The tables under http://167.71.161.149/data/tables/ are rebuilt every hour; a number read there today can differ from the same query tomorrow. A release is a copy of that directory frozen on a date, so a paper or a notebook can cite a version that will not change.
http://167.71.161.149/data/tables/releases/<date>/<file>, with <file> any path of the live directory: market_daily.parquet, board_entry/source=vercel/year=2026/part.parquet, board_entry_latest.csv.gz, tables.json, overview.json, and so on. The list of releases is http://167.71.161.149/data/tables/releases/index.json (date, file count, bytes, schema version, the generated stamp of the tables it froze).
Each release directory holds a MANIFEST.json: for every file its path, bytes and sha256; the schema_version and generated of the tables.json it copied; the row count of every table; and not_frozen, the files of the live directory that are not in the copy.
Not frozen: routerflow.duckdb. It is a single file of the same rows the Parquet files hold, and copying it every month would double the size of each release; open the Parquet files instead, or the live DuckDB file for an interactive session. Also left out: _build_state.json and the exporter's lock and temporary files, which describe the build and not the data.
Monthly, and additive: a release, once listed in index.json, is not changed or removed. When the tables' schema changes, the release that follows carries the new schema_version in its manifest; a query written against an older release keeps working against that release. The schema's own history is in CHANGELOG-DATA.md.
Name the release date and the file, and check the file's hash against the manifest when it matters:
routerflow.art, tables release 2026-09-22, board_entry (schema 1.2.1),
http://167.71.161.149/data/tables/releases/2026-09-22/, MANIFEST.json sha256 per file.
Reading a frozen board_entry with DuckDB is the same as reading the live one, with the release path in place of the base URL:
import duckdb, json, urllib.request
base = 'http://167.71.161.149/data/tables/releases/2026-09-22'
files = [base + '/' + f['path'] for f in json.load(urllib.request.urlopen(base + '/board_entry_files.json'))['files']]
con = duckdb.connect(); con.execute('INSTALL httpfs; LOAD httpfs')
con.read_parquet(files).create_view('board_entry')
Verifying a downloaded file:
curl -sO http://167.71.161.149/data/tables/releases/2026-09-22/market_daily.parquet
curl -s http://167.71.161.149/data/tables/releases/2026-09-22/MANIFEST.json | python3 -c 'import json,sys; print([f["sha256"] for f in json.load(sys.stdin)["files"] if f["path"]=="market_daily.parquet"][0])'
sha256sum market_daily.parquet
routerflow's tables are CC BY 4.0. Every row keeps its upstream license: source.license and source.citation give each source's license and the citation line it asks for, and board.license the license of a board where it differs from its source. Rows are published only from sources whose redistribution is yes or attribution; when you publish a number from a release, publish the citation line of its source with it.
ops/release.sh [YYYY-MM-DD] [--force] on the collection server, from the repository root. It waits for a running build to finish, copies every file of data/tables/ except the ones listed above into data/tables/releases/<date>/, verifies the copy against the hashes it took, writes MANIFEST.json, and rewrites releases/index.json from the manifests on disk. A second run on the same date verifies the frozen copy and, if the tables have not changed since, does nothing; if they have (an hourly rebuild ran in between), the release stands and the script exits 1 unless --force replaces it. ROUTERFLOW_TABLES_DIR overrides the tables path. The private directory data/tables-private/ is never read.
| date | files | bytes | schema | tables generated | notes |
|---|---|---|---|---|---|
| 2026-09-22 | 99 | 70,327,382 bytes (70.3 MB) | 1.2.1 | 2026-09-22T17:28Z | First release. The 21 classic tables plus the leaderboard tables of schema 1.2 (public board_entry, 12 sources, 177 boards; source, board, board_metric, board_coverage, name_map; entity and alias tables). |