Method.

Two kinds of thing.

Source-locked. Geography, coastlines and river courses by date, named structures, dates, merges and administrative units. Each resolves to a node in an evidence graph of 340 nodes and 189 edges, with the source, the page and a confidence. A structure appears only with period evidence recorded there. Nothing modern is backdated.

Illustrative. Textures, counts of houses and boats and people, the form of unnamed buildings, the pattern of bare ground on the Island. These exist for visual completeness. They are marked illustrative in the graph and in the record, never on the picture.

How the world is built.

Code owns placement. Every instance in the world is placed by script from a scene specification written from the sources; no scene is generated as an image. Generated pixels supply appearance only: a texture, a normal map, a single object turned into a mesh. The fort's geometry is authored in Blender from the period plans, not generated. Every asset carries its generation receipt and its input hashes.

Work moves in phases. A coordinator writes a brief with numeric targets and a forbidden list; an implementer builds and reports with every measurement; the coordinator re-measures everything independently, scores the result against a fixed visual target on composition, lighting, materials and detail, and against the specification on historical conformance, and writes the next brief from the gaps. Every judgment and every ruling is in a public decision log. Nothing is approved by passing a check; I approve nothing myself.

Where the money goes.

Every paid generation call is declared before it runs and logged with its inputs and their hashes, so any asset can be traced to the call that made it. The ledger shows what the work has taken so far in sessions, hours, tokens, renders and disk, and a rough cost at list prices.

Primary sources, in short.

• H. D. Love, Vestiges of Old Madras, three volumes, 1913, on archive.org; the spine of the c.1640 to c.1700 slice.
• Period plans and descriptions quoted in Love: Fryer, Lockyer, Salmon, Pitt's Prospect.
• Greater Chennai Corporation pages, the Acts and orders for names and units, and the Madras Musings heritage columns for later periods.
• OpenStreetMap and SRTM for the modern registration plate, used as modern evidence only.

The full list, generated from the evidence graph with every node, date, source and confidence, is on its own page. It is long on purpose.

What I have not done.

• No dated historical map has yet been aligned to the modern plate; the c.1700 layout follows Love's distances and the Fort's plans.
• No external historian has reviewed the world. Corrections are welcome and are logged.
• Later eras, c.1915, c.1980 and 2026, are researched and not yet built.