NeuronScope
Interactive diagramBetasince v0.7.0

Fingerprint flow

Trace a Python value through canonicalization, chunked hashing, and namespace tagging to a stable 32-byte digest.

Audience: All usersRead: 4 minEdit on GitHub

Each stage below runs once per fingerprint computation. Click a node to open its source module, or focus it with the keyboard to inspect it inline.

The pipeline

Interactive pipelinehover · focus · click
InputPython objectStageCanonicalizerStageChunk schedulerCoreBLAKE3 digestStageNamespace tagOutputfp:*
Input
Python object

A model, dataset, prompt, or environment reference. Any object that implements the fingerprintable protocol.

Solid edges are the primary path; the canonicalizer and chunk scheduler execute concurrently for tensor-heavy inputs.

Interpretation notes

  • The canonicalizer is the only stage that varies by namespace. Everything after it is namespace-agnostic.
  • The chunk scheduler exists solely so a fingerprint can be computed against inputs that exceed available memory.
  • The namespace tag is what lets a reader validate a digest without knowing which object produced it.
Related