement is settled, the reference value can stay.

4. 4. Derived

   Values worked out from other data, for example a capability read from the class notation or a figure taken from descriptive text.

   Fills gaps only. Any named source replaces a derived value.

• Higher tier wins. When sources disagree, the value from the higher tier is meant to be published. Not every disagreement has been settled yet; until it is, the earlier value can stay on the page.
• The class register settles technical facts. For the dimensions, tonnage, class notation and build year it states, a class-register or IACS value closes a disagreement.
• Same tier, different values: a person decides. Values are never averaged or blended.
• A document must state the value clearly. An owner or shipyard figure is used when the document names the quantity (for example “Deadweight 5000 t”) and the value passes basic checks. Unclear readings go to review first.
• A web page must be about the right vessel. Vessel names are reused. A value from an owner's web page counts as primary once the page is confirmed to describe this hull, for example because it prints the IMO number.
• Broker and charter listings are not primary sources, even when they look like a spec sheet.

A value is marked corroborated when two or more sources agree on it. Agreement is counted per source family: several documents from one publisher count once. Each commercial reference database counts as its own source, so two of them agreeing earns the mark, but that agreement alone does not make a value trusted.

Corroborated means the sources agree. It does not prove that they checked the vessel separately: some publishers may draw on the same upstream data. A primary source still outranks a corroborated value from lower tiers.

Sometimes a page shows a value that differs from the figure on the owner's website or spec sheet. This usually means one of these:

• The disagreement is not settled yet. When a source disagrees with a published value, the published value stays until the disagreement is checked. Nothing is replaced automatically on an unclear reading.
• The figures use a different basis. For example design versus maximum draught, deadweight in tonnes versus long tons, or brake horsepower versus kilowatts.
• The owner's page is not yet matched to the vessel. A page that does not show the IMO number waits until other details confirm it describes the same hull.
• We have it wrong. The rules say the owner wins; if the page shows otherwise, please tell us and we will correct it.

Grades and internal confidence scores are not published. The tag and the listed sources are what the record rests on.

vesselspecification.com is the public surface for approved, public-safe outputs from the offshore data group.

Commercial workflows such as brokerage, charter, and fleet operations live in separate paid products. This site stays free, public, and indexable.

VesselSpecification is currently operated as an experimental public vessel-reference project. A formal operating entity is intended before paid services, API access, bulk data products, accounts, or broader commercial launch.

Send the vessel name or IMO, the field, the value you expect and a link to your source (an owner page, spec sheet or class record) to [email protected]. A person checks every correction against these rules. We reply with the sources behind the value, the rule that applies and what we changed.

Vessel data may be incomplete or outdated and must not be used for operational decisions. Legal, privacy, copyright, database-rights, and takedown notices should be sent to [email protected].

Verify with owners, operators, class, flag, or other primary sources before making decisions.