Field notes

Updates from the archive

Research notes and updates, written when there is something worth reporting rather than on a schedule — so the gaps are real. No newsletter; everything lives here.

Earlier posts

Scouting 3,373 Maps: Building a Discovery Pipeline for Vietnam Cartography

We built a multi-source discovery pipeline that pulls Vietnam-related map records from BnF Gallica, Humazur, David Rumsey, and the Library of Congress into a single reviewable grid. 3,373 candidates surfaced — 758 high-confidence — now curatable from /admin/scout with thumbnails, scoring, and one-click bulk-ingest into the catalog.

New Collection: 63 Tonkin Topographic Sheets (1903–1927)

A batch of 63 colonial-era topographic sheets covering the Red River delta — Hà Nội, Hải Phòng, Nam Định, Thanh Hóa, and surrounding provinces — has been tiled to R2 and added to the catalog. All from the Service Géographique de l'Indochine, surveyed between 1903 and 1927.

Platform Notes: Universal IIIF, Gemini OCR, and the MapShell Refactor

Self-hosted tiles on R2 for better performance, a Gemini-powered OCR pipeline for toponym discovery, and a core architecture refactor around MapShell. April's updates focus on scaling our data foundation and unifying the user experience.

Platform Notes: Retiring a Pipeline and Rebuilding the Data Foundation

The L7014 pipeline is done and its code is gone. In its place: a modular maps data layer with proper source provenance, multi-IIIF support, and metadata backfilled for all 37 maps in the collection. What we built, why we built it that way, and what it reveals about running a research archive as software.

Buildings as Ground Control: A New Method for Vectorizing Colonial Maps

The 1882 and 1898 Saigon maps share hundreds of the same buildings. We use those stable structures as automatic ground control points — letting maps georeference each other, and producing a 1880–1900 building dataset as a byproduct.

March 2026 Update: Maps, Methods, and What's Next

Featured in Saigoneer. Automated 500+ L7014 maps. Now planning the knowledge graph, 3D pipeline, and community tier system. Here's where things stand.

Dissecting Space: The Six-Layer Method Behind the Archive

How we treat a historical city as a data stack — translating the modern world's six-layer spatial model into a framework that works with 19th-century maps, panoramic paintings, and archival photos.

How We Automate Georeferencing of 500 Historical Maps

A plain-language explanation of the GCP propagation algorithm that turns a 500-sheet military map series into an automatically georeferenced archive.

We're in Saigoneer — and What It Means

Saigoneer covered us in January 2026. Here's the project vision in plain language — recovering the body and soul of a city lost to wartime imagery.