One page per real task. Each recipe states the problem, gives a copy-paste solution in Python and CLI where relevant, shows expected output, and links to the concepts the recipe rests on.
Deterministic identity over a streamed HF split.
Row-order-independent hashing for tabular data.
Layer-wise diff and interpretation.
Reproducible diff across environments.
Render an artifact as an HTML report.
Chaining ns subcommands in Bash and PowerShell.
Regression tests against a pinned reference artifact.
Small, real plugins you can copy.
Reading and writing .nsz artifacts against S3, GCS, and Azure.
Chunked, memory-bounded recording.
True streaming with deterministic replay.
Running many experiments in parallel with the same seed contract.