HRF Roadmap: ML fee prediction engine

[Bortlesboat]

Why this matters

Fee selection is a practical financial-freedom problem: users with limited resources should not have to overpay because they lack good mempool context. The HRF proposal commits to fee forecasts that help users decide when to send and how much they can save by waiting.

Scope

Build a CPU-friendly fee prediction engine for Satoshi API with 1hr, 6hr, and 24hr forecast windows.

Acceptance criteria

• Add forecast endpoint(s) for 1hr, 6hr, and 24hr fee conditions
• Include confidence intervals or confidence bands in responses
• Document the model inputs and retraining/update path
• Include accuracy metrics, starting with mean absolute error on 6hr forecasts
• Keep runtime practical for self-hosted nodes and low-cost hardware
• Add tests and example responses

HRF proposal alignment

This maps to the submitted deliverable: ML fee prediction engine with calibrated accuracy metrics, running without GPU requirements.