Module hash 

Exact and normalized content hashes plus the dataset fingerprint.

Implemented by plan 02-03. Exact and normalized content hashes plus the dataset fingerprint.

Two hashes per row, for two different jobs (D-17): the exact SHA-256 over the raw input bytes is identity and provenance; the normalized hash (nfc-trim-ws-v1 — NFC, trimmed, internal whitespace collapsed, deliberately NO casefolding) is leakage detection.

This module is a LEAF

It depends on nothing else in this crate except the error type. In particular it does not know that typed split roles exist: both fingerprints take raw parts ([SplitFingerprintInput], [DatasetFingerprintInput]) rather than a typed split, so the hashing story is complete and testable on its own. The one place the typestate and the hashes meet is the prepared-dataset constructor, which assembles these inputs from its own splits.

One construction, two domain tags

A dataset fingerprint and a single-split fingerprint absorb split parts through the same private helper. They differ only in their domain-tag prefix, which is what makes them differ even for a one-split dataset — so a split fingerprint can never be mistaken for a dataset fingerprint by a consumer comparing hex strings.

Structs

DatasetFingerprint

Identity of a WHOLE dataset.

SplitFingerprint

Identity of ONE split alone.

Constants

CONTENT_NORMALIZATION_VERSION

The version tag of the normalization pipeline behind normalized_hash.

Functions

exact_hash

SHA-256 over the raw bytes of input. Identity and provenance.

hex

Lowercase hex rendering of a digest.

normalized_hash

SHA-256 over the nfc-trim-ws-v1 normalization of input. Leakage detection.