docs(admin): hash conflicts page — the duplicate-content queue

Sixth admin page. Documents what creates conflict groups (scanner at
scan time + the hash backfill retroactively), the group card anatomy
(count badge, library + short SHA-256, per-copy rows with usage summary
highlighted when that copy holds reading data), and both resolutions
with verified semantics: Keep this copy (reparents the other copies'
children into the kept copy, deletes their catalog rows, records the
resolver — files on disk never touched, consistent with platform-wide
delete semantics) and Keep both (mark resolved, stop flagging, nothing
merged). Glossary gains Keep both and Keep this copy.

docs/admin/hash-conflicts.md

@@ -0,0 +1,56 @@
+<!-- Ground truth: templates/admin_hash_conflicts.templ (cards, actions,
+     confirms), internal/handlers/hash_conflicts.go (ResolveHashConflict:
+     keep_all marks resolved; keep reparents children to the kept copy,
+     deletes the other catalog rows, records who resolved), internal/
+     services/media_scanner.go (recordHashConflictIfAny), internal/services/
+     hash_backfill.go (backfill-created conflicts). Re-verify against these
+     when the page changes. -->
+
+# Hash Conflicts
+
+Hash Conflicts is the duplicate-content queue: groups of catalog items
+whose files are **byte-identical** (same SHA-256) but stored at more than
+one path in the same library. The scanner flags these as it goes — and
+the hash backfill job flags them retroactively for content that predates
+hash tracking — so both fresh scans and old libraries surface their
+duplicates here.
+
+Duplicates usually mean the same book was copied into two folders, or a
+library was reorganized and both the old and new copies got scanned. The
+harm is quiet but real: split reading histories, double-counted library
+stats, and two entries for one book.
+
+## What a conflict looks like
+
+One card per group: a **Duplicate content** heading, a badge with the
+copy count, the library and a short fingerprint of the shared hash, and
+one row per copy — title, author, full path, file size, and a usage
+summary that's **highlighted when that copy holds reading data** (someone
+has actually read it).
+
+## Resolving a group
+
+Two outcomes, both confirm-guarded:
+
+### Keep this copy
+
+Per-copy button. Keeps the copy you chose and folds the **other copies'
+reading data into it** (their catalog rows are removed; the underlying
+files on disk are never touched — this platform's deletes act on the
+catalog, not your files). The row-level usage summaries exist precisely
+so you can pick the copy that already carries the reading history.
+
+### Keep both
+
+The group-level button, for duplicates that are intentional. It records
+the resolution and stops flagging this group — every copy stays in the
+catalog exactly as it is, nothing merged, nothing removed.
+
+After either choice the card is replaced with the outcome, and the group
+never comes back (unless the same duplicate situation re-arises from a
+*new* scan, which creates a fresh group).
+
+## When the page is empty
+
+"No content duplicates detected." — the queue drains from the top; groups
+appear as scans and the backfill find them.

docs/developer/ui-controls.md

@@ -284,6 +284,14 @@ label: User Access
 where: Admin → Libraries → a library's panel → section heading
 does: Per-user visibility checkboxes — checked (default) shows the library to that user; unchecked hides it everywhere for them, enforced in the database.
 
+label: Keep both
+where: Admin → Hash Conflicts → a conflict group card
+does: Records the resolution and stops flagging the group; every copy stays, nothing merged or removed. For intentional duplicates.
+
+label: Keep this copy
+where: Admin → Hash Conflicts → a copy row
+does: Keeps that copy, folds the other copies' reading data into it, and removes their catalog rows. Files on disk are never touched.
+
 label: Resume rewind
 where: audio player → player settings; profile settings
 does: Rewinds playback a configured few seconds when you resume.