archive-v1 — the MIT SportsDataverse archive, 2021-2025

The archive this project runs on, published so a stranger can run it too.

28 assets, 549,177,654 bytes (~0.55 GB), seasons 2021-2025. No account, no API
key, no token, no permission required from anyone.

git clone https://github.com/vyhlidal/cfb-poll && cd cfb-poll
make .venv
make archive        # cfbpoll archive sync --source sportsdataverse --verify
make rankings       # a poll, from raw data, on your machine

make archive reads
data/manifests/sportsdataverse.lock.json,
downloads every asset below, and checks its sha256 and byte count before any
consumer reads it. A mismatch is a hard failure, not a warning. Downloads land
on <name>.part and are renamed only once the digest matches, so an interrupted
sync is resumable and any file that is there is a file that was checked.

What is in here

Series	Files	What it is
play_by_play_{2021..2025}.parquet	5	play-by-play, ~0.52 GB of the total
cfb_schedules_{2021..2025}.parquet	5	schedules and final scores (cfb_schedules_*, not schedules_* — they are different series)
cfb_ratings_weekly_{2021..2025}.csv	5	third-party ratings, benchmarks only, never model inputs
cfb_schedule_crosswalk_*, cfb_teams_crosswalk_*, cfb_rosters_crosswalk.parquet, cfb_crosswalk_in_data_repo.csv	12	ESPN ↔ CFBD id mapping
LICENSE-MIT-sportsdataverse.txt	1	the upstream notice, unmodified

Licence

These assets are a republication of MIT-licensed work by
SportsDataverse, whose
notice ships with them as LICENSE-MIT-sportsdataverse.txt and is reproduced here
as MIT requires:

MIT License. Copyright (c) 2023 SportsDataverse.

MIT License. Copyright (c) 2021 cfbfastR authors.

Files are byte-identical to what upstream served; the upstream_url and
fetched_at of every one of them is in the lockfile. Two things stated rather
than glossed: MIT covers SportsDataverse's compilation work and is not an upstream
rights transfer, and the underlying scores, dates, sites and opponents are facts.
The full position, including why that matters, is in
LICENSE-DATA.md.

What is deliberately NOT here: anything from CollegeFootballData.com. CFBD's
terms prohibit operating "a raw feed, public database mirror, proxy, substitute
API", so no raw CFBD response is republished by this project, ever. This project
depends on CFBD and credits it; that dependency is a reason to
support them, not to mirror them.
One consequence is honest and worth knowing before you run anything: our 2021 and
2022 postseason comes from a private CFBD backfill, so a fork's hindsight surface
for those two seasons stops at conference championship weekend. Every run records
which archives it actually read in _run.json, so that difference shows up as a
number rather than as a mystery.

Project code is MIT. Published ratings are CC BY 4.0.