This article summarizes a completed deep-learning project on American put option pricing, built as a reproducible research workflow and fully documented in the public GitHub repository linked at the end. The question driving it wasn't whether a neural network can fit option prices — of course it can — but whether deep learning adds anything a competent numerical method doesn't already give you.

In a fair value world, RIM is no longer plug-and-play. When OCI reshapes book value without flowing through earnings, residual income can mislead. Valuing banks now demands more than formulas—it requires aligning capital, returns, and economic substance to avoid accounting distortions masquerading as insight.