# Duplicate audio hashing

## Problem

Matching filenames do not prove matching content. Two files can be different DJ
edits, masters, encodes, or entirely different tracks. Removing either without
stronger evidence is unsafe.

## Design

Serato Doctor computes SHA-256 only for the duplicate group currently being
reviewed. This avoids hashing an entire 22,000-track library during every scan.
The review queue labels a group as byte-for-byte identical or warns that its
files differ. File size and a short fingerprint remain available as supporting
detail.

The final batch dry-run recomputes every approved comparison server-side and
includes the status beside each keeper decision. The hash is evidence, not an
automatic repair decision; the DJ remains in control.

## Edge cases

- Identical audio stored under different filenames is recognized.
- Metadata changes inside an audio container produce a different byte hash and
  therefore a conservative warning.
- Missing or unreadable files fail verification rather than being treated as
  identical.
- Hashing is streamed in chunks rather than loading whole tracks into memory.

## Tests

- Equal byte content produces identical SHA-256 fingerprints.
- Different content is labeled different.
- Batch previews carry the server-computed confidence status.