# Native Sparse Attention Audit

## Description
Audits Native Sparse Attention (NSA) compression ratios, block-sparse top-k selection, and memory bandwidth reduction.

## Category
`architecture`

## Tags
`nsa`, `sparse-attention`, `native-sparse`, `compression`, `bandwidth`

## When to Use
Audit memory bandwidth and FLOP savings in sparse attention serving engines.

## Operational Limitations
Evaluates block-sparse index selection math. Does not train sparse kernels.

## Verification Script
Run the deterministic verification checker:
```bash
python3 scripts/check.py examples/synthetic.json
```
Exit codes:
- `0`: Check passed and thresholds satisfied.
- `1`: Threshold breach or operational violation detected.
- `2`: Schema error or invalid usage.
