cascade-fullctx-spectral-v1

A differentiated, NumPy-only full-context generator derived from the public custom-fullctx-v4 design.

Design

• Emits only 4096-point series, matching the 128-patch training/eval geometry and minimizing the one non-target context patch paid per series.
• Samples trend as total end-to-end excursion, so trend strength is independent of series length.
• Mixes twelve families: trend/seasonality, structural regimes, multiplicative series, AR/integrated/nonlinear dynamics, smooth spectral GP, power-law long memory, intermittent demand, and outliers.
• Replaces the predecessor's slow 48-pass random-Fourier GP with one batched inverse FFT and adds persistent/anti-persistent spectral paths.
• Generates lazy 1024-row random-family chunks, keeping every stream prefix mixed while amortizing Python dispatch.

Validate

python -m cascade.miner.cli verify ./cascade-v2 --chain-toml chain.toml

Contract validity and CPU throughput do not establish forecasting quality. Run a production-faithful GPU A/B score against the current king before deploying this candidate.