# topics/tartaria First topic. Chosen deliberately (2026-09-06), against the rule that a chosen topic collapses the pipeline into that topic. The guard: nothing under `sources/` or `derived/` may know the word "tartaria" except as a parameter; the same connectors run on the control topics. ``` sources// raw bytes as fetched, content-addressed; NEVER edited sources/manifest.jsonl one line per fetch: id (sha256), url, connector, fetched_at, fetched_by, via_proxy, bytes, path — the POSITION of the noticing beliefs/-.md frozen, dated weights per claim, per origin; never edited after freeze — a new file is a new freeze claims/ extracted statements, each pointing at a source id + quote span derived/ vectors, projections, instrument outputs; reproducible from sources/ + code hash ``` A fetch is a noticing from a position. The manifest records the position. Nothing here is frame-free. ## What is in here (2026-09-07) Sources are append-only; this list is a guide, `sources/manifest.jsonl` is the record. - `wikipedia/` 10 pages (core + controls) · `web/` 5 arc-of-interest pieces · `x/` 4 search runs - `openalex/` 2 queries · `ia/` ~25 period texts (Tartary 1655-1706, fair reports 1892-93, Children's Aid Society 1854, 1871 fires, fortification 1636-73, incubator 1906) - `ca/` 36 newspaper pages (orphan trains, incubators, Peshtigo, Chicago fire, Jackson Park, Tartary) - `commons/` ~2,200 images: Maps of Tartary, Old maps of Asia, Maps of Siberia, the 1893 / 1904 / 1915 / 1889 expositions, star forts (241 subcategories), neonatal incubators, article images - `locpic/` ~1,100 Library of Congress photographs: the 1893 fair, Detroit Publishing Co. court houses / city halls / high schools / post offices / capitols / libraries / stations, 1871 fire, orphan asylums, incubators, fortifications Every image has a vision vector in the same space as the text (see pipeline/embed_images.py).