the invite

Four ways in. Pick the one that fits.

  1. 1. An opinion

    You haveA view about what a season should be worth
    What you doMove a setting on this page
    What happensThe same season ranks itself again, beside the published one
  2. 2. An argument

    You haveA paragraph
    What you doSend it to challenge@thepoll.ai
    What happensOne of two answers back: why not, or it gets built with your name on the credit
  3. 3. A setting worth sending

    You haveSettings you moved and a reason
    What you doSend your settings to challenge@thepoll.ai
    What happensYour settings arrive attached to your argument
  4. 4. A working entry

    You haveA file, and an AI to write it
    What you doOpen a pull request
    What happensScored on the identical games, with the scorecard posted beside this poll's

Thirty seconds to an afternoon. Three of the four need no account and no code. Only the pull request needs a GitHub login.

Send what it gets wrong to challenge@thepoll.ai. An AI reads it first. A person decides. You get one of two answers. Why not, argued from the same numbers, and a concession on anything you turn out to be right about. Or it gets built, run in public, with your name on the credit.

Challenges and answers

no entries yet · this list is kept by hand

Nobody has sent a finished run yet. The first goes up here with the name on it, whichever way the answer went.

Build your own.

Point the AI you already have at github.com/vyhlidal/cfb-poll. It sets up Python, pulls the data and runs your version against this one.

What your AI is actually going to runfor anyone who wants to look first

Point your AI at it.

git clone https://github.com/vyhlidal/cfb-poll && cd cfb-poll make .venv make archive SEASONS=2023

Answer the questions.

# your version of the poll, as a settings file cp configs/challengers/beta-w-4.toml configs/challengers/<yourname>.toml # or write the ranking yourself, in about forty lines # rate(games, plays, through_week, config=None, state=None) -> dict[str, float]

Put it next to this one.

make challenge CHALLENGE_ENTRY=configs/challengers/<yourname>.toml

That archive line is a real download, about half a gigabyte. One season is enough to start.

Open a pull request with your file in it and the same machine runs your version on the identical games against the same set of rating systems, then posts your scorecard beside this poll’s. It needs no key and no account of yours, which is why this works at all.

Before you start

This poll’s constants were set on 2021 through 2023 and scored once against 2025 without being tuned on it. Yours may learn from any season that finished before the one it is ranking, and never from that one.

The bar you would be beating, and every line where this poll fails it →

What this could get to.

Calling the playoff field as well as the polls do, or better, by the end of this season. That is worth taking to the people who run the conferences.