Method

The source hierarchy

Primary accounts, logs and manuscripts anchor the record. Scholarly editions supply transmission history and interpretation — how a text reached us, and how reliable that path was. Reference works can orient a route but never outrank a primary or scholarly record.

Every published stop resolves to at least one entry in the central register. That is not an aspiration: an unresolved source id fails the build. Each register entry carries a note saying what the source is good for and where it is weak, because a teaching edition of excerpts and a digitised manuscript are not the same evidence.

What the certainty labels mean

Certainty is a claim about evidence, not a confidence mood. It is spelled out in words beside every route — colour and line pattern alone are not accessible — and it is drawn as a distinct line style on the map.

• Documented — directly supported by the cited record.
• Probable — strong but incomplete support; the kind of claim a careful historian makes with a qualifier.
• Approximate — a broad location, a month rather than a day, or a connection we interpolated.
• Disputed — scholarship materially disagrees, and the page says so rather than picking a winner.
• Traditional — a received literary or cultural placement, repeated for centuries, not established by evidence.
• Fictional reconstruction — an illustrative narrative map. Never a historical claim.

Coordinates cannot outrun their evidence

A coordinate with four decimal places asserts a position to about ten metres. Almost nothing in this atlas is known to ten metres, so the schema caps decimal places by certainty: four for documented, three for probable, two for approximate and disputed, one for traditional and fictional reconstruction.

Give an approximate stop four decimals and the build stops and names the rule. The fix is always to lower the coordinate, never to raise the certainty. Stop coordinates are modern reference points — a harbour area, an island, an archipelago — and each carries a note saying so, because a modern harbour centroid is not a claim about where a ship tied up.

The line between two stops

Stops and dates can be well documented while the water between them is not. Route segments therefore carry their own certainty, a basis, and an interpolation method, and any segment that is not documented needs an editorial note explaining what it is standing in for.

Ocean legs use a great-circle approximation; coastal legs and overland legs use scholarly waypoints. None of them is a daily track, and no overland line is a modern road. A dashed line means we do not know the path — not that the ship wandered.

Dates, calendars and narrative order

A date here keeps five things apart: how it is displayed, its calendar, its precision, its certainty, and the sources that support it. Ranges stay ranges and approximate years keep their "c.". Magellan, Columbus and Polo are recorded in the Julian calendar they were written in, and are not silently converted.

The Odyssey uses book and episode order and carries no calendar position at all. Assigning BCE dates to a poem would be inventing evidence, so the schema forbids a narrative date from holding an ISO value. A dated journey must also run forward in time; a stop out of sequence fails the build.

Quotations

A direct quotation needs an edition, a translator where relevant, an exact locator, a rights status, and an attribution naming who is speaking. No locator, no quotation — the build enforces it.

Because that bar is real, this edition carries only two verified verbatim quotations, both Darwin's, both checked word for word against the linked transcription. Everywhere else the excerpt is labelled an editorial paraphrase or a piece of context and is rendered without quotation marks. Where the original design had quotation marks around invented logbook prose, that prose is gone.

Distance and difficulty

Distances are not stored anywhere in the content. They are computed at build time from the curated geometry, adjusted by a declared detour factor for the interpolation method — a coastal approximation is longer than the straight line it is drawn as — and then rounded to a precision the reconstruction can actually carry: to the nearest 500 km above 10,000, the nearest 100 km above 1,000, the nearest 10 km below that. You will see "about 60,000 km" and never "60,417.32 km", because the second number would be a fiction dressed as arithmetic.

Difficulty is a five-axis editorial comparison, labelled as editorial wherever it appears. It is a reading of the record, not a measurement of it, and the fictional journey is scored on narrative adversity rather than pretending to a physical one.

The gate

All of the above is executable. The rules live in the build's schema, and a violation stops the build rather than raising a warning nobody reads — an unsourced stop, a quotation without a locator, a coordinate too precise for its certainty, an uncited date, a route segment missing its basis, a narrative date carrying an ISO value, a public-domain claim with no stated reason for it.

Automated checks catch the mechanical failures only. They cannot tell whether a source was read correctly, whether the framing is fair, or whether a claim deserves the label it was given. That still needs a person, and it is where corrections do their work.