Democracy's Library by place: what metadata alone can place (sampled 14/10/2026)
DL scope: 10,520,879 items. Sample: up to 400 random items per pool, weighted by pool size.
| pool | items | sampled | city | county | state/prov. | court district | coverage/subject only | none |
|---|---|---|---|---|---|---|---|---|
| USGovernmentDocuments | 9,441,916 | 400 | 2% | 1% | 2% | 92% | 0% | 4% |
| fedlink | 393,841 | 400 | 22% | 2% | 15% | 0% | 8% | 54% |
| nasa | 208,897 | 400 | 8% | 0% | 7% | 0% | 46% | 40% |
| governmentpublications | 104,550 | 400 | 0% | 0% | 23% | 0% | 12% | 65% |
| us_census | 24,320 | 400 | 61% | 1% | 36% | 0% | 0% | 2% |
| IGS (instituteofgovernmentalstudieslibrary) | 9,369 | 400 | 72% | 6% | 6% | 0% | 0% | 16% |
| Alberta | 21,275 | 400 | 0% | 0% | 64% | 0% | 2% | 34% |
| LAC (lacbac / library_and_archives_canada) | 852 | 400 | 0% | 0% | 11% | 0% | 37% | 52% |
| democracys-library (direct) | 25,424 | 400 | 14% | 1% | 5% | 0% | 4% | 76% |
| sim_microfilm (Government Documents) | 114,078 | 400 | 0% | 0% | 1% | 0% | 0% | 99% |
Placeable from metadata alone (city, county, state or court district): about 91% of DL, weighted by pool. The rest (~992,185 items) would need page text: at 2 pages each with claude-haiku-5-5 in batch (measured $1.24 per 1,000 pages), about $2,461. Not run: needs Daniel's go.
Without the federal court dockets (USGovernmentDocuments, 9.4 M items placed only by their court district): of the other 902,606 items, about 15% name a city or county in their metadata and 29% at least a state or province. City matches need the state named too (Census 2023 place names), so they lean to precision over recall.