Fingerprint frozen fitted state, including Python parameters and buffers

Computes SHA-256 of R state and deterministic content snapshots of nested Python objects. Python snapshots include estimator preprocessing, training context, neural parameters, buffers and evaluation flags. Snapshots use numerical content rather than shared-object identities or pickle bytes. Autograd ownership, TabICL checkpoint paths and temporary inference workspaces are excluded; frozen context caches and estimator inference settings remain included. Fingerprints from earlier package versions must be regenerated. Unknown opaque state is rejected, rather than hashed as a pointer placeholder.

Usage

singlesample_state_fingerprint(model)

Arguments

model

A fitted model or single-sample deployment object.

Value

One SHA-256 string. Equality tests detect state changes, not validity of the fitted model or clinical performance.