jgrusewski
f7cf02f363
plan: OFI momentum + dense micro-reward — 9 tasks, full gradient flow
9-task implementation plan covering:
- fxcache target_dim 4→6 (raw_open + mid_price_open from MBP-10)
- OFI data pipeline fix (ofi_gpu wiring, indices 42→66, PORTFOLIO_STRIDE 30→38)
- Dense micro-reward (OFI momentum × price confirm × adaptive cost tolerance)
- OFI embed MLP (16→8 cuBLAS) with full backward gradient flow
- Mamba2 history enrichment (SH2→SH2+8) + d_h_history backward
- Attention enrichment (D→D+8) + d_input_scratch exposure
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>