Design a carbon-aware scheduling policy for {{workload_type}} on {{cluster_desc}}.
Include:
- Grid carbon intensity inputs ({{region_codes}}) + renewable forecasts
- GPU tactics: mixed-precision, quantization, sparsity, gradient checkpointing
- Job classes: deferrable vs latency-critical
- SLA-safe shifting windows and priority queue rules
- Emissions accounting: Scope 2 location vs market-based
Provide a pseudocode scheduler and test scenarios.
