[manifold-baseline-curated](https://huggingface.co/datasets/jdpressman/manifold-baseline-curated-v0) is a dataset I've made to make it easier to tune a model on Manifold questions without training on difficult to resolve, personal, or meta/joke/luck markets. There is currently substantial interest in LLM based forecasting bots and this dataset should help people get started if they want to try tuning local models. However another purpose I have in making it is to help provide absolute standards to measure forecasting bots against instead of the anti-inductive relative standard of actual in practice market performance (since even a slightly better bot will eat the alpha that a slightly worse bot would have).

Towards that end I would like to see people start to publish the performance that their bot gets on the validation and test split of manifold-baseline-curated. The question then is simple: Will at least three Manifold bot teams advertise their performance on manifold-baseline-curated or a successor dataset[0] before the end of 2024?

Question resolves YES if three separate bot teams advertise or otherwise disclose their performance on the validation and test sets of manifold-baseline-curated or a successor dataset. If three separate bot teams don't exist by the end of 2024 the question resolves NO, if three don't publish it resolves NO.

I will not be betting in this market. Instead I will effect it by making manifold-baseline-curated better in response to community criticism.

[0]: A dataset is a successor dataset if it is called "manifold-baseline-curated", published by myself, and intended to supercede use of the previous iteration of manifold-baseline-curated.

[1]: The teams do not need to publish their performance on the same iteration of manifold-baseline-curated if multiple iterations exist.