SELFDOUBT: Uncertainty Quantification for Reasoning LLMs via the Hedge-to-Verify Ratio

Build an open-source library that implements the Hedge-to-Verify ratio for black-box LLM APIs. This would allow developers to quantify reasoning uncertainty without direct access to logprobs.

Suggested repo: hedgeverify

"Quantify reasoning confidence for any closed-source model."

Estimated effort: 30h