2025.11.21

• log constant optimal performance values to compare against
• sweep on new scaled FFT
• make T be sampled from $U(T_{\min },T_{\max })$ (128 → 1024)
• plot the analytical optimal terminal riskless for $T_{\min }$ and $T_{\max }$, like how it is already for 256
• sort the values into buckets based on episode T then log terminal riskless in buckets
• to log the Sharpe ratio, probably would have to log std then just expression with terminal riskless / std