Priority list

[markwhiting]

Technical bits:

• persistent paper storage on S3 Save gpt reponse assigned to paper in DB #55
• move from mongo to document DB
• accounting for multiple truth_types of ratings on a paper-feature (e.g., default mechanism, ground truth)
• metric for each feature type¹ (for now R^2 for numerical and unbiased F1 for categorical, LLM concept comparison for strings)
• black box model wrapper (this is mostly unknown right now)

Functional goals:

• basic projects (with URL, persistent features and papers) Set up the Projects concept #79
• basic feature control (add and remove features from projects) (related to Set up the Projects concept #79)
• quality functions (scores and some kind of secondary robustness model) relates to Set up feature Validation #80

Don't need right now:

1. dev environment

Footnotes

1. we should have a simple way to deal with type from the feature source so that it doesn't need to be redefined everywhere. ↩