How to read

How this site works

Yearline organises history by year, from AD 1 to 2026. A record appears only when an exact source passage supports it, and it carries a note on its source and on what that passage can and cannot establish. The collection is researched and reviewed by AI, with the method and its limits shown.

A year page

A year page lists the records that a source places in that year. Each one names its source, states its limits and links to the quoted passage. Below them, clearly separated, is any period material whose date range merely overlaps the year. A year page is never a complete account of the year.

Kinds of record

Placed in a year: A source places this in one year. It is listed under that year; no day precision is invented.
What a document says: The wording of a document, which is not proof of what happened in practice.
Approximate range: A date range given by the source. It is not an event in every year of the range.
No year assigned: Something a source reports, with no single year established.

Dates

Placed in a year. The year under which a source enters the record: an annal heading, a regnal or consular year, the date on a document. Where a source’s calendar had to be converted, the record says how. For Chinese annals the year comes from the lunar month the entry stands under; the cyclical day is quoted as the source gives it and is not always checked against calendar tables. A record’s date reference is the project’s reading and may name an era in a normalised form; the dating notes say what the source prints.
Approximate range. A range from the source, such as a museum catalogue’s date. It does not mean something happened in every year of the range.
No year assigned. The source’s date wording is kept and no year is guessed, for example where two chronicles place the same event in different years.

Regions

A region says where the source places the act itself. An official’s seat, an author’s home, a destination or the place an object was found is not enough. “No region assigned” means this collection has not fixed the place; it does not mean the place is unknown.

Sources and translations

Each record quotes a key passage of the source text it relies on. For catalogue and database rows the quote is often a single field, and the statement summarises the rest of the row. Original-language text, published translations and AI working translations are labelled separately. Where a record was read in a published translation, in an editor’s published English abstract (calendared records) or in an editor’s English rendering, that is what is quoted; such records are flagged, and the original-language text was not the basis of the record. An AI working translation is not official or authoritative. A law’s wording is not evidence of how it worked in practice. A catalogue entry is the catalogue’s description. Limits are the research record’s own wording, so they sometimes speak in working terms. “The retained file”, “supplied text”, “permitted” or “admissible” text and “packet” mean the copy of the source that was read and checked; “physical line” or “row” and line numbers refer to that copy; “selected” marks the one event taken from a longer entry; “region is null” means no region is assigned; “this revision” means the current version of the record; a “pinned” revision or edition is the exact version that was read; “fixed” or “native” rows and fields are the database’s own values; a “candidate” is an event that was considered and not taken. Some limits are short notes on the source and add no caution of their own. In papyri, inscriptions and legal texts, letters in round brackets resolve an abbreviation: they are the editor’s reading, not what is written.

Checks

Each statement is tied to an exact source version and passed an independent AI evidence review before it was accepted. Before this release, a random sample of the records was audited again against the retained sources:

Two audits were run on samples of the public records. The first round checked 168 records: 156 drawn at random and 12 chosen because earlier audits had faulted an earlier version. The second round checked 156 drawn at random. Each record was checked again against the retained source text by a reviewer that had not seen the research.
No record was rejected by the auditors in either round (some were withdrawn afterwards, where a needed correction could not be made). In the first round 15 of 156 randomly drawn records needed a correction or an added limit; after corrections, 6 of 156 did in the second round. The audit’s own pass mark, no such finding at all, was not reached.
The findings were wrong credits, links and notes, turns of translation, and a few dates stated more firmly than the sources allow. None was an invented event. Each record was corrected and reviewed again, or withdrawn, and each cause was searched for across the whole collection.
At the second round’s rate, a few records in every hundred are likely to carry a flaw of the same kind. A sample audit estimates; it does not clear every record.

AI agreement is not proof of historical truth, and a sample audit is not a guarantee for every record. If a source shows something is wrong, prepare a correction.

The name

Yearline is this project’s name. It has not been legally cleared as a trademark.