Round 17, 2025

To remain on the Squiggle platform, a model must:

1. Perform competently over the long-term.

2. Not take human opinion as an input (including betting odds).

Model authors may update their models mid-season. However, they must not attempt to compensate for their model's deficiency with a one-off adjustment for unusual circumstances, as this would constitute taking human opinion as an input. For example, if a team sacks their coach, authors cannot consequently adjust one particular tip because they now think it looks too low or high, unless they ensure this same modifier will be equally applied to all future sackings, not only this one.

Usually errors in tips supplied to Squiggle cannot be corrected after the start of the relevant match. However, I will manually correct a tip when:

1. The error was the fault of the Squiggle Bot (e.g. bot failed to run and include an updated tip), not the model nor the model's supplied data; or

2. The error was obvious, i.e. would be immediately apparent as an error to a reasonable observor.