test(anime): synchronize stale Jimaku open test

- Hide the standalone browser during the Jimaku handoff
- Skip stale or windowless playback flows safely

.agents/plugins/marketplace.json

@@ -1,20 +0,0 @@
-{
-  "name": "subminer-local",
-  "interface": {
-    "displayName": "SubMiner Local"
-  },
-  "plugins": [
-    {
-      "name": "subminer-workflow",
-      "source": {
-        "source": "local",
-        "path": "./plugins/subminer-workflow"
-      },
-      "policy": {
-        "installation": "AVAILABLE",
-        "authentication": "ON_INSTALL"
-      },
-      "category": "Productivity"
-    }
-  ]
-}

.agents/skills/subminer-change-verification/SKILL.md

@@ -1,22 +1,45 @@
 ---
-name: 'subminer-change-verification'
-description: 'Compatibility shim. Canonical SubMiner change verification workflow now lives in the repo-local subminer-workflow plugin.'
+name: subminer-change-verification
+description: Verify SubMiner changes with repo-native cheap-first test lanes. Use after code, config, launcher, plugin, runtime, stats, documentation, or workflow changes; do not use for read-only questions.
 ---
 
-# Compatibility Shim
+# SubMiner Change Verification
 
-Canonical source:
+Verify the behavior claimed by a change without running unrelated expensive checks by default.
 
-- `plugins/subminer-workflow/skills/subminer-change-verification/SKILL.md`
+## Workflow
 
-Canonical helper scripts:
+1. Inspect the requested scope and changed paths with `git status --short` and `git diff`.
+2. Read `docs/workflow/verification.md` as the source of truth for maintained lanes.
+3. Run the cheapest lane or lanes that cover the changed behavior.
+4. Escalate to the full handoff gate only for substantial or cross-boundary changes.
+5. Report exact commands, results, skipped checks, blockers, and remaining risk.
 
-- `plugins/subminer-workflow/skills/subminer-change-verification/scripts/classify_subminer_diff.sh`
-- `plugins/subminer-workflow/skills/subminer-change-verification/scripts/verify_subminer_change.sh`
+Do not use hidden wrapper commands. Verification commands are owned by `package.json` and the workflow documentation.
 
-When this shim is invoked:
+## Lane Selection
 
-1. Read the canonical plugin-owned skill.
-2. Follow the plugin-owned skill as the source of truth.
-3. Use the wrapper scripts in this shim directory only for compatibility with existing commands and docs.
-4. Do not duplicate workflow changes here; update the plugin-owned skill and scripts instead.
+- Internal docs, `AGENTS.md`, or `.agents/skills/**`: `bun run test:docs:kb`
+- User-facing `docs-site/**`: `bun run docs:test`, then `bun run docs:build`
+- Config/schema/defaults: `bun run test:config`
+  - If defaults or templates changed, also run `bun run generate:config-example` and `bun run verify:config-example`.
+- General TypeScript source: `bun run typecheck`, then `bun run test:fast`
+- Launcher or mpv plugin: `bun run test:launcher` or `bun run test:env`, based on the behavior changed
+- Runtime compatibility or dist-sensitive wiring: `bun run test:runtime:compat`
+- Stats dashboard: `bun run test:stats`
+- Build/release scripts: `bun run test:scripts`
+
+For substantial changes, use the full gate documented in `AGENTS.md` and `docs/workflow/verification.md`.
+
+## Runtime Escalation
+
+Real runtime checks are required when the claim depends on actual Electron, mpv, overlay, focus, window tracking, launch, or socket behavior. Run the relevant application flow when the environment supports it. Otherwise, report the missing runtime dependency and do not present cheaper checks as authoritative runtime validation.
+
+## Pre-Handoff Checks
+
+Before handoff, reconcile both questions:
+
+1. Do behavior, defaults, flags, shortcuts, ports, APIs, architecture, or workflow changes require documentation updates?
+2. Does the change require a current-outcome fragment under `changes/` according to `changes/README.md`?
+
+Complete required updates before handoff or report the blocker.

.agents/skills/subminer-change-verification/scripts/classify_subminer_diff.sh

@@ -1,13 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
-REPO_ROOT=$(cd "$SCRIPT_DIR/../../../.." && pwd)
-TARGET="$REPO_ROOT/plugins/subminer-workflow/skills/subminer-change-verification/scripts/classify_subminer_diff.sh"
-
-if [[ ! -x "$TARGET" ]]; then
-  echo "Missing canonical script: $TARGET" >&2
-  exit 1
-fi
-
-exec "$TARGET" "$@"

.agents/skills/subminer-change-verification/scripts/verify_subminer_change.sh

@@ -1,13 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
-REPO_ROOT=$(cd "$SCRIPT_DIR/../../../.." && pwd)
-TARGET="$REPO_ROOT/plugins/subminer-workflow/skills/subminer-change-verification/scripts/verify_subminer_change.sh"
-
-if [[ ! -x "$TARGET" ]]; then
-  echo "Missing canonical script: $TARGET" >&2
-  exit 1
-fi
-
-exec "$TARGET" "$@"

.agents/skills/subminer-release/SKILL.md

@@ -0,0 +1,34 @@
+---
+name: subminer-release
+description: Prepare, cut, publish, or repair SubMiner stable and prerelease releases. Use for hands-on release work; do not use for general release questions.
+---
+
+# SubMiner release
+
+Carry out the requested release phase using the repository's current release process.
+
+## Source of truth
+
+Read `docs/RELEASING.md` completely before changing files or release state. Treat it as canonical. Read `changes/README.md` when the work touches change fragments or generated release notes.
+
+Do not copy release commands or policy into this skill. If this skill disagrees with the release guide, follow the guide and reconcile the skill before handoff.
+
+## Workflow
+
+1. Identify whether the request is for a stable release, prerelease, release preparation, publication, or repair.
+2. Inspect the current branch, worktree status, package version, pending change fragments, relevant tags, and latest CI state before making changes.
+3. Follow the matching procedure in `docs/RELEASING.md` in order. Review generated changelog and release-note Markdown before it can be committed or published.
+4. Run every required gate for the requested release phase. Do not treat a cheaper test lane as a substitute for the documented release gate.
+5. Before a stable tag, confirm the package and tag versions match and no pending `changes/*.md` fragments remain. Preserve fragments for prereleases as documented.
+6. Report the resulting version, completed checks, local commit and tag state, remote publication state, skipped platform checks, and any remaining manual work.
+
+## Authorization boundaries
+
+- A request to prepare a release stops before commit, tag, push, or remote publication unless the user also authorizes those actions.
+- A clear request to cut or publish a release includes the documented commit, tag, and push steps. Ask before the first remote mutation when the wording is ambiguous.
+- Do not edit an existing GitHub release, publish to the AUR, change secrets, or alter signing configuration unless the user explicitly requests that operation.
+- Do not switch branches without consent.
+
+## Stop conditions
+
+Stop and report the blocker when required CI or a release gate fails, authentication is missing, versions disagree, required artifacts are absent, or the worktree contains unexpected changes that overlap the release. Do not tag or publish a partially verified release.

.agents/skills/subminer-scrum-master/SKILL.md

@@ -1,18 +0,0 @@
----
-name: 'subminer-scrum-master'
-description: 'Compatibility shim. Canonical SubMiner scrum-master workflow now lives in the repo-local subminer-workflow plugin.'
----
-
-# Compatibility Shim
-
-Canonical source:
-
-- `plugins/subminer-workflow/skills/subminer-scrum-master/SKILL.md`
-
-When this shim is invoked:
-
-1. Read the canonical plugin-owned skill.
-2. Follow the plugin-owned skill as the source of truth.
-3. Do not duplicate workflow changes here; update the plugin-owned skill instead.
-
-This shim exists so existing repo references and prompts keep resolving during the migration to the repo-local plugin workflow.

.github/workflows/docs-pages.yml

@@ -32,9 +32,11 @@ jobs:
       - name: Guard stable docs tag shape
         id: tag_guard
         if: github.ref_type == 'tag'
+        env:
+          TAG_NAME: ${{ github.ref_name }}
         run: |
-          if [[ ! "${{ github.ref_name }}" =~ ^v[0-9]+\.[0-9]+\.[0-9]+$ ]]; then
-            echo "::notice::Skipping non-stable docs tag ${{ github.ref_name }}"
+          if [[ ! "$TAG_NAME" =~ ^v[0-9]+\.[0-9]+\.[0-9]+$ ]]; then
+            echo "::notice::Skipping non-stable docs tag $TAG_NAME"
             echo "stable_tag=false" >> "$GITHUB_OUTPUT"
             exit 0
           fi

.github/workflows/prerelease.yml

@@ -274,7 +274,8 @@ jobs:
             config.example.jsonc \
             plugin/subminer \
             plugin/subminer.conf \
-            assets/themes/subminer.rasi
+            assets/themes/subminer.rasi \
+            assets/thumbnailers/subminer-ffmpegthumbnailer.thumbnailer
 
       - name: Generate checksums
         run: |
@@ -296,15 +297,22 @@ jobs:
         run: echo "VERSION=${GITHUB_REF#refs/tags/}" >> "$GITHUB_OUTPUT"
 
       - name: Verify committed prerelease notes
+        env:
+          RELEASE_VERSION: ${{ steps.version.outputs.VERSION }}
         run: |
           if [ ! -s release/prerelease-notes.md ]; then
             echo "::error::release/prerelease-notes.md is missing or empty. Run 'bun run changelog:prerelease-notes --version <version>' locally and commit the file before tagging."
             exit 1
           fi
+          if ! bun run changelog:check-prerelease-notes --version "$RELEASE_VERSION"; then
+            echo "::error::release/prerelease-notes.md was not generated for $RELEASE_VERSION. Rerun 'bun run changelog:prerelease-notes --version $RELEASE_VERSION' locally, commit, and retag."
+            exit 1
+          fi
 
       - name: Publish Prerelease
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          RELEASE_VERSION: ${{ steps.version.outputs.VERSION }}
         run: |
           set -euo pipefail
 
@@ -326,27 +334,27 @@ jobs:
             exit 1
           fi
 
-          if gh release view "${{ steps.version.outputs.VERSION }}" >/dev/null 2>&1; then
-            gh release edit "${{ steps.version.outputs.VERSION }}" \
+          if gh release view "$RELEASE_VERSION" >/dev/null 2>&1; then
+            gh release edit "$RELEASE_VERSION" \
               --draft \
               --prerelease \
-              --title "${{ steps.version.outputs.VERSION }}" \
+              --title "$RELEASE_VERSION" \
               --notes-file release/prerelease-notes.md
           else
-            gh release create "${{ steps.version.outputs.VERSION }}" \
+            gh release create "$RELEASE_VERSION" \
               --draft \
               --latest=false \
               --prerelease \
-              --title "${{ steps.version.outputs.VERSION }}" \
+              --title "$RELEASE_VERSION" \
               --notes-file release/prerelease-notes.md
           fi
 
           for asset in "${artifacts[@]}"; do
-            gh release upload "${{ steps.version.outputs.VERSION }}" "$asset" --clobber
+            gh release upload "$RELEASE_VERSION" "$asset" --clobber
           done
 
-          gh release edit "${{ steps.version.outputs.VERSION }}" \
+          gh release edit "$RELEASE_VERSION" \
             --draft=false \
             --prerelease \
-            --title "${{ steps.version.outputs.VERSION }}" \
+            --title "$RELEASE_VERSION" \
             --notes-file release/prerelease-notes.md

.github/workflows/release.yml

@@ -273,7 +273,8 @@ jobs:
             config.example.jsonc \
             plugin/subminer \
             plugin/subminer.conf \
-            assets/themes/subminer.rasi
+            assets/themes/subminer.rasi \
+            assets/thumbnailers/subminer-ffmpegthumbnailer.thumbnailer
 
       - name: Generate checksums
         run: |
@@ -295,33 +296,40 @@ jobs:
         run: echo "VERSION=${GITHUB_REF#refs/tags/}" >> "$GITHUB_OUTPUT"
 
       - name: Guard against pending changelog fragments
+        env:
+          RELEASE_VERSION: ${{ steps.version.outputs.VERSION }}
         run: |
           if find changes -maxdepth 1 -name '*.md' -not -name README.md -print -quit | grep -q .; then
-            echo "::error::Pending changelog fragments detected. Run 'bun run changelog:build --version ${{ steps.version.outputs.VERSION }}' locally and commit the polished CHANGELOG.md before tagging. CI no longer auto-builds the changelog because the polish step requires the local 'claude' CLI."
+            echo "::error::Pending changelog fragments detected. Run 'bun run changelog:build --version $RELEASE_VERSION' locally and commit the polished CHANGELOG.md before tagging. CI no longer auto-builds the changelog because the polish step requires the local 'claude' CLI."
             exit 1
           fi
 
       - name: Verify changelog is ready for tagged release
-        run: bun run changelog:check --version "${{ steps.version.outputs.VERSION }}"
+        env:
+          RELEASE_VERSION: ${{ steps.version.outputs.VERSION }}
+        run: bun run changelog:check --version "$RELEASE_VERSION"
 
       - name: Generate release notes from changelog
-        run: bun run changelog:release-notes --version "${{ steps.version.outputs.VERSION }}"
+        env:
+          RELEASE_VERSION: ${{ steps.version.outputs.VERSION }}
+        run: bun run changelog:release-notes --version "$RELEASE_VERSION"
 
       - name: Publish Release
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          RELEASE_VERSION: ${{ steps.version.outputs.VERSION }}
         run: |
           set -euo pipefail
 
-          if gh release view "${{ steps.version.outputs.VERSION }}" >/dev/null 2>&1; then
+          if gh release view "$RELEASE_VERSION" >/dev/null 2>&1; then
             # Do not pass the prerelease flag here; gh defaults to a normal release.
-            gh release edit "${{ steps.version.outputs.VERSION }}" \
+            gh release edit "$RELEASE_VERSION" \
               --draft=false \
-              --title "${{ steps.version.outputs.VERSION }}" \
+              --title "$RELEASE_VERSION" \
               --notes-file release/release-notes.md
           else
-            gh release create "${{ steps.version.outputs.VERSION }}" \
-              --title "${{ steps.version.outputs.VERSION }}" \
+            gh release create "$RELEASE_VERSION" \
+              --title "$RELEASE_VERSION" \
               --notes-file release/release-notes.md
           fi
 
@@ -344,7 +352,7 @@ jobs:
           fi
 
           for asset in "${artifacts[@]}"; do
-            gh release upload "${{ steps.version.outputs.VERSION }}" "$asset" --clobber
+            gh release upload "$RELEASE_VERSION" "$asset" --clobber
           done
 
   aur-publish:
@@ -420,9 +428,10 @@ jobs:
         if: steps.aur_prereqs.outputs.skip != 'true' && steps.aur_ssh.outputs.skip != 'true' && steps.aur_clone.outputs.skip != 'true'
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          RELEASE_VERSION: ${{ steps.version.outputs.VERSION }}
         run: |
           set -euo pipefail
-          version="${{ steps.version.outputs.VERSION }}"
+          version="$RELEASE_VERSION"
           install -dm755 .tmp/aur-release-assets
           gh release download "$version" \
             --dir .tmp/aur-release-assets \
@@ -432,15 +441,17 @@ jobs:
 
       - name: Update AUR packaging metadata
         if: steps.aur_prereqs.outputs.skip != 'true' && steps.aur_ssh.outputs.skip != 'true' && steps.aur_clone.outputs.skip != 'true'
+        env:
+          RELEASE_VERSION: ${{ steps.version.outputs.VERSION }}
         run: |
           set -euo pipefail
-          version_no_v="${{ steps.version.outputs.VERSION }}"
+          version_no_v="$RELEASE_VERSION"
           version_no_v="${version_no_v#v}"
           cp packaging/aur/subminer-bin/PKGBUILD aur-subminer-bin/PKGBUILD
           cp packaging/aur/subminer-bin/.SRCINFO aur-subminer-bin/.SRCINFO
           bash scripts/update-aur-package.sh \
             --pkg-dir aur-subminer-bin \
-            --version "${{ steps.version.outputs.VERSION }}" \
+            --version "$RELEASE_VERSION" \
             --appimage ".tmp/aur-release-assets/SubMiner-${version_no_v}.AppImage" \
             --wrapper ".tmp/aur-release-assets/subminer" \
             --assets ".tmp/aur-release-assets/subminer-assets.tar.gz"
@@ -450,6 +461,7 @@ jobs:
         working-directory: aur-subminer-bin
         env:
           GIT_SSH_COMMAND: ssh -i ~/.ssh/aur -o IdentitiesOnly=yes
+          RELEASE_VERSION: ${{ steps.version.outputs.VERSION }}
         run: |
           set -euo pipefail
           if git diff --quiet -- PKGBUILD .SRCINFO; then
@@ -459,7 +471,7 @@ jobs:
           git config user.name "github-actions[bot]"
           git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
           git add PKGBUILD .SRCINFO
-          git commit -m "Update to ${{ steps.version.outputs.VERSION }}"
+          git commit -m "Update to $RELEASE_VERSION"
 
           attempts=3
           for attempt in $(seq 1 "$attempts"); do

.gitignore

@@ -49,6 +49,7 @@ tests/*
 !.agents/skills/
 .agents/skills/*
 !.agents/skills/subminer-change-verification/
+!.agents/skills/subminer-release/
 !.agents/skills/subminer-scrum-master/
 .agents/skills/subminer-change-verification/*
 !.agents/skills/subminer-change-verification/SKILL.md
@@ -56,6 +57,8 @@ tests/*
 .agents/skills/subminer-change-verification/scripts/*
 !.agents/skills/subminer-change-verification/scripts/classify_subminer_diff.sh
 !.agents/skills/subminer-change-verification/scripts/verify_subminer_change.sh
+.agents/skills/subminer-release/*
+!.agents/skills/subminer-release/SKILL.md
 .agents/skills/subminer-scrum-master/*
 !.agents/skills/subminer-scrum-master/SKILL.md
 favicon.png

AGENTS.md

@@ -1,4 +1,4 @@
-# AGENTS.MD
+# AGENTS.md
 
 ## Internal Docs
 
@@ -13,7 +13,7 @@ Start here, then leave this file.
 
 `docs-site/` is user-facing. Do not treat it as the canonical internal source of truth.
 
-`CLAUDE.md` is a symlink to this file — there is one project instruction file, not two.
+`CLAUDE.md` is a symlink to this file; there is one project instruction file, not two.
 
 ## Quick Start
 
@@ -25,8 +25,9 @@ Start here, then leave this file.
 
 ## Build / Test
 
-- Runtime/package manager: Bun (`packageManager: bun@1.3.5`)
-- Default handoff gate:
+- Runtime/package manager: Bun; use the version pinned by `package.json`.
+- Follow [`docs/workflow/verification.md`](./docs/workflow/verification.md) and start with the cheapest sufficient lane.
+- Full handoff gate for substantial changes:
   `bun run typecheck`
   `bun run test:fast`
   `bun run test:env`
@@ -44,13 +45,15 @@ Start here, then leave this file.
 - Runtime-compat / dist-sensitive: `bun run test:runtime:compat`
 - Stats dashboard UI (`stats/`): `bun run test:stats`
 - Build/release scripts (`scripts/**`): `bun run test:scripts`
-- Docs-only: `bun run docs:test`, then `bun run docs:build`
+- Internal docs, `AGENTS.md`, or repo skills: `bun run test:docs:kb`
+- User-facing `docs-site/`: `bun run docs:test`, then `bun run docs:build`
+- macOS mpv window helper: `bun test scripts/get-mpv-window-macos.test.ts`
 - Test lanes are directory-discovered via `scripts/test-lanes.ts`; never hand-list test files in `package.json`
 
 ## Docs Upkeep
 
 - Docs ship with the change, not after. If a change alters behavior, defaults, flags, shortcuts, ports, or APIs, update the matching docs in the same PR. Touching code without reconciling its docs is an incomplete change.
-- Source of truth for config defaults is the generated `config.example.jsonc`. Never write a default value into prose you didn't read from it — and don't restate the same default across multiple docs; cite/link to one place so there's a single thing to update.
+- Source of truth for config defaults is the generated `config.example.jsonc`. Never write a default value into prose you didn't read from it, and don't restate the same default across multiple docs; cite/link to one place so there's a single thing to update.
 - Trigger map (touch left → update right):
   - `src/config/definitions/**` (schema/defaults/template) → `bun run generate:config-example`, then reconcile `docs-site/configuration.md` + any feature doc that cites that default
   - shortcuts/keybindings (`shortcuts.*`, `keybindings`, `stats.*Key`, `subtitleSidebar.toggleKey`, controller bindings) → `docs-site/shortcuts.md`
@@ -71,16 +74,10 @@ Start here, then leave this file.
 
 ## Release / PR Notes
 
-- User-visible PRs need reconciled current-outcome fragment(s) in `changes/*.md` — format and rules in [`changes/README.md`](./changes/README.md) (`type` + `area` keys required; inspect existing same-PR fragments, then update/remove stale bullets or add only genuinely separate outcomes; apply the `skip-changelog` label to opt out)
+- User-visible PRs need reconciled current-outcome fragment(s) in `changes/*.md`. Format and rules live in [`changes/README.md`](./changes/README.md) (`type` + `area` keys required; inspect existing same-PR fragments, then update/remove stale bullets or add only genuinely separate outcomes; apply the `skip-changelog` label to opt out).
 - User-visible docs changes get a `type: docs` fragment
 - CI enforces `bun run changelog:lint` and `bun run changelog:pr-check`
 - PR review helpers:
-  - `gh pr view --json number,title,url --jq '"PR #\\(.number): \\(.title)\\n\\(.url)"'`
+  - `gh pr view --json number,title --jq '"PR #\\(.number): \\(.title)"'`
   - `gh api repos/:owner/:repo/pulls/<num>/comments --paginate`
-
-## Runtime Notes
-
-- Use Codex background for long jobs; tmux only when persistence/interaction is required
-- CI red: `gh run list/view`, rerun, fix, repeat until green
-- TypeScript: keep files small; follow existing patterns
-- Only Swift is the `scripts/get-mpv-window-macos.swift` helper (macOS mpv window detection); validate via `bun test scripts/get-mpv-window-macos.test.ts`
+- For CI debugging, inspect runs with `gh run list/view`; rerun or fix only within the requested scope.

CHANGELOG.md

@@ -1,5 +1,61 @@
 # Changelog
 
+## v0.19.5 (2026-08-30)
+
+### Fixed
+
+- **Anki Card Update Progress**: The card-update spinner now stays visible until audio and image updates finish, instead of disappearing early.
+- **Anki Word-Card Fields**: Word-card enrichment now writes sentence text and audio to the fields configured in AnkiConnect, while the dedicated sentence-card and audio-card actions keep their existing compatible field names.
+- **Overlapping Subtitles**:
+  - Subtitle lines that start while another line is still on screen now appear alongside it, instead of staying hidden until a track switch or seek.
+  - Subtitles shown at the same time now stack by their authored screen position, with top signs and song lines above bottom dialogue.
+  - Half-size ASS furigana is no longer shown as if it were a dialogue line.
+- **YouTube Auto Captions**:
+  - Auto-generated captions now follow their intended timing and two-row roll-up layout.
+  - Long speech is paged instead of covering the video with a wall of text.
+  - Explicitly timed sound cues like `[音楽]` no longer cover later dialogue.
+
+## v0.19.4 (2026-08-25)
+
+### Added
+- **Library Merge & Move**: Duplicate library cards for the same show can now be combined. Select cards in the library grid and use "Merge Selected" to pick which entry to keep and move every episode onto it, preserving sessions, mined cards, and watch time. Episodes can also be reassigned individually via the "→" button, useful when a file lands under a stray title; manual assignments survive later filename parsing, Jellyfin refreshes, and season repair. Exact AniList title matches with compatible seasons now merge automatically, while fuzzy matches surface as dismissible "Possible duplicate" reviews instead of merging silently.
+- **Duplicate Line Cleanup Tool**: The Vocabulary tab's new "Duplicates" button scans a chosen time window (7 days through all time) for old karaoke/typeset duplicate-line bursts, shows what it found, and collapses each run to one line once confirmed; `subminer stats cleanup --duplicate-lines` does the same from the terminal, with `--dry-run` and `--lookback-days <n>` options. Watch time and lines-seen totals are left unchanged.
+
+### Changed
+- **Prerelease Release Notes**: Prerelease notes now open with a "Changes since" section listing only what changed versus the previous beta/RC of the same version, above the cumulative highlights, and CI rejects prerelease tags whose committed notes were generated for a different beta/RC.
+
+### Fixed
+- **Subtitle & Karaoke Duplication**:
+  - Karaoke and animated signs are reconstructed once from their authored text and shown only while actually sung, with original word spacing preserved, instead of flooding the overlay, subtitle sidebar, immersion history, mining, or stats with glyph fragments, per-frame color phases, and repeated animation events.
+  - Decorative layers (highlight sweeps, glow/shadow copies, symbol-font decoration, particle swarms, hidden or zero-scaled text) stay out of published text, while ordinary repeated dialogue, positioned signs, wrapped lyric rows, and multi-row CC-style blocks still display correctly.
+  - Embedded subtitle tracks on network-mounted (SMB/NFS) media are extracted and parsed again instead of falling back to live-text-only, restoring karaoke reconstruction, sidebar cues, and mining for releases that only ship subtitles inside the container.
+  - Secondary subtitles go through the same deduplication pipeline as primary subtitles and no longer clip display after about four lines.
+  - Event-heavy karaoke files that previously stalled subtitle loading for several seconds now parse in well under a second.
+- **Character Dictionary Reliability**:
+  - Generation, merged rebuilds, and imports no longer freeze the app on large dictionaries; snapshot I/O, archive building, and image/name lookup caches moved off the UI's critical path.
+  - Dictionaries are reused instead of regenerated when MeCab finds no name splits.
+  - Cached portraits restore correctly after the portrait index finishes loading post-tokenization.
+  - Desktop progress notifications on Linux AppImage installs update in place instead of flickering, fixing a bug where the AppImage's bundled libraries broke the system notification helper.
+- **Overlay Startup & Modals**:
+  - The macOS window-tracking helper targets macOS 12.0+ instead of requiring the build machine's exact macOS version, fixing crashes on older systems like Ventura that left the overlay stuck on "Overlay loading".
+  - mpv IPC connection attempts time out and retry, showing an actionable error if content still isn't ready after 30 seconds.
+  - Dedicated overlay modals are prewarmed on macOS and Windows so shortcuts open them promptly.
+  - On macOS, reused modals and the stats window open above fullscreen mpv on its current Space instead of jumping to another desktop.
+- **Wayland File Drop**: Fixed native Wayland drag-and-drop from file managers such as Thunar, so subtitle and video files dropped on the visible overlay are resolved and forwarded to mpv.
+- **Windows Mouse Lag**: Fixed system-wide mouse lag on Windows while SubMiner is running, caused by the overlay's global mouse hook for click-through forwarding and by the mpv window tracker blocking the app on repeated PowerShell lookups.
+- **Sentence Mining Audio & Clips**: Sentence-audio generation no longer times out on slow network-mounted media with many subtitle/font streams (bounded FFmpeg probing, two-minute extraction budget, clearer error reporting), and mined audio/animated AVIF clips now capture the subtitle line that was actually mined by snapshotting the clip range at lookup time instead of reading live mpv state later.
+- **Stats Performance & Reliability**: Immersion stats storage now sets its SQLite busy timeout before WAL setup, avoiding transient lock errors under concurrent writes. Deletes in the stats dashboard no longer freeze the UI, run proportional to what's deleted instead of rebuilding full lifetime summaries, retry safely if the delete worker crashes, and no longer rescan the whole library when deleting very common words; a new index also makes large session deletes drop from minutes to milliseconds. Library merges, video moves, and AniList reassignments got the same lifetime-summary fix.
+- **Vocabulary Stats Accuracy**: Vocabulary totals and charts now count all tracked vocabulary instead of only the first page, new-word history uses corrected daily rollups (fixing legacy timestamp and time-zone issues), summary cards refresh automatically after edits to the exclusion list, and rapid exclusion edits no longer race each other.
+- **Rofi MKV Thumbnails**: Fixed missing MKV thumbnails in the Linux rofi picker when system thumbnailer registrations only advertise legacy Matroska MIME aliases.
+
+<details>
+<summary>Internal changes</summary>
+
+### Internal
+- Docs Site Indexing: Excluded the `/main/` and `/v/<version>/` docs trees from search indexing (self-referential canonical, `noindex,follow`, matching `X-Robots-Tag`) so crawlers focus on current docs instead of ~30 archived copies of every page, and restored `<lastmod>` dates in the docs sitemap that were silently dropped by production builds.
+
+</details>
+
 ## v0.19.3 (2026-08-13)
 
 ### Added

Makefile

@@ -2,8 +2,10 @@
 
 APP_NAME := subminer
 THEME_SOURCE := assets/themes/subminer.rasi
+THUMBNAILER_SOURCE := assets/thumbnailers/subminer-ffmpegthumbnailer.thumbnailer
 LAUNCHER_OUT := dist/launcher/$(APP_NAME)
 THEME_FILE := subminer.rasi
+THUMBNAILER_FILE := subminer-ffmpegthumbnailer.thumbnailer
 
 # Default install prefix for the wrapper script.
 PREFIX ?= $(HOME)/.local
@@ -221,11 +223,13 @@ docs-dev: ensure-bun
 
 
 install-linux: build-launcher
-	@printf '%s\n' "[INFO] Installing Linux wrapper/theme artifacts"
+	@printf '%s\n' "[INFO] Installing Linux wrapper/support artifacts"
 	@install -d "$(BINDIR)"
 	@install -m 0755 "$(LAUNCHER_OUT)" "$(BINDIR)/$(APP_NAME)"
 	@install -d "$(LINUX_DATA_DIR)/themes"
 	@install -m 0644 "./$(THEME_SOURCE)" "$(LINUX_DATA_DIR)/themes/$(THEME_FILE)"
+	@install -d "$(LINUX_DATA_DIR)/thumbnailers"
+	@install -m 0644 "./$(THUMBNAILER_SOURCE)" "$(LINUX_DATA_DIR)/thumbnailers/$(THUMBNAILER_FILE)"
 	@install -d "$(LINUX_DATA_DIR)/plugin/subminer"
 	@cp -R ./plugin/subminer/. "$(LINUX_DATA_DIR)/plugin/subminer/"
 	@if [ -n "$(APPIMAGE_SRC)" ]; then \
@@ -234,7 +238,7 @@ install-linux: build-launcher
 		printf '%s\n' "[WARN] No release/SubMiner-*.AppImage found; skipping AppImage install"; \
 		printf '%s\n' "       Build one with: make build"; \
 	fi
-	@printf '%s\n' "Installed to:" "  $(BINDIR)/subminer" "  $(LINUX_DATA_DIR)/themes/$(THEME_FILE)"
+	@printf '%s\n' "Installed to:" "  $(BINDIR)/subminer" "  $(LINUX_DATA_DIR)/themes/$(THEME_FILE)" "  $(LINUX_DATA_DIR)/thumbnailers/$(THUMBNAILER_FILE)"
 
 install-macos: build-launcher
 	@printf '%s\n' "[INFO] Installing macOS wrapper/theme/app artifacts"
@@ -275,8 +279,9 @@ uninstall:
 uninstall-linux:
 	@rm -f "$(BINDIR)/subminer" "$(BINDIR)/SubMiner.AppImage"
 	@rm -f "$(LINUX_DATA_DIR)/themes/$(THEME_FILE)"
+	@rm -f "$(LINUX_DATA_DIR)/thumbnailers/$(THUMBNAILER_FILE)"
 	@rm -rf "$(LINUX_DATA_DIR)/plugin/subminer"
-	@printf '%s\n' "Removed:" "  $(BINDIR)/subminer" "  $(BINDIR)/SubMiner.AppImage" "  $(LINUX_DATA_DIR)/themes/$(THEME_FILE)" "  $(LINUX_DATA_DIR)/plugin/subminer"
+	@printf '%s\n' "Removed:" "  $(BINDIR)/subminer" "  $(BINDIR)/SubMiner.AppImage" "  $(LINUX_DATA_DIR)/themes/$(THEME_FILE)" "  $(LINUX_DATA_DIR)/thumbnailers/$(THUMBNAILER_FILE)" "  $(LINUX_DATA_DIR)/plugin/subminer"
 
 uninstall-macos:
 	@rm -f "$(BINDIR)/subminer"

README.md

@@ -15,7 +15,7 @@ Integrates Yomitan and mpv - on-screen lookups, mine to Anki, and track immersio
 [![License](https://img.shields.io/github/license/ksyasuda/SubMiner?style=flat-square&color=1a1a2e)](https://www.gnu.org/licenses/gpl-3.0)
 [![TypeScript](https://img.shields.io/badge/TypeScript-1a1a2e?style=flat-square&logo=typescript&logoColor=3178c6)](https://www.typescriptlang.org)
 
-[![SubMiner demo](./assets/minecard.webp)](https://github.com/user-attachments/assets/89e61895-e2b7-4b47-8d50-a35afe4132b2)
+[![SubMiner demo](./assets/minecard.webp)](https://github.com/user-attachments/assets/7abab8a9-4e4e-4f06-9f3c-9783e15a3807)
 
 </div>
 
@@ -86,6 +86,10 @@ Browse sibling episode files and the active mpv queue in one overlay modal. Open
     <td><b>Jellyfin</b></td>
     <td>Browse, launch, and cast media from your Jellyfin server with setup and discovery controls in the app tray</td>
   </tr>
+  <tr>
+    <td><b>Anime Browser</b></td>
+    <td>Search anime sources you supply as <a href="https://github.com/aniyomiorg/aniyomi">Aniyomi</a> extension APKs and play an episode in mpv with the overlay attached (<code>subminer anime</code>); SubMiner ships no repositories and bundles no sources</td>
+  </tr>
   <tr>
     <td><b>Jimaku</b></td>
     <td>Search and download Japanese subtitles</td>
@@ -178,6 +182,8 @@ See the [full requirements list](https://docs.subminer.moe/installation#_1-insta
 
 ```bash
 paru -S subminer-bin
+# optional: the anime browser bridge, shared with Mangatan and updated by pacman
+paru -S mangatan-extension-server
 ```
 
 </details>
@@ -256,19 +262,28 @@ Full guides on configuration, Anki setup, Jellyfin, immersion tracking, and more
 
 SubMiner builds on the work of these open-source projects:
 
-| Project                                                                                     | Role                                                                    |
-| ------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------- |
-| [ani-skip](https://github.com/synacktraa/ani-skip)                                          | AniSkip API client for anime intro/outro skip timestamps                |
-| [Anacreon-Script](https://github.com/friedrich-de/Anacreon-Script)                          | Inspiration for the mining workflow                                     |
-| [asbplayer](https://github.com/killergerbah/asbplayer)                                      | Inspiration for subtitle sidebar and logic for YouTube subtitle parsing |
-| [Bee's Character Dictionary](https://github.com/bee-san/Japanese_Character_Name_Dictionary) | Character name recognition in subtitles                                 |
-| [GameSentenceMiner](https://github.com/bpwhelan/GameSentenceMiner)                          | Inspiration for Electron overlay with Yomitan integration               |
-| [jellyfin-mpv-shim](https://github.com/jellyfin/jellyfin-mpv-shim)                          | Jellyfin integration                                                    |
-| [Jimaku.cc](https://jimaku.cc)                                                              | Japanese subtitle search and downloads                                  |
-| [Renji's Texthooker Page](https://github.com/Renji-XD/texthooker-ui)                        | Base for the WebSocket texthooker integration                           |
-| [Yomitan](https://github.com/yomidevs/yomitan)                                              | Dictionary engine powering all lookups and the morphological parser     |
-| [yomitan-jlpt-vocab](https://github.com/stephenmk/yomitan-jlpt-vocab)                       | JLPT level tags for vocabulary                                          |
+| Project                                                                                     | Role                                                                         |
+| ------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------- |
+| [ani-skip](https://github.com/synacktraa/ani-skip)                                          | AniSkip API client for anime intro/outro skip timestamps                     |
+| [Anacreon-Script](https://github.com/friedrich-de/Anacreon-Script)                          | Inspiration for the mining workflow                                          |
+| [Aniyomi](https://github.com/aniyomiorg/aniyomi)                                            | Anime extension API and data model the anime browser targets                 |
+| [asbplayer](https://github.com/killergerbah/asbplayer)                                      | Inspiration for subtitle sidebar and logic for YouTube subtitle parsing      |
+| [Bee's Character Dictionary](https://github.com/bee-san/Japanese_Character_Name_Dictionary) | Character name recognition in subtitles                                      |
+| [GameSentenceMiner](https://github.com/bpwhelan/GameSentenceMiner)                          | Inspiration for Electron overlay with Yomitan integration                    |
+| [jellyfin-mpv-shim](https://github.com/jellyfin/jellyfin-mpv-shim)                          | Jellyfin integration                                                         |
+| [Jimaku.cc](https://jimaku.cc)                                                              | Japanese subtitle search and downloads                                       |
+| [M-Extension-Server](https://github.com/1Selxo/M-Extension-Server)                          | Runs Aniyomi extension APKs off Android; the bridge the anime browser drives |
+| [Mangatan](https://github.com/1Selxo/Mangatan)                                              | Reference client for the bridge protocol the anime browser speaks            |
+| [Renji's Texthooker Page](https://github.com/Renji-XD/texthooker-ui)                        | Base for the WebSocket texthooker integration                                |
+| [Yomitan](https://github.com/yomidevs/yomitan)                                              | Dictionary engine powering all lookups and the morphological parser          |
+| [yomitan-jlpt-vocab](https://github.com/stephenmk/yomitan-jlpt-vocab)                       | JLPT level tags for vocabulary                                               |
 
 ## License
 
 [GNU General Public License v3.0](LICENSE)
+
+The anime browser drives [M-Extension-Server](https://github.com/1Selxo/M-Extension-Server),
+downloaded from its upstream releases at runtime rather than
+bundled or redistributed here; its bundles carry their own dependencies, including a JRE and
+GPL-3.0 NewPipe Extractor. SubMiner includes none of them and talks to the bridge
+over its own HTTP protocol. It ships no extensions or repositories by default.

assets/thumbnailers/subminer-ffmpegthumbnailer.thumbnailer

@@ -0,0 +1,4 @@
+[Thumbnailer Entry]
+TryExec=ffmpegthumbnailer
+Exec=ffmpegthumbnailer -i %i -o %o -s %s -f
+MimeType=video/matroska;video/matroska-3d;video/x-matroska;video/x-matroska-3d;

bun.lock

@@ -38,7 +38,7 @@
     "app-builder-lib": "26.15.3",
     "brace-expansion": "5.0.9",
     "electron-builder-squirrel-windows": "26.15.3",
-    "fast-uri": "3.1.5",
+    "fast-uri": "3.1.6",
     "form-data": "4.0.6",
     "ip-address": "10.2.0",
     "js-yaml": "4.3.1",
@@ -406,7 +406,7 @@
 
     "fast-levenshtein": ["fast-levenshtein@2.0.6", "", {}, "sha512-DCXu6Ifhqcks7TZKY3Hxp3y6qphY5SJZmrWMDrKcERSOXWQdMhU9Ig/PYrzyw/ul9jOIyh0N4M0tbC5hodg8dw=="],
 
-    "fast-uri": ["fast-uri@3.1.5", "", {}, "sha512-gHwA1O9LDIcKunMKhObS/HimwtehO1nPUECKAu5TpKgaO19fcWEl4bliWe1jWxVFvIXztJjjQ4L8XQ1EU9f7Jw=="],
+    "fast-uri": ["fast-uri@3.1.6", "", {}, "sha512-7Ical1vFEMr0onbVzEDIreM22I4khW+fzyQPwvAFWBp1iwdshSZRsL4jjRvPG9JP1uiqMHRto+YU6R2/CzDz5Q=="],
 
     "fdir": ["fdir@6.5.0", "", { "peerDependencies": { "picomatch": "^3 || ^4" }, "optionalPeers": ["picomatch"] }, "sha512-tIbYtZbucOs0BRGqPJkshJUYdL+SDH7dVM8gjy+ERp3WAUjLEFJE+02kanyHtwjWOnwrKYBiwAmM0p4kLJAnXg=="],
 

changes/README.md

@@ -42,13 +42,14 @@ How fragments turn into a release:
 
 - At release time, `bun run changelog:build` (and `bun run changelog:prerelease-notes`) pipes every pending fragment through `claude -p` to merge related items, drop noise, and rewrite into a clean user-facing release body. Write fragments as raw, informative notes — don't worry about polished prose, deduping across PRs, or line-by-line phrasing. The polish step handles all of that.
 - The polish step treats pending fragments as the final release outcome, not prerelease history. If a feature is added and then renamed or fixed before the stable cut, ship the final feature bullet instead of separate prerelease-only breaking/fix entries.
-- GitHub release notes and prerelease notes use short top-level items with nested bullets for the change, user benefit, and any useful action note. The stable `CHANGELOG.md` can stay in compact single-line bullets.
+- `CHANGELOG.md`, GitHub release notes, and prerelease notes all use short top-level items with one nested bullet per distinct change, instead of packing a release's worth of detail into a single paragraph bullet. An item with only one thing to say stays inline on the top-level bullet. Release notes and prerelease notes additionally cover user benefit and any useful action note in their nested bullets.
 - `internal` fragments stay in `CHANGELOG.md` (inside a collapsed `<details>` block) but are dropped from the GitHub release notes entirely.
 - The polished `CHANGELOG.md` and `release/release-notes.md` are committed and reviewed before tagging — edit the Markdown by hand if Claude misses something.
 
 Prerelease notes:
 
 - prerelease tags like `v0.11.3-beta.1` and `v0.11.3-rc.1` reuse the current pending fragments to generate `release/prerelease-notes.md`
+- from the second prerelease of a base version onward, the notes also open with a `## Changes since <previous tag>` section generated from the fragment diff against the previous beta/RC tag; keep fragment edits meaningful. Editorial-only rewording is filtered out of that section, while genuinely changed behavior and deleted fragments (reverted changes) are reported
 - existing prerelease notes are a reviewed baseline; later prerelease runs should replace stale beta/RC wording with the current outcome instead of appending fix churn
 - prerelease note generation does not consume fragments and does not update `CHANGELOG.md` or `docs-site/changelog.md`
 - the final stable release is the point where `bun run changelog:build` consumes fragments into the stable changelog and release notes

changes/anilist-season-resolution.md

@@ -0,0 +1,7 @@
+type: fixed
+area: anilist
+
+- Resolving "season N" against AniList no longer lands one season short for franchises whose broadcast seasons are listed as several entries. AniList records the back half of a split cour ("… Season 2 Part 2", "… Cour 2", "第2クール") as its own sequel, and the resolver counted each as a season — so Mushoku Tensei season 3 resolved to the season 2 entry and Re:ZERO season 4 to season 3, which then drove the character dictionary and AniList progress updates to the wrong series. A sequel carrying a part or cour marker that matches the season it continues is now followed without advancing the season count, in both the relation walk and the air-order fallback. Titles are compared ignoring punctuation, and the marker is looked for in every title and synonym rather than just the display title. A wrong match cached before this fix is remembered in `character-dictionaries/anilist-resolution-cache.json` and needs removing (or overriding from the character dictionary picker) for the affected series.
+- A special or OVA sitting in a franchise's sequel chain is walked through without counting as a season. Dr. STONE links STONE WARS to New World through a one-episode special, which made New World resolve as season 4 and every later season shift with it.
+- An AniList rate limit or network failure while walking sequel relations is reported so the caller can retry, instead of falling through to the air-date fallback. That fallback is for a franchise with missing relation edges; running it after a failed lookup turned a transient 429 into a confidently wrong season.
+- The shared title parser recognizes spelled-out episode labels (`Episode 4`, `第4話`) and a season named at the end of the title (`… Season 3`, `… 3rd Season`, `… S3`), which now fill the Season field instead of being searched for as part of the series name. The TsukiHime modal gained a Season field to match Jimaku, and seasons after the first are included in its search query so a later season's releases are actually found.

changes/anime-browser.md

@@ -0,0 +1,34 @@
+type: added
+area: anime
+
+- Added an anime browser window that searches Aniyomi extension sources, shows cover art and episode lists, and plays an episode in mpv so the overlay and mining tools attach as usual.
+- Added `subminer anime` and the `--anime` flag to open the browser, plus a "Browse Anime" tray entry.
+- Anime extensions are read from `<userData>/anime-extensions`; drop Aniyomi `.apk` files there to add sources.
+- Added a source settings tab so extensions that need configuration (server address, credentials, quality) can be set up from the browser; values persist per extension and source, and updated extension schemas replace stale saved field definitions without losing values. Older unscoped preferences are discarded once because their package ownership cannot be proven safely.
+- Added an Extensions tab for adding repository URLs and installing, updating, or removing extensions in place; extensions that fail to load are listed with the reason. Repository requests time out instead of hanging, APK updates are staged before replacing the installed copy, and removing an extension refreshes the runtime even when saved credential cleanup fails while still reporting that failure.
+- Browse, Extensions, and Source settings are tabs, so each one gets the full window instead of sharing it with the search results.
+- Repository URLs only need to be an https URL to a `.json` index; the file name is not restricted to `index.min.json`.
+- The source picker offers "All sources", which searches every installed source at once. Results stream in as each source answers, with per-source progress in the status bar. Results are tagged with their source, failures do not blank the grid, and **Load more** appends later pages without duplicating streamed entries.
+- The Extensions tab opens with an Installed section listing every extension on disk with Remove, including ones added by hand or whose repository has since been removed. It compares APK and repository version codes, enables Update only for newer builds, marks current extensions as Up to date, and offers Update all when multiple updates are waiting.
+- Added `anime.repos`, `anime.extensionsDir`, `anime.preferredQuality`, and `anime.defaultSource` config keys; **Set default** on an Installed row in the Extensions tab makes that source (or All sources) the one the browser opens on, and the current default carries a tag. SubMiner ships no extension repositories and performs no discovery.
+- The settings app now groups Anime Browser config under an **Aniyomi** section in Integrations. A new `anime.autoOpenJimaku` option pauses newly loaded Anime Browser episodes, waits for mpv's video window to appear, closes the in-player browser, hides the standalone browser window so it is not pulled in front of mpv when the modals close, opens Jimaku with the episode details filled in, and resumes after a subtitle loads or the modal closes without overriding playback that was already paused.
+- Anime playback targets Japanese audio: dub-labelled entries are skipped when the source offers an alternative, `alang` prefers Japanese, and the source's own audio and subtitle tracks are loaded into mpv (Japanese selected) instead of being discarded, so all of them can be switched from mpv's track menu.
+- The primary subtitle slot stays reserved for Japanese: a source that only has, say, English subtitles gets them added with a normalized language tag (`English` → `en`) but not selected, so the regular `secondarySub` auto-load can route them to the secondary slot instead.
+- HLS streams pass through a local strip proxy that removes fake image headers some hosts glue onto their video segments and gives disguised segment URLs (`.image`, `.jpg`, `.css`, and other rotating fake extensions) a media-safe local alias, so those streams play in mpv and support Anki audio and image extraction with current ffmpeg releases. Upstream responses use identity encoding so playlists remain readable for URL rewriting.
+- The strip proxy retries a failed segment fetch once after a short pause and logs upstream error statuses; a host that errors on the very first fetches right after an episode resolves no longer kills the whole playback, disconnecting clients release their active upstream fetch instead of consuming the socket and bandwidth in the background, stalled response bodies retain the upstream inactivity timeout, and only origin-form requests can reach the configured local bridge.
+- The strip proxy no longer forwards `Range` headers to the bridge: ffmpeg opens every HLS segment with `Range: bytes=0-`, the bridge answers some of those with 206, and a partial response bypassed the disguise strip, so whether an episode played depended on the bridge's cache state.
+- A bridge that dies out from under the app (killed, crashed, or stopped mid-operation) no longer leaves the browser failing every request until an app restart: the exit is detected, surfaced in the status bar, and the bridge restarts on the next request.
+- "Playing" is only reported once mpv actually configures a video output; when a stream fails to decode, the browser shows mpv's error instead of claiming playback started while no window ever appeared.
+- The bridge bundle is downloaded from the newest upstream release that ships a bundle for the platform (skipping releases older than the oldest server SubMiner is known to work with), the same way Mangatan fetches it, so all four published bundles (macOS arm64 and x64, Linux x64, Windows x64) work without a maintainer hashing each one first.
+- The Extensions tab's available list has a language chip row: pick one or more languages to narrow a repository index that otherwise lists every language it knows, or "All" to clear the filter. Rows name the language ("Japanese" instead of `ja`) and the Available heading counts what the filter leaves.
+- A stream's subtitle tracks are downloaded to a temp directory and loaded into mpv as files rather than streamed from the source URL, so they can serve as the alass reference in Subsync the way Jellyfin subtitles do. The format is detected from the file's own content, a track that fails to download falls back to its URL so the episode still plays, and the directory is removed when the next episode starts or the app exits.
+- An episode launched from the browser carries its series, season and episode number through the app instead of a single joined string: stats group by series (every stream previously landed in one entry named `m3u8`), rewatching reuses the existing entry, the Jimaku and TsukiHime modals prefill Title/Season/Episode from the source's own listing, AniList updates use those fields directly, and the mpv title reads `Series S03E04 - Episode Name`.
+- Opening the browser shows a tray icon on every platform and — on macOS — puts the app in the Cmd+Tab switcher and Dock, so you can switch between it and mpv; the Dock icon is released again when the window closes during playback. Launching a video starts a regular SubMiner session, and in `subminer anime` standalone mode closing the window during playback no longer quits the app and kills the stream.
+- The episode list has a filter box: type an episode number (`12`), a range (`12-18`), or part of an episode name to narrow a long list, with a `6 of 25` counter while it is applied. Escape inside the box clears the filter instead of leaving the detail page.
+- Episodes already watched are dimmed and marked in the episode list, with a watched count in the header. The marks come from the stats history playback already writes (an episode is marked once a session passes the completion threshold), so they match the stats window and survive the stream URL changing between playbacks. They refresh when the browser window comes back to the front, and stay empty when immersion tracking is disabled.
+- Right-clicking an episode opens a menu for marking it watched or unwatched by hand, plus "Mark this and N below watched/unwatched" for the episode and every episode listed under it. Sources list newest first, so a span covers the back catalogue, which is how a series watched elsewhere gets caught up. A filter never narrows what a span covers, and the status bar reports how many episodes were touched.
+- Marking an episode that was never played creates its stats row, carrying the same series, season and episode fields playback would have recorded. Both stats library views join the lifetime tables, so a manual mark does not show up there as watch time nobody spent, and clearing a mark creates nothing.
+- Episodes can be queued instead of replacing what is playing. Every episode row has **Play** and **Queue** buttons (clicking the row still plays now), the right-click menu offers the same two, and a queued episode shows its place in line ("next up", "#2 in queue") with a queue count and **Clear queue** in the episode header. The queue spans anime, resolves and appends each episode to mpv's real playlist as soon as it is queued while subtitle tracks cache in the background, so next/previous navigation works immediately and the next episode starts without a resolution pause when the current one ends. Queueing with nothing playing just plays.
+- Added an in-player Anime Browser modal on `Ctrl+Alt+A`. The shortcut toggles it without losing its page or scroll position. It stays within the player bounds and shares the active episode, playback queue, source configuration, and watch history with the standalone browser, while each surface keeps independent search and navigation state. The modal validates its embedded page before changing overlay state, so a load setup failure leaves it closed rather than revealing a broken modal. The SubMiner logo identifies the Anime Browser in both the modal and standalone window.
+- The bridge is reused from a package-manager install when one exists: on Arch the AUR `mangatan-extension-server` package (shared with Mangatan) is picked up from `/usr/share/mangatan/extension_server`, so nothing is downloaded and pacman keeps it current. `anime.bridgeDir` points SubMiner at a bundle anywhere else. `subminer-bin` lists the package as an optional dependency.
+- SubMiner records which bridge release it installed and, once the bridge is running, checks GitHub for a newer one; when there is, the banner offers an **Update to vX** button. The new release is downloaded beside the running bridge, then the bridge restarts on it. The updater waits for a bridge still starting to stop and keeps the previous bundle until the replacement is active. The Extensions tab shows the bridge version, where it lives, and who updates it without describing an unchecked install as current.

changes/caption-row-sentence-join.md

@@ -0,0 +1,4 @@
+type: fixed
+area: overlay
+
+- Broadcast-style Japanese caption tracks (Crunchyroll JA subs) that split one sentence across two positioned events now publish it as a single line, so `preserveLineBreaks: false` flattens it, the sidebar lists it once, and mined sentences are whole. Rows from two different speakers, sound effects, and labeled turns still stay on separate lines.

changes/jellyfin-zero-subtitle-delay.md

@@ -0,0 +1,4 @@
+type: fixed
+area: jellyfin
+
+- Jellyfin subtitle files now load with zero mpv delay instead of inferring and saving an offset from Japanese and English cue timelines.

changes/linux-overlay-content-ready-fallback.md

@@ -0,0 +1,4 @@
+type: fixed
+area: overlay
+
+- Fixed the overlay never loading (stuck on the "Overlay loading" OSD spinner) when the Yomitan content-script reload raced overlay window creation at startup, most visible when playing from the anime browser on Linux/Wayland: a hidden window stops painting after that reload, so the ready-to-show signal that gates showing the overlay never fired. Content-ready now falls back to did-finish-load after a short grace period.

changes/macos-window-activation.md

@@ -0,0 +1,4 @@
+type: fixed
+area: macos
+
+- `subminer anime` (and `--settings` / `--sync`) now bring their window to the front on macOS. `show()`/`focus()` only reorder windows inside the app that is already active, so the window opened behind the terminal that launched it; SubMiner now activates itself when opening one. The anime browser also restores its Dock icon before showing rather than after, because the overlay's fullscreen transform leaves the app as an accessory process that macOS refuses to bring forward at all.

changes/multiline-copy-timeline.md

@@ -0,0 +1,4 @@
+type: fixed
+area: mining
+
+- Multi-line copy and mining now select backward from the current subtitle in timeline order after seeking, instead of copying lines in playback encounter order. Jumping back to a short previous line also counts as a seek with external subtitle files, so that line becomes the current one.

changes/secondary-subtitle-track-visibility.md

@@ -0,0 +1,4 @@
+type: fixed
+area: overlay
+
+- Native mpv secondary subtitles stay hidden when switching secondary subtitle tracks during playback.

changes/senren-field-grouping.md

@@ -0,0 +1,5 @@
+type: added
+area: anki
+
+- Senren note type support for duplicate-card field grouping: enable `ankiConnect.isSenren` to merge duplicate mined cards using Senren's scene-switching markup, with grouped sentence, furigana, audio, picture, and miscInfo entries.
+- Senren field grouping supports the same auto/manual/disabled modes as Kiku, including the manual merge modal, and is mutually exclusive with Kiku (only one can be enabled at a time).

changes/subsync-file-url-tracks.md

@@ -0,0 +1,4 @@
+type: fixed
+area: subsync
+
+- Subsync no longer fails with `Protocol "file:" not supported` on a subtitle that was dropped onto mpv. mpv reports such a track as a percent-encoded `file://` URL, which subsync read as a stream and tried to fetch over HTTP; the URL is now decoded back to its path, so both the retimed target and an alass reference work. A `file://` video path is treated as local too, which restores the video reference and ffsubsync for a dropped file.

changes/subsync-stream-tracks.md

@@ -0,0 +1,7 @@
+type: fixed
+area: subsync
+
+- Subsync now works when mpv loaded a subtitle track from a URL, which is how Aniyomi extension streams and Jellyfin add theirs. The track is downloaded to a temporary file first — reusing mpv's own request headers, so authenticated and referer-gated hosts stay reachable — instead of being rejected with "Subtitle file not found: https://…". This applies to both the sync target and the alass reference, so a Jimaku or TsukiHime download can be retimed against a stream's own subtitles. Internal tracks of a stream also pass mpv's headers through to `ffmpeg`.
+- Alass now works on tracks that arrive as WebVTT, which is what Aniyomi extension streams serve. Alass picks its parser from the file extension and has no WebVTT support, so it treated a `.vtt` reference as a video file and failed with "no audio stream in file". Both the target and the reference are rewritten as SRT for alass, keeping the cue text as-is, and the originals are left untouched.
+- Leaving `subsync.alass_path`, `ffsubsync_path`, or `ffmpeg_path` empty now actually auto-discovers the binary, as the config help has always claimed. Previously it fell back to a hard-coded `/usr/bin/<tool>`, which does not exist on macOS and broke subsync for every default-config install there. Discovery searches `PATH` plus the usual install prefixes (a GUI launch inherits a minimal `PATH`) and accepts `alass-cli` as well as `alass`. An explicitly configured path is still used verbatim and never silently substituted.
+- Subsync failures are written to the application log. Previously the only trace was an OSD toast that vanished after a few seconds, leaving nothing to diagnose from.

changes/subtitle-recorders-drop-ass-furigana.md

@@ -0,0 +1,4 @@
+type: fixed
+area: subtitles
+
+- Copying the current subtitle, Anki sentence mining from recent lines, and immersion stats no longer include the separate furigana lines that broadcast-caption ASS files place above a word; recorders now use the same furigana-free text the overlay displays.

config.example.jsonc

@@ -341,6 +341,12 @@
         "__playlist-browser-open"
       ] // Command setting.
     },
+    {
+      "key": "Ctrl+Alt+KeyA", // Key setting.
+      "command": [
+        "__anime-browser-open"
+      ] // Command setting.
+    },
     {
       "key": "Ctrl+Shift+KeyH", // Key setting.
       "command": [
@@ -523,7 +529,7 @@
   // ==========================================
   // AnkiConnect Integration
   // Automatic Anki updates and media generation options.
-  // Hot-reload: ankiConnect.ai.enabled, media.normalizeAudio/mirrorMpvVolume, knownWords, nPlusOne, fields.word/audio/image/sentence/miscInfo, behavior.autoUpdateNewCards, isLapis.sentenceCardModel, isKiku.fieldGrouping, and lapisKiku.wordCardKind update live while SubMiner is running.
+  // Hot-reload: ankiConnect.ai.enabled, media.normalizeAudio/mirrorMpvVolume, knownWords, nPlusOne, fields.word/audio/image/sentence/miscInfo, behavior.autoUpdateNewCards, isLapis.sentenceCardModel, isKiku.fieldGrouping, isSenren.fieldGrouping, and lapisKiku.wordCardKind update live while SubMiner is running.
   // Shared AI provider transport settings are read from top-level ai and typically require restart.
   // Most other AnkiConnect settings still require restart.
   // ==========================================
@@ -606,11 +612,31 @@
       "fieldGrouping": "disabled", // Kiku duplicate-card field grouping mode. Values: auto | manual | disabled
       "deleteDuplicateInAuto": true // When Kiku field grouping is "auto", delete the duplicate source card after grouping completes. Values: true | false
     }, // Is kiku setting.
+    "isSenren": {
+      "enabled": false, // Enable Senren-specific duplicate handling (scene-switching field grouping, including miscInfo grouping). Mutually exclusive with isKiku.enabled. Values: true | false
+      "fieldGrouping": "auto", // Senren duplicate-card field grouping mode (scene switching). Values: auto | manual | disabled
+      "deleteDuplicateInAuto": true // When Senren field grouping is "auto", delete the duplicate source card after grouping completes. Values: true | false
+    }, // Is senren setting.
     "lapisKiku": {
       "wordCardKind": "word-and-sentence" // Card-type flag SubMiner marks on Kiku/Lapis word cards. Only one flag is set at a time; the others are cleared. Requires isKiku.enabled or isLapis.enabled. Values: word-and-sentence | click | sentence | audio | none
     } // Lapis kiku setting.
   }, // Automatic Anki updates and media generation options.
 
+  // ==========================================
+  // Anime Browser
+  // Anime browser sources. SubMiner ships no extension repositories and bundles no sources;
+  // add a repository index URL here (or drop .apk files in the extensions directory) to have any.
+  // Hot-reload: autoOpenJimaku applies to the next episode; other anime changes apply the next time the anime browser opens.
+  // ==========================================
+  "anime": {
+    "autoOpenJimaku": false, // Pause Anime Browser playback and open Jimaku when an episode loads. Playback resumes after a subtitle loads or the modal closes only when auto-open initiated the pause. Values: true | false
+    "extensionsDir": "", // Directory holding Aniyomi extension .apk files. Empty uses <userData>/anime-extensions.
+    "repos": [], // Extension repository index URLs (any https .json index, e.g. https://.../index.min.json). Empty by default; SubMiner ships no repositories.
+    "preferredQuality": "", // Preferred stream quality label, matched as a substring (for example: 1080). Empty uses the source order.
+    "defaultSource": "", // Source the Anime Browser selects when it opens: a source id (<package>:<source>) or "all" for every installed source. Empty selects the first installed source. "Set default" in the Extensions tab writes this value.
+    "bridgeDir": "" // Directory holding an M-Extension-Server bundle (java runtime plus server jar) to run instead of the copy SubMiner downloads. Empty checks the package-manager install (Arch: mangatan-extension-server), then <userData>/anime-bridge.
+  }, // Anime browser sources. SubMiner ships no extension repositories and bundles no sources;
+
   // ==========================================
   // Jimaku
   // Jimaku API configuration and defaults.

docs-site/.vitepress/config.ts

@@ -1,3 +1,4 @@
+import { spawnSync } from 'node:child_process';
 import { existsSync, readFileSync, statSync } from 'node:fs';
 import { extname, join, posix, resolve, sep } from 'node:path';
 import type { DefaultTheme, HeadConfig, TransformContext, UserConfig } from 'vitepress';
@@ -26,6 +27,9 @@ function optionalEnv(value: string | undefined): string | undefined {
 const base = normalizeBase(optionalEnv(process.env.SUBMINER_DOCS_BASE) ?? '/');
 const outDir = optionalEnv(process.env.SUBMINER_DOCS_OUT_DIR);
 const docsSourceDir = optionalEnv(process.env.SUBMINER_DOCS_SOURCE_DIR) ?? process.cwd();
+// The tracked `docs-site/` checkout, which stays a git working tree even when
+// `docsSourceDir` points at an untracked release snapshot. Used for git lookups only.
+const repoDocsDir = optionalEnv(process.env.SUBMINER_DOCS_REPO_DIR) ?? process.cwd();
 const channel = normalizeChannel(optionalEnv(process.env.SUBMINER_DOCS_CHANNEL));
 const docsVersion = optionalEnv(process.env.SUBMINER_DOCS_VERSION);
 const latestStable = optionalEnv(process.env.SUBMINER_DOCS_LATEST_STABLE) ?? 'v0.18.0';
@@ -82,15 +86,18 @@ function pageToRoute(page: string): string | null {
   return route ? `/${route}` : '/';
 }
 
+// Only the root channel is indexable. `main` and every /v/<version>/ archive are
+// near-verbatim copies of it, so they own their URL via a self-referential canonical
+// and are excluded from the index instead of being consolidated onto root. Uniform
+// self-canonical plus noindex avoids mixing noindex with a cross-page canonical,
+// which Google treats as a conflicting signal.
+const isIndexableChannel = channel === 'stable-root';
+
 function pageToCanonicalHref(page: string): string | null {
   const route = pageToRoute(page);
   if (!route) return null;
 
-  if (channel === 'main') {
-    return `${DOCS_HOSTNAME}${canonicalRouteWithBase(route)}`;
-  }
-
-  if (channel === 'stable-archive' && docsVersion !== latestStable) {
+  if (!isIndexableChannel) {
     return `${DOCS_HOSTNAME}${canonicalRouteWithBase(route)}`;
   }
 
@@ -106,7 +113,9 @@ function transformPageHead({ page }: TransformContext): HeadConfig[] {
   const href = pageToCanonicalHref(page);
   const head: HeadConfig[] = href ? [['link', { rel: 'canonical', href }]] : [];
 
-  if (channel === 'main') {
+  // Crawlable so links still pass through, but out of the index: ~30 archived copies
+  // of every page otherwise soak up the crawl budget the current docs need.
+  if (!isIndexableChannel) {
     head.push(['meta', { name: 'robots', content: 'noindex,follow' }]);
   }
 
@@ -287,6 +296,39 @@ const versionItems = [
   })),
 ];
 
+function sitemapUrlToPage(url: string): string {
+  const route = url.replace(/\.html$/, '').replace(/^\/+|\/+$/g, '');
+  return route ? `${route}.md` : 'index.md';
+}
+
+// VitePress derives <lastmod> by running `git log` inside its source dir. Production
+// builds point that at an untracked snapshot of the release tag, so the lookup comes
+// back empty and the sitemap ships with no dates at all. Resolve it from the tracked
+// checkout at the ref being built instead.
+function lastModifiedFor(url: string): string | undefined {
+  const ref = docsVersion && docsVersion !== 'main' ? docsVersion : 'HEAD';
+  const result = spawnSync('git', ['log', '-1', '--format=%cI', ref, '--', sitemapUrlToPage(url)], {
+    cwd: repoDocsDir,
+    encoding: 'utf8',
+  });
+
+  return (result.status === 0 && result.stdout.trim()) || undefined;
+}
+
+// Only the root channel publishes a sitemap. Archived and `main` builds would emit
+// their own copies listing the same canonical URLs, which just advertises the
+// duplicate trees we are trying to keep out of the index.
+const sitemap: UserConfig['sitemap'] = isIndexableChannel
+  ? {
+      hostname: DOCS_HOSTNAME,
+      transformItems(items) {
+        return items
+          .filter((item) => item.url !== 'README' && item.url !== `${DOCS_HOSTNAME}/README`)
+          .map((item) => ({ ...item, lastmod: item.lastmod ?? lastModifiedFor(item.url) }));
+      },
+    }
+  : undefined;
+
 const nav: DefaultTheme.NavItem[] = [
   { text: 'Home', link: '/' },
   { text: 'Get Started', link: '/installation' },
@@ -306,7 +348,6 @@ const sidebar: DefaultTheme.SidebarItem[] = [
       { text: 'Usage', link: '/usage' },
       { text: 'Mining Workflow', link: '/mining-workflow' },
       { text: 'Launcher Script', link: '/launcher-script' },
-      { text: 'Feature Demos', link: '/demos' },
     ],
   },
   {
@@ -327,6 +368,7 @@ const sidebar: DefaultTheme.SidebarItem[] = [
       { text: 'Anki', link: '/anki-integration' },
       { text: 'Jellyfin', link: '/jellyfin-integration' },
       { text: 'YouTube', link: '/youtube-integration' },
+      { text: 'Anime Browser', link: '/anime-browser' },
       { text: 'Jimaku', link: '/jimaku-integration' },
       { text: 'TsukiHime', link: '/tsukihime-integration' },
       { text: 'AniList', link: '/anilist-integration' },
@@ -420,14 +462,7 @@ const config: UserConfig = {
   appearance: 'dark',
   cleanUrls: true,
   metaChunk: true,
-  sitemap: {
-    hostname: DOCS_HOSTNAME,
-    transformItems(items) {
-      return items.filter(
-        (item) => item.url !== 'README' && item.url !== `${DOCS_HOSTNAME}/README`,
-      );
-    },
-  },
+  sitemap,
   transformHead: transformPageHead,
   lastUpdated: true,
   srcExclude: ['subagents/**', 'README.md'],

docs-site/README.md

@@ -38,8 +38,10 @@ bun run docs:dev
 The public docs root is stable-only:
 
 - `/` serves the latest stable release docs.
-- `/main/` serves development docs from `main` and is marked `noindex,follow`.
+- `/main/` serves development docs from `main`.
 - `/v/<version>/` serves stable release archives.
 - Prerelease tags do not update the docs site.
 
+Only `/` is indexable. `/main/` and every `/v/<version>/` page carries a self-referential canonical plus `noindex,follow`, and the generated `_headers` file repeats that as an `X-Robots-Tag`. They stay crawlable so their links still resolve, but ~30 archived copies of every page would otherwise consume the crawl budget the current docs need. Only the root build emits `sitemap.xml`, and its `<lastmod>` dates come from `git log` against the tracked checkout at the released tag, because the build renders from an untracked snapshot that VitePress cannot date itself.
+
 Keep Cloudflare Git auto-deploy disabled. The production deploy is `.github/workflows/docs-pages.yml`, which uploads `.tmp/docs-versioned-site` with `--branch main` so tag-triggered runs update Production instead of creating preview deployments.

docs-site/anime-browser.md

@@ -0,0 +1,393 @@
+# Anime Browser
+
+Search anime sources, pick an episode, and play it in mpv with SubMiner's overlay
+and mining tools attached — the same way a local file or a Jellyfin stream works.
+
+Open it with `subminer anime`, with `SubMiner.AppImage --anime`, or from
+**Browse Anime** in the tray menu. The window stays open while you watch, so you
+can queue the next episode without reopening it.
+
+During playback, `Ctrl+Alt+A` toggles the same browser as a modal inside the mpv
+player bounds. It uses a dedicated modal surface, so it stays above fullscreen
+playback and closes like the other in-player tools. Toggling it off keeps its
+current page and scroll position ready for the next toggle. The standalone
+window and the modal keep their own search, selected source, tab, and scroll
+state, so using one does not replace or cancel what you were doing in the other.
+
+Both surfaces use the same playback queue, source configuration, and stats
+history. Queue changes and the currently playing episode appear in both
+immediately, and watched marks come from the same history that playback and the
+stats window update. Closing and reopening the modal therefore picks up progress
+made from either browser surface.
+
+While the window is open, SubMiner shows a tray icon and — on macOS — appears
+in the Cmd+Tab switcher and the Dock (macOS ties the two together), so you can
+flip between the browser and mpv. SubMiner normally hides itself from the Dock
+because the subtitle overlay needs that to float above fullscreen video; it
+hides again when the window closes during playback.
+
+Launching an episode starts a full SubMiner playback session, the same as
+playing a local file: the overlay and mining tools attach, and the tray icon
+stays available. In standalone `subminer anime` mode, closing the window while
+a video is playing leaves playback running — reopen the browser from the tray
+(**Browse Anime**). The app only exits with the window when nothing is playing.
+
+## How it works
+
+SubMiner does not implement any anime source itself. It runs **Aniyomi extension
+APKs** through a bundled JVM sidecar ([M-Extension-Server][mes]), asks the
+selected extension to resolve an episode, and hands the resulting URL to mpv.
+
+```
+extension APK  →  bridge (JVM)  →  { url, headers }  →  mpv  →  SubMiner overlay
+```
+
+Because the extension resolves the stream, whichever sources you install decide
+what is available. SubMiner only hosts them.
+
+## Installing extensions
+
+**SubMiner ships no extension repositories and bundles no sources.** There is no
+default repository, no suggested list, and no discovery. Until you add one, the
+browser has nothing to search — that is deliberate, and it is what keeps SubMiner
+a neutral host rather than a distributor.
+
+There are two ways to add extensions.
+
+### From a repository
+
+The window has three tabs — **Browse**, **Extensions**, and **Source settings** —
+and each one fills the window, so a long extension list is not squeezed in above
+the search results.
+
+Open the **Extensions** tab, paste a repository index URL, and choose
+**Add repository**. The URL must be `https` and point at a `.json` index file —
+`index.min.json` is the common Aniyomi name, but repositories are free to publish
+under another one (for example `video.min.json`). Anything else is rejected
+immediately rather than failing later. Everything before the file name is treated
+as the repository root, so `.apk` and icon URLs are resolved relative to it.
+
+Extensions your repositories offer but you do not have appear under
+**Available**, each with **Install**. Repositories are stored in config under
+`anime.repos`, so you can also manage them there and keep them in a dotfile.
+
+Every row carries the extension's icon, as the repository publishes it, so a
+site is recognisable before you read the name. A repository row shows its host's
+favicon instead. Icons are the only part of a row that is fetched from the
+network, and a row whose icon is missing falls back to the first letter of its
+name rather than an empty box.
+
+A repository index lists every language it knows about, which is far more than
+any one person reads, so the **Available** list has a language chip row above
+it. Pick one or more languages to narrow it, or **All** to clear the filter;
+picking a language replaces **All** rather than sitting beside it. Rows name the
+language in full ("Japanese" rather than `ja`), extensions whose sources span
+languages are grouped under **Multi-language**, and the Available heading counts
+how many of the offered extensions the filter leaves.
+
+### Managing what is installed
+
+The Extensions tab opens with an **Installed** section listing everything in the
+extensions directory, with the sources each one provides and a **Remove**
+button. It is built from the directory rather than from a repository, so an
+extension you dropped in by hand — or one whose repository you have since
+removed — is still listed and still removable. An installed extension borrows
+its icon from the catalogue, so one no repository carries shows its monogram.
+
+SubMiner compares the APK's Android version code with the newest build in the
+configured repositories. **Update** is clickable only when the repository has a
+newer build. A current extension says **Up to date**. If an unusual APK has no
+readable version code, its disabled button says **Version unknown**. **Update
+all** installs every newer build in one pass and disables itself when there is
+nothing to update.
+
+### From a file
+
+Drop Aniyomi `.apk` files into the extensions directory, shown at the top of the
+Extensions tab. It defaults to `<userData>/anime-extensions` — on macOS,
+`~/Library/Application Support/SubMiner/anime-extensions` — and can be moved with
+`anime.extensionsDir`.
+
+A single APK may provide several sources; each appears separately in the
+**Source** picker. Extensions that fail to load are listed in the Installed
+section with the reason, so a broken APK is visible rather than silently
+missing.
+
+An extension that fails to load is skipped rather than blocking the others, so
+one bad APK will not hide the rest.
+
+### Default source
+
+The browser opens on the first installed source. To open on a different one,
+press **Set default** on its row in the Installed list; the row then carries a
+**default** tag, and only one source can carry it. The choice is written to
+`anime.defaultSource` and applies the next time a browser window or the
+in-player modal opens. With more than one source installed, an **All sources**
+row at the top of the list can be the default too. A default that is no longer
+installed falls back to the first source.
+
+## Searching every source at once
+
+With more than one source installed, the **Source** picker gains an
+**All sources** entry. Searching with it selected runs the query against every
+installed source at once, and each source's results appear the moment that
+source answers — a fast source is on screen while a slow one is still
+resolving. The status bar counts sources as they finish
+(`Searching… 3/5 sources · 42 results`).
+
+Each cover is labelled with the source it came from, and opening one always
+queries that source, whatever the picker says afterwards.
+
+A source that errors is named in the status bar and the rest still show their
+results; one extension that needs a login cannot blank the grid. If every
+source fails, the first error is shown in full.
+
+Typing a new search while one is still running simply starts over: results
+from the superseded search are discarded, even if its sources answer late.
+
+When a source reports another page, **Load more** appears below the covers.
+It appends the next page without duplicating entries that already arrived in
+the live result stream. A failed next-page request remains available to retry.
+
+Source settings belong to a single extension, so the **Source settings** tab
+asks you to pick one while **All sources** is selected.
+
+## Finding an episode, and what you have watched
+
+An episode list can run to hundreds of entries, so the episode header carries a
+filter box:
+
+- A number, `12`, keeps that episode. Sources that report no numbers at all
+  are still searched by name, so `12` also matches `Episode 12` in a title.
+- A range, `12-18`, keeps the episodes between the two, in either order
+  (`18-12` reads the same). Episodes the source gave no number are left out of
+  a range.
+- Anything else is a case-insensitive substring of the episode name, so `beach`
+  finds `OVA: Beach Special`.
+
+The counter next to **Episodes** reads `6 of 25` while a filter is applied.
+Pressing Escape in the filter box clears it; pressing it anywhere else goes back
+to the results grid.
+
+Episodes you have already watched are dimmed and marked `✓ watched`, with a
+count in the header. This is not a separate list the browser keeps: it reads the
+same stats history the rest of SubMiner writes to, where an episode is marked
+watched once a session runs past the completion threshold. Streams are recorded
+under a stable per-episode identity, so the mark survives the stream URL
+changing between playbacks, and it is the same mark the stats window and
+`--mark-watched` use.
+
+Because playback marks an episode partway through the session, the marks
+refresh when the browser window comes back to the front: finish an episode in
+mpv, switch back, and it is marked. With
+[immersion tracking](configuration.md) disabled there is no history to read, so
+no episode is marked.
+
+### Marking by hand, and catching up
+
+Right-click an episode for:
+
+- **Mark watched** / **Mark unwatched**: the single episode, whichever way it
+  is not already.
+- **Mark this and N below watched** / **... unwatched**: that episode and every
+  episode listed below it. Sources list newest first, so "below" is the back
+  catalogue: right-click the last episode you saw and mark everything down to
+  the start, which is how you catch up a series you watched somewhere else.
+
+A filter narrows what you are looking at, not what you mark: a span always
+covers the full episode list, and the status bar says how many episodes it
+touched. The oldest episode has nothing below it, so it only offers the single
+entry. Escape closes the menu.
+
+Marking an episode you have never played creates its stats row so the mark has
+somewhere to live. That row carries the same series, season and episode fields
+playback would have recorded, and both stats library views join the lifetime
+tables, so a manually marked episode does not appear there as watch time you
+never spent. Clearing a mark never creates anything.
+
+Marks are written to the stats history, so with immersion tracking disabled
+there is nowhere to write them and the status bar says so.
+
+## Settings
+
+| Key                      | Purpose                                                                  |
+| ------------------------ | ------------------------------------------------------------------------ |
+| `anime.autoOpenJimaku`   | Pause new episodes and open Jimaku for Japanese subtitles.               |
+| `anime.repos`            | Repository index URLs. Empty by default.                                 |
+| `anime.extensionsDir`    | Where APKs are read from. Empty uses `<userData>/anime-extensions`.      |
+| `anime.preferredQuality` | Preferred stream label, matched as a substring (for example `1080`).     |
+| `anime.defaultSource`    | Source the browser opens on: a source id or `all`. Empty uses the first. |
+| `anime.bridgeDir`        | A bridge bundle to run instead of the downloaded one. Empty by default.  |
+
+Enable `anime.autoOpenJimaku` to hand each newly loaded Anime Browser episode
+to Jimaku. SubMiner pauses playback on its first frame, waits for mpv's video
+window to appear, closes the in-player browser if it is open, hides the
+standalone browser window, and opens Jimaku with the source title, season, and
+episode already filled in. Playback resumes after the selected subtitle loads.
+Closing Jimaku also releases the automatic pause, while playback that was
+already paused stays paused. A stream that never shows a window resumes without
+opening Jimaku. Reopen the browser window with `subminer anime`, the tray
+entry, or the shortcut to keep browsing.
+
+## Source settings
+
+Most extensions need configuration before they return anything — a server
+address and credentials, a preferred quality, a language filter. Open the
+**Source settings** tab to edit them. Changes save as you make them and
+persist across restarts in `<userData>/anime-source-preferences.json`.
+
+Each save is handed back to the extension, so it can react: the Jellyfin source
+logs in when the address and password land, then fills in its media-library
+picker. Password-like fields are masked. Because that file can hold
+credentials, it is written with owner-only permissions. Values are scoped to
+the exact extension package and source, so two extensions that reuse the same
+internal source ID cannot read each other's settings. Preferences saved by an
+older build without package ownership are discarded; re-enter those source
+settings once after upgrading.
+
+## The bridge
+
+SubMiner looks for the server in three places, in order, and runs the first
+one it finds:
+
+1. `anime.bridgeDir`, if set. It must hold a Java runtime and the
+   `MExtensionServer-*.jar`, laid out as upstream ships them. A directory that
+   holds neither is an error rather than a silent fallback.
+2. A package-manager install. On Arch that is the AUR
+   [`mangatan-extension-server`](https://aur.archlinux.org/packages/mangatan-extension-server)
+   package, which Mangatan also uses, at `/usr/share/mangatan/extension_server`.
+   Installing it means no download, and pacman keeps the bridge current. The
+   `subminer-bin` package lists it as an optional dependency.
+3. SubMiner's own copy in `<userData>/anime-bridge`. The first launch downloads
+   the newest upstream release that ships a bundle for your platform (~130 MB,
+   containing the server and a matching Java runtime, so no system JDK is
+   required), the same way Mangatan does. Releases older than the oldest server
+   SubMiner is known to work with are skipped. Progress appears in the banner
+   at the top of the window.
+
+The **Bridge** note at the bottom of the Extensions tab says which one is in
+use, its version, and who updates it.
+
+### Updating the bridge
+
+Only the copy SubMiner downloaded is ever updated by SubMiner. A package-manager
+install or an `anime.bridgeDir` belongs to whoever put it there.
+
+SubMiner records the release it installed and, once the bridge is running,
+asks GitHub for the newest one. When upstream has published a newer release,
+the banner reads "Extension bridge v… is installed; v… is available" with an
+**Update to v…** button. Clicking it downloads the new release beside the
+running bridge, so a failed download changes nothing, then stops the bridge,
+swaps the directories, and starts it again. The restart takes a few seconds
+and kills the stream of an episode that is playing, the same as when the
+bridge exits for any other reason; the queue and browser state survive. The
+check is one unauthenticated GitHub API call per bridge start; if it fails
+(offline, rate limited) it is logged and no update is offered until the next
+start.
+
+Upstream publishes no checksums, so the download is trusted the way Mangatan
+and the AUR package trust it: TLS to GitHub and the maintainer's account. That
+is the same trust running the server implies in the first place.
+
+The bridge stays running while the window is open. Resolved video URLs point at
+its own loopback proxy so the extension's cookies and headers apply, which means
+those URLs stop working once it exits — the window keeps it alive for the whole
+session.
+
+If the bridge dies anyway (killed by hand, crashed, or stopped mid-operation),
+the exit is detected and named in the status bar, and the next request starts a
+new one. Playback already in flight still ends when its stream URL dies, but the
+browser recovers without an app restart.
+
+Two known limits:
+
+- There is no Android WebView, so extensions that need one (typically for
+  Cloudflare challenges) will fail with an error from the source.
+- Bundles are published for macOS (arm64, x64), Linux (x64), and Windows (x64).
+  Other platforms are unsupported outright.
+
+## Playback
+
+Selecting an episode resolves the best available stream, applies the source's
+required headers as mpv `http-header-fields`, and loads it. The headers are
+readable back off mpv, so Anki card audio and screenshots fetch correctly too.
+
+HLS streams are routed through a small local proxy before mpv sees them. Some
+hosts disguise their video segments by prepending a fake image header (a real
+1x1 PNG) so scrapers back off; Aniyomi's own player strips this, but ffmpeg
+probes the segment as a picture and playback dies with "no audio or video data
+played". The proxy scans each segment for the first genuine MPEG-TS packet run
+and drops whatever junk sits in front of it. Segments that are not TS (fMP4,
+subtitles, encryption keys) pass through untouched, and direct-file streams
+skip the proxy entirely. Segment URLs disguised behind fake extensions
+(`.image`, `.jpg`, `.css`, and friends) are exposed locally with a `.ts`
+suffix so current ffmpeg releases accept them when Anki extracts audio,
+screenshots, or animated images from the playing stream.
+
+"Playing" in the status bar means playing: after handing mpv the stream,
+SubMiner waits until mpv actually configures a video output before reporting
+success. If mpv gives up instead — a dead host, an undecodable stream — the
+browser shows mpv's error rather than pretending playback started (a failed
+load leaves no mpv window, because the player idles windowless).
+
+Choosing **Queue** resolves the episode and appends the playable stream to mpv's
+own playlist immediately, while its subtitle tracks cache in the background.
+That makes mpv's next command available at once and lets the next episode begin
+automatically when the current one ends, without waiting for another source
+request. Resolved stream URLs can be short-lived, so a very long queue can still
+outlive what its source issued; dequeue and queue that episode again to refresh
+it.
+
+### Japanese audio, and switching tracks
+
+Sources often return a dub and the original audio as two separate entries — or
+as two audio tracks of one stream — and the dub is frequently listed first.
+SubMiner always aims at the Japanese audio:
+
+- Entries labelled as a dub are skipped as long as another entry exists. This
+  outranks `anime.preferredQuality`: a 1080p dub is the wrong file, not a better
+  one. If every entry is a dub, it still plays.
+- mpv's `alang` is set to `ja,jpn,jp,japanese` before the file loads, so a
+  stream carrying several audio tracks starts on the Japanese one. With no
+  Japanese track, mpv falls back to the first one as usual.
+- Any audio or subtitle tracks the extension supplies separately are added to
+  mpv with `audio-add` / `sub-add`, tagged with their language, and the
+  Japanese one is selected.
+
+The primary subtitle slot is reserved for Japanese — it is what the overlay
+mines. A source that only carries, say, English subtitles does not get them
+promoted to primary; instead the track is added with a normalized language tag
+(`English` → `en`), and the regular [dual-subtitle settings](configuration.md)
+apply: with `secondarySub.autoLoadSecondarySub` enabled and the language listed
+in `secondarySub.secondarySubLanguages`, it is picked up as the secondary
+subtitle, exactly as it would be for a local file.
+
+Every track is added, including the ones that are not selected, so all of them
+appear in mpv's track menu and can be switched by hand while watching
+(`#` cycles audio, `j` cycles subtitles by default).
+
+The extension hands over subtitle tracks as URLs, but SubMiner downloads each
+one to a temporary directory and gives mpv the local file. mpv is happy either
+way; [Subsync](/troubleshooting#subtitle-sync-subsync) is not, because alass
+needs a file on disk to use as the timing reference. A track that fails to
+download falls back to its URL so the episode still plays, the format is
+detected from the file's own content rather than its URL, and the directory is
+removed when the next episode starts or the app exits.
+
+### Series, season, and episode
+
+The episode's identity travels with it instead of being guessed back out of the
+stream URL, which carries nothing but a proxy path and a file extension. The
+title and episode label from the source's own listing are split into series,
+season, and episode number once, at launch, and everything downstream reads
+those fields:
+
+- mpv's title reads `Series S03E04 - Episode Name`.
+- Stats group by series, and rewatching an episode reuses its entry instead of
+  creating a new one.
+- The [Jimaku](/jimaku-integration) and [TsukiHime](/tsukihime-integration)
+  modals open with Title, Season, and Episode already filled in, so a subtitle
+  search is one keypress rather than a retype.
+- [AniList](/anilist-integration) progress updates use those fields directly.
+
+[mes]: https://github.com/1Selxo/M-Extension-Server

docs-site/anki-integration.md

@@ -136,6 +136,8 @@ SubMiner maps its data to your Anki note fields. Configure these under `ankiConn
 
 Field names are matched against your Anki note type case-insensitively (an exact match wins, then a lowercase comparison). If a configured field does not exist on the note type, SubMiner skips it without error.
 
+These mappings always control normal word-card enrichment, including Yomitan proxy/polling updates and manual clipboard updates. Enabling Lapis or Kiku does not replace the configured word-card sentence and audio fields with `Sentence` and `SentenceAudio`. The dedicated sentence-card and audio-card shortcuts still use those Lapis/Kiku field names.
+
 Two related options live alongside `fields`: `ankiConnect.deck` (target deck; empty falls back as described above) and `ankiConnect.tags` (tags added to mined cards, default `["SubMiner"]`; set `[]` to disable tagging). The `miscInfo` content is controlled by `ankiConnect.metadata.pattern` (default `[SubMiner] %f (%t)`; tokens: `%f` filename, `%F` filename with extension, `%t` timestamp, `%T` timestamp with milliseconds, `<br>` newline).
 
 ### Minimal Config
@@ -233,7 +235,7 @@ Animated AVIF requires an AV1 encoder (`libaom-av1`, `libsvtav1`, or `librav1e`)
 
 When media is available, mined-card overlay and system notifications include the same current-frame thumbnail.
 
-`overwriteAudio` applies to automatic card updates and duplicate-card enrichment. Manual clipboard subtitle updates (`Ctrl/Cmd+C`, then `Ctrl/Cmd+V`) always replace generated sentence audio, while leaving the word audio field unchanged.
+`overwriteAudio` applies to automatic card updates and duplicate-card enrichment. Manual clipboard subtitle updates (`Ctrl/Cmd+C`, then `Ctrl/Cmd+V`) always replace generated sentence audio in `ankiConnect.fields.audio`, even when `overwriteAudio` is disabled.
 
 ## AI Translation
 
@@ -287,6 +289,8 @@ Sentence card creation and audio card marking require a non-empty `ankiConnect.i
 
 Trigger with the mine sentence shortcut (`Ctrl/Cmd+S` by default). The card is created directly via AnkiConnect with the sentence, audio, and image filled in.
 
+The dedicated sentence-card and audio-card shortcuts use the Lapis/Kiku-compatible `Sentence` and `SentenceAudio` fields. This does not affect the configured fields used to enrich normal word cards.
+
 To mine multiple subtitle lines as one sentence card, use `Ctrl/Cmd+Shift+S` followed by a digit (1–9) to select how many recent lines to combine.
 
 ## Word Card Type (Kiku/Lapis)
@@ -304,9 +308,9 @@ Word cards get a card-type flag when SubMiner fills their sentence, whether that
 
 `click` marks `IsClickCard`, `sentence` marks `IsSentenceCard`, `audio` marks `IsAudioCard`, and `none` leaves the flags untouched for templates that manage them elsewhere. Whichever flag is chosen, the other card-type flags are cleared so the note never claims two card types. The setting is only read when `isKiku` or `isLapis` is enabled, and cards mined with Mine Sentence or Mine Audio keep their own flag.
 
-## Field Grouping (Kiku)
+## Field Grouping (Kiku/Senren)
 
-When you mine the same word multiple times, SubMiner can merge the cards instead of creating duplicates. This is designed for note types like [Kiku](https://github.com/youyoumu/kiku) that support grouped sentence/audio/image fields.
+When you mine the same word multiple times, SubMiner can merge the cards instead of creating duplicates. This is designed for note types that support grouped fields: [Kiku](https://github.com/youyoumu/kiku) and [Senren](https://github.com/BrenoAqua/Senren) (which calls the feature scene switching).
 
 ```jsonc
 "ankiConnect": {
@@ -318,6 +322,18 @@ When you mine the same word multiple times, SubMiner can merge the cards instead
 }
 ```
 
+For Senren note types, enable `isSenren` instead. Kiku and Senren write incompatible markup into the same fields, so only one can be enabled at a time; if both are enabled, Kiku wins and a config warning is emitted.
+
+```jsonc
+"ankiConnect": {
+  "isSenren": {
+    "enabled": true,
+    "fieldGrouping": "auto",           // "auto" (default), "manual", or "disabled"
+    "deleteDuplicateInAuto": true      // delete new card after auto-merge
+  }
+}
+```
+
 ### Modes
 
 **Disabled** (`"disabled"`): No duplicate detection. Each card is independent.
@@ -333,9 +349,12 @@ When you mine the same word multiple times, SubMiner can merge the cards instead
 | Sentence | Both cards' sentences kept as grouped entries |
 | Audio    | Both cards' `[sound:...]` entries kept        |
 | Image    | Both cards' images kept                       |
+| MiscInfo | Both cards' source info kept as grouped entries |
 
 Identical values from both cards are kept as separate grouped entries; the merge does not deduplicate.
 
+The merge markup depends on the note type. Kiku entries are wrapped in `<span data-group-id="...">` spans ordered newest first. Senren entries follow the [scene switching](https://github.com/BrenoAqua/Senren/blob/main/docs/scene_switching.md) format: sentence, sentenceFurigana, and miscInfo entries use `group` spans when ordinal order is sufficient and numbered `groupN` spans when they need an absolute scene target. Audio and pictures are appended positionally, and the number of sentenceAudio entries drives Senren's scene count. Ungrouped legacy content is wrapped into a group span on first merge, and source `groupN` spans are rebased after the kept note's existing audio scenes.
+
 ### Keyboard Shortcuts in the Modal
 
 | Key         | Action                             |

docs-site/architecture.md

@@ -74,8 +74,8 @@ src/
   shared/ipc/              # Cross-process IPC channel constants + payload validators
   renderer/                # Overlay renderer (modularized UI/runtime)
     handlers/              # Keyboard/mouse/gamepad interaction modules
-    modals/                # Modal flows (Jimaku, Kiku, subsync, runtime options, session help,
-                           #   changelog, character dictionary, playlist browser, subtitle
+    modals/                # Modal flows (Anime Browser, Jimaku, Kiku, subsync, runtime options,
+                           #   session help, changelog, character dictionary, playlist browser, subtitle
                            #   sidebar, YouTube track picker, controller config/debug/select)
     positioning/           # Subtitle position controller (drag-to-reposition)
   settings/                # Settings window UI (model, controls, markup)

docs-site/changelog.md

@@ -1,5 +1,61 @@
 # Changelog
 
+## v0.19.5 (2026-08-30)
+
+**Fixed**
+
+- **Anki Card Update Progress**: The card-update spinner now stays visible until audio and image updates finish, instead of disappearing early.
+- **Anki Word-Card Fields**: Word-card enrichment now writes sentence text and audio to the fields configured in AnkiConnect, while the dedicated sentence-card and audio-card actions keep their existing compatible field names.
+- **Overlapping Subtitles**:
+  - Subtitle lines that start while another line is still on screen now appear alongside it, instead of staying hidden until a track switch or seek.
+  - Subtitles shown at the same time now stack by their authored screen position, with top signs and song lines above bottom dialogue.
+  - Half-size ASS furigana is no longer shown as if it were a dialogue line.
+- **YouTube Auto Captions**:
+  - Auto-generated captions now follow their intended timing and two-row roll-up layout.
+  - Long speech is paged instead of covering the video with a wall of text.
+  - Explicitly timed sound cues like `[音楽]` no longer cover later dialogue.
+
+## v0.19.4 (2026-08-25)
+
+**Added**
+- **Library Merge & Move**: Duplicate library cards for the same show can now be combined. Select cards in the library grid and use "Merge Selected" to pick which entry to keep and move every episode onto it, preserving sessions, mined cards, and watch time. Episodes can also be reassigned individually via the "→" button, useful when a file lands under a stray title; manual assignments survive later filename parsing, Jellyfin refreshes, and season repair. Exact AniList title matches with compatible seasons now merge automatically, while fuzzy matches surface as dismissible "Possible duplicate" reviews instead of merging silently.
+- **Duplicate Line Cleanup Tool**: The Vocabulary tab's new "Duplicates" button scans a chosen time window (7 days through all time) for old karaoke/typeset duplicate-line bursts, shows what it found, and collapses each run to one line once confirmed; `subminer stats cleanup --duplicate-lines` does the same from the terminal, with `--dry-run` and `--lookback-days <n>` options. Watch time and lines-seen totals are left unchanged.
+
+**Changed**
+- **Prerelease Release Notes**: Prerelease notes now open with a "Changes since" section listing only what changed versus the previous beta/RC of the same version, above the cumulative highlights, and CI rejects prerelease tags whose committed notes were generated for a different beta/RC.
+
+**Fixed**
+- **Subtitle & Karaoke Duplication**:
+  - Karaoke and animated signs are reconstructed once from their authored text and shown only while actually sung, with original word spacing preserved, instead of flooding the overlay, subtitle sidebar, immersion history, mining, or stats with glyph fragments, per-frame color phases, and repeated animation events.
+  - Decorative layers (highlight sweeps, glow/shadow copies, symbol-font decoration, particle swarms, hidden or zero-scaled text) stay out of published text, while ordinary repeated dialogue, positioned signs, wrapped lyric rows, and multi-row CC-style blocks still display correctly.
+  - Embedded subtitle tracks on network-mounted (SMB/NFS) media are extracted and parsed again instead of falling back to live-text-only, restoring karaoke reconstruction, sidebar cues, and mining for releases that only ship subtitles inside the container.
+  - Secondary subtitles go through the same deduplication pipeline as primary subtitles and no longer clip display after about four lines.
+  - Event-heavy karaoke files that previously stalled subtitle loading for several seconds now parse in well under a second.
+- **Character Dictionary Reliability**:
+  - Generation, merged rebuilds, and imports no longer freeze the app on large dictionaries; snapshot I/O, archive building, and image/name lookup caches moved off the UI's critical path.
+  - Dictionaries are reused instead of regenerated when MeCab finds no name splits.
+  - Cached portraits restore correctly after the portrait index finishes loading post-tokenization.
+  - Desktop progress notifications on Linux AppImage installs update in place instead of flickering, fixing a bug where the AppImage's bundled libraries broke the system notification helper.
+- **Overlay Startup & Modals**:
+  - The macOS window-tracking helper targets macOS 12.0+ instead of requiring the build machine's exact macOS version, fixing crashes on older systems like Ventura that left the overlay stuck on "Overlay loading".
+  - mpv IPC connection attempts time out and retry, showing an actionable error if content still isn't ready after 30 seconds.
+  - Dedicated overlay modals are prewarmed on macOS and Windows so shortcuts open them promptly.
+  - On macOS, reused modals and the stats window open above fullscreen mpv on its current Space instead of jumping to another desktop.
+- **Wayland File Drop**: Fixed native Wayland drag-and-drop from file managers such as Thunar, so subtitle and video files dropped on the visible overlay are resolved and forwarded to mpv.
+- **Windows Mouse Lag**: Fixed system-wide mouse lag on Windows while SubMiner is running, caused by the overlay's global mouse hook for click-through forwarding and by the mpv window tracker blocking the app on repeated PowerShell lookups.
+- **Sentence Mining Audio & Clips**: Sentence-audio generation no longer times out on slow network-mounted media with many subtitle/font streams (bounded FFmpeg probing, two-minute extraction budget, clearer error reporting), and mined audio/animated AVIF clips now capture the subtitle line that was actually mined by snapshotting the clip range at lookup time instead of reading live mpv state later.
+- **Stats Performance & Reliability**: Immersion stats storage now sets its SQLite busy timeout before WAL setup, avoiding transient lock errors under concurrent writes. Deletes in the stats dashboard no longer freeze the UI, run proportional to what's deleted instead of rebuilding full lifetime summaries, retry safely if the delete worker crashes, and no longer rescan the whole library when deleting very common words; a new index also makes large session deletes drop from minutes to milliseconds. Library merges, video moves, and AniList reassignments got the same lifetime-summary fix.
+- **Vocabulary Stats Accuracy**: Vocabulary totals and charts now count all tracked vocabulary instead of only the first page, new-word history uses corrected daily rollups (fixing legacy timestamp and time-zone issues), summary cards refresh automatically after edits to the exclusion list, and rapid exclusion edits no longer race each other.
+- **Rofi MKV Thumbnails**: Fixed missing MKV thumbnails in the Linux rofi picker when system thumbnailer registrations only advertise legacy Matroska MIME aliases.
+
+<details>
+<summary>Internal changes</summary>
+
+**Internal**
+- Docs Site Indexing: Excluded the `/main/` and `/v/<version>/` docs trees from search indexing (self-referential canonical, `noindex,follow`, matching `X-Robots-Tag`) so crawlers focus on current docs instead of ~30 archived copies of every page, and restored `<lastmod>` dates in the docs sitemap that were silently dropped by production builds.
+
+</details>
+
 ## v0.19.3 (2026-08-13)
 
 **Added**

docs-site/configuration.md

@@ -148,12 +148,13 @@ The configuration file includes several main sections:
 
 - [**Shared AI Provider**](#shared-ai-provider) - Canonical OpenAI-compatible provider config shared by Anki and YouTube subtitle fixing
 - [**AnkiConnect**](#ankiconnect) - Automatic Anki card creation with media
-- [**Kiku/Lapis Integration**](#kiku-lapis-integration) - Sentence cards and duplicate handling for Kiku/Lapis note types
+- [**Kiku/Lapis Integration**](#kiku-lapis-integration) - Sentence cards and duplicate handling for Kiku/Lapis/Senren note types
 - [**N+1 Word Highlighting**](#n-1-word-highlighting) - Known-word cache and single-target highlighting
-- [**Field Grouping Modes**](#field-grouping-modes) - Kiku/Lapis duplicate card merging
+- [**Field Grouping Modes**](#field-grouping-modes) - Kiku/Senren duplicate card merging
 
 **External Integrations**
 
+- [**Anime Browser**](#anime-browser) - Extension repositories and stream preferences for the anime browser
 - [**Jimaku**](#jimaku) - Jimaku API configuration and defaults
 - [**TsukiHime**](#tsukihime) - Multi-language subtitle search and download
 - [**Subtitle Sync**](#subtitle-sync) - Sync current subtitle with `alass`/`ffsubsync`
@@ -592,6 +593,7 @@ See `config.example.jsonc` for detailed configuration options and more examples.
 | `KeyJ`                  | `["cycle", "sid"]`            | Cycle primary subtitle track            |
 | `Shift+KeyJ`            | `["cycle", "secondary-sid"]`  | Cycle secondary subtitle track          |
 | `Ctrl+Alt+KeyP`         | `["__playlist-browser-open"]` | Open playlist browser                   |
+| `Ctrl+Alt+KeyA`         | `["__anime-browser-open"]`    | Toggle Anime Browser in the player      |
 | `Ctrl+Alt+KeyC`         | `["__youtube-picker-open"]`   | Open the manual YouTube subtitle picker |
 | `ArrowRight`            | `["seek", 5]`                 | Seek forward 5 seconds                  |
 | `ArrowLeft`             | `["seek", -5]`                | Seek backward 5 seconds                 |
@@ -633,7 +635,7 @@ See `config.example.jsonc` for detailed configuration options and more examples.
 { "key": "Space", "command": null }
 ```
 
-**Special commands:** Commands prefixed with `__` are handled internally by the overlay rather than sent to mpv. `__playlist-browser-open` opens the split-pane playlist browser for the current file's parent directory and the live mpv queue. `__replay-subtitle` replays the current subtitle and pauses at its end. `__play-next-subtitle` seeks to the next subtitle, plays it, and pauses at its end. `__runtime-options-open` opens the runtime options palette. `__runtime-option-cycle:<id>[:next|prev]` cycles a runtime option value.
+**Special commands:** Commands prefixed with `__` are handled internally by the overlay rather than sent to mpv. `__playlist-browser-open` opens the split-pane playlist browser for the current file's parent directory and the live mpv queue. `__anime-browser-open` opens the Anime Browser inside the player bounds. `__replay-subtitle` replays the current subtitle and pauses at its end. `__play-next-subtitle` seeks to the next subtitle, plays it, and pauses at its end. `__runtime-options-open` opens the runtime options palette. `__runtime-option-cycle:<id>[:next|prev]` cycles a runtime option value.
 
 **Supported commands:** Any valid mpv JSON IPC command array (`["cycle", "pause"]`, `["seek", 5]`, `["script-binding", "..."]`, etc.)
 
@@ -1051,6 +1053,7 @@ This example is intentionally compact. The option table below documents availabl
 | `metadata.pattern`                                | string                                      | Format pattern for metadata: `%f`=filename, `%F`=filename+ext, `%t`=time, `%T`=time with milliseconds, `<br>`=newline                                                                                                           |
 | `isLapis`                                         | object                                      | Lapis/shared sentence-card config: `{ enabled, sentenceCardModel }`. Sentence/audio field names are fixed to `Sentence` and `SentenceAudio`.                                                                                    |
 | `isKiku`                                          | object                                      | Kiku-only config: `{ enabled, fieldGrouping, deleteDuplicateInAuto }` (shared sentence/audio/model settings are inherited from `isLapis`)                                                                                       |
+| `isSenren`                                        | object                                      | Senren-only config: `{ enabled, fieldGrouping, deleteDuplicateInAuto }`. Merges duplicates using Senren's scene-switching markup. Mutually exclusive with `isKiku.enabled`.                                                     |
 
 `ankiConnect.ai` only controls feature-local enablement plus optional `model` / `systemPrompt` overrides.
 API key resolution, base URL, and timeout live under the shared top-level [`ai`](#shared-ai-provider) config.
@@ -1080,6 +1083,7 @@ SubMiner is intentionally built for [Kiku](https://kiku.youyoumu.my.id/) and [La
 - Enable `isKiku` to turn on duplicate merge behavior for mined Word/Expression hits.
 - When both are enabled, Kiku behavior is applied for grouping while sentence-card model settings are still read from `isLapis`.
 - `isKiku.fieldGrouping` supports `disabled`, `auto`, and `manual` merge modes; see [Field Grouping Modes](#field-grouping-modes).
+- For [Senren](https://github.com/BrenoAqua/Senren) note types, enable `isSenren` instead of `isKiku`. Duplicate merges then use Senren's scene-switching markup (including grouped `miscInfo` entries), and `isSenren.fieldGrouping` supports the same three modes (default: `auto`). Kiku and Senren are mutually exclusive; if both are enabled, Kiku wins and Senren is turned off with a config warning.
 - `lapisKiku.wordCardKind` picks the card-type flag set on word cards; see [Word Card Type](#word-card-type). It is read only while `isLapis` or `isKiku` is enabled.
 
 ### Word Card Type
@@ -1154,6 +1158,35 @@ When the manual merge popup opens, SubMiner pauses playback and closes any open
 
 ## External Integrations
 
+### Anime Browser
+
+Sources for the [anime browser](/anime-browser). SubMiner ships no extension repositories and bundles no sources, so these are empty until you add one:
+
+```json
+{
+  "anime": {
+    "autoOpenJimaku": false,
+    "extensionsDir": "",
+    "repos": [],
+    "preferredQuality": "",
+    "defaultSource": ""
+  }
+}
+```
+
+| Option                   | Type       | Default | Description                                                                                                                                                                                                                                         |
+| ------------------------ | ---------- | ------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `anime.autoOpenJimaku`   | `boolean`  | `false` | Pause Anime Browser playback and open Jimaku when an episode loads. Playback resumes after a subtitle loads or the modal closes. Playback that was already paused stays paused.                                                                     |
+| `anime.extensionsDir`    | `string`   | `""`    | Directory holding Aniyomi extension `.apk` files. Empty uses `<userData>/anime-extensions`.                                                                                                                                                         |
+| `anime.repos`            | `string[]` | `[]`    | Extension repository index URLs. Any `https` URL ending in `.json` works; `index.min.json` is only the common name.                                                                                                                                 |
+| `anime.preferredQuality` | `string`   | `""`    | Preferred stream quality label, matched as a substring (for example `1080`). Empty keeps the source's own order. A Japanese-audio entry always outranks a higher-quality dub.                                                                       |
+| `anime.defaultSource`    | `string`   | `""`    | Source the Anime Browser selects when it opens: a source id (`<package>:<source>`) or `all` for every installed source. Empty selects the first installed source. **Set default** in the Extensions tab writes this value.                          |
+| `anime.bridgeDir`        | `string`   | `""`    | Directory holding an M-Extension-Server bundle (Java runtime plus server jar) to run instead of the downloaded copy. Empty checks the package-manager install first, then `<userData>/anime-bridge`. See [the bridge](anime-browser.md#the-bridge). |
+
+Repositories added from the browser's Extensions tab are written back to `anime.repos`, so the list can also be kept in a dotfile. Changes apply the next time the anime browser opens.
+
+Per-source settings (server addresses, credentials, per-extension quality or language options) are not part of `config.jsonc`. They belong to the extension, are edited in the browser's **Source settings** tab, and persist in `<userData>/anime-source-preferences.json` with owner-only permissions.
+
 ### Jimaku
 
 Configure Jimaku API access and defaults:
@@ -1214,11 +1247,15 @@ Sync a subtitle track from the overlay picker using `alass` or `ffsubsync`. The
 
 | Option           | Values          | Description                                                                                                               |
 | ---------------- | --------------- | ------------------------------------------------------------------------------------------------------------------------- |
-| `alass_path`     | string path     | Path to `alass` executable. Empty falls back to `/usr/bin/alass`. `alass` must be installed separately.                   |
-| `ffsubsync_path` | string path     | Path to `ffsubsync` executable. Empty falls back to `/usr/bin/ffsubsync`. `ffsubsync` must be installed separately.       |
-| `ffmpeg_path`    | string path     | Path to `ffmpeg` (used for internal subtitle extraction). Empty or `null` falls back to `/usr/bin/ffmpeg`.                |
+| `alass_path`     | string path     | Path to `alass` executable. Empty auto-discovers `alass` or `alass-cli`. `alass` must be installed separately.            |
+| `ffsubsync_path` | string path     | Path to `ffsubsync` executable. Empty auto-discovers `ffsubsync`. `ffsubsync` must be installed separately.               |
+| `ffmpeg_path`    | string path     | Path to `ffmpeg` (used for internal subtitle extraction). Empty or `null` auto-discovers `ffmpeg`.                        |
 | `replace`        | `true`, `false` | When `true` (default), overwrite the active subtitle file on successful sync. When `false`, write `<name>_retimed.<ext>`. |
 
+Auto-discovery searches `PATH`, then the usual install prefixes (`/opt/homebrew/bin`, `/usr/local/bin`, `/opt/local/bin`, `/usr/bin`, `/bin`) — a GUI launch inherits a minimal `PATH` that often omits the first two. Set the option explicitly if your binary lives elsewhere.
+
+Subtitle tracks that mpv loaded from a URL (Aniyomi extension streams, Jellyfin) are downloaded to a temporary file first, reusing mpv's own request headers, so they can be used as either the sync target or the alass reference.
+
 Stats dashboard sentence mining also uses `alass_path` when available to align a local English sidecar against the local Japanese sidecar before filling the card translation field. This stats-only retime writes a temporary cached copy and never edits the original subtitle files.
 
 Default trigger is `Ctrl+Alt+S` via `shortcuts.triggerSubsync`.

docs-site/demos.md

@@ -5,7 +5,7 @@ Short recordings of SubMiner's key features and integrations from real playback
 <script setup>
 import { withBase } from 'vitepress';
 
-const v = '20260301-1';
+const v = '20260819-1';
 </script>
 
 ## Anki Card Mining & Enrichment

docs-site/development.md

@@ -6,7 +6,7 @@ For internal architecture/workflow guidance, use `docs/README.md` at the repo ro
 
 - [Bun](https://bun.sh)
 - A system `lua` interpreter for `bun run test:launcher` / `bun run test:plugin:src`
-- macOS builds compile a Swift helper via `scripts/build-macos-helper.sh` (skip with `SUBMINER_SKIP_MACOS_HELPER_BUILD=1`)
+- macOS builds compile a Swift helper via `scripts/prepare-build-assets.mjs` (skip with `SUBMINER_SKIP_MACOS_HELPER_BUILD=1`)
 
 ## Setup
 

docs-site/immersion-tracking.md

@@ -34,7 +34,7 @@ The same immersion data powers the stats dashboard.
 - In-app overlay: focus the visible overlay, then press the key from `stats.toggleKey` (default: `` ` `` / `Backquote`).
 - Launcher command: run `subminer stats` to start the local stats server on demand (it also opens the dashboard in your browser when `stats.autoOpenBrowser` is enabled; the default is `false`).
 - Background server: run `subminer stats -b` to start or reuse a dedicated background stats daemon without keeping the launcher attached, and `subminer stats -s` to stop that daemon.
-- Maintenance commands: run `subminer stats cleanup` or `subminer stats cleanup -v` to backfill/repair vocabulary metadata (`headword`, `reading`, POS) and purge stale or excluded rows from `imm_words` on demand; `subminer stats cleanup -l` repairs lifetime summary tables. `subminer stats rebuild` and `subminer stats backfill` rebuild or backfill rollup data.
+- Maintenance commands: run `subminer stats cleanup` or `subminer stats cleanup -v` to backfill/repair vocabulary metadata (`headword`, `reading`, POS) and purge stale or excluded rows from `imm_words` on demand; `subminer stats cleanup -l` repairs lifetime summary tables non-destructively (recomputed from per-episode history, so lifetime totals older than the session retention window are kept); `subminer stats cleanup --duplicate-lines` collapses repeated lines left behind by typeset subtitles (see [Repeated Line Cleanup](#repeated-line-cleanup)). `subminer stats rebuild` and `subminer stats backfill` rebuild or backfill rollup data.
 - Browser page: open `http://127.0.0.1:6969` directly if the local stats server is already running.
 
 ### Dashboard Tabs
@@ -57,6 +57,13 @@ Jellyfin stream URLs are normalized to stable item links before stats titles are
 
 When YouTube channel metadata is available, the Library tab groups videos by creator/channel and treats each tracked video as an episode-like entry inside that channel section.
 
+A library entry is identified by its parsed title plus any detected season, so the same show can end up on several cards when releases disagree about the title or omit the season tag. Two fixes are available:
+
+- **Merge duplicates.** Hit **Select** above the grid, tick the cards that are the same show, and choose **Merge Selected**. Pick which entry to keep in the dialog; every episode moves onto it and the other cards are removed. Nothing is deleted, so sessions, mined cards and watch time all carry over. SubMiner remembers the merged title variants, so future episodes parsed with one of those names join the kept entry instead of recreating a duplicate card.
+- **Move a single episode.** Hover an episode row in a title's episode list and use the **→** button to reassign it to another library entry. The correction is remembered, so later filename parsing or Jellyfin metadata cannot move that episode back. For local files, later episodes in the same directory inherit the correction when their detected seasons are compatible and every manual correction there points to the same entry; a file that parses to a title which already has its own library entry keeps that identity instead. Conflicting seasons or manual destinations are left for review. If the move empties the old entry, that card is removed and you are returned to the grid.
+
+Once cover art resolves a series to an AniList entry, cards with compatible seasons are folded together automatically only when the searched title exactly matches an AniList title or synonym. A fuzzy result that points at an AniList entry already used by another card appears as a **Possible duplicate** review above the Library grid instead. Choose **Review merge** to compare the cards and pick which one to keep, or **Not duplicates** to dismiss that suggestion permanently. Entries with conflicting explicit season numbers are left alone rather than merged or suggested.
+
 Open a title and use **Delete Entry** in its header to remove a mistakenly tracked show outright. This deletes every episode of that title along with their sessions, subtitle lines, rollups and cover art, drops the words and kanji that were only seen there, and removes the card from the Library grid. Individual episodes and sessions can still be deleted on their own from the episode list and session rows. Entry deletion is refused while that title is the one currently playing.
 
 ![Stats Library](/screenshots/stats-library.png)
@@ -75,7 +82,7 @@ Expandable session history with new-word activity, cumulative totals, and pause/
 
 #### Vocabulary
 
-Top repeated words (click a bar to open the word), new-word timeline, cross-title and frequency rank tables with Hide Known / Hide Kana filters, kanji breakdown, word exclusion list, and click-through occurrence drilldown with Mine Word / Mine Sentence / Mine Audio buttons.
+The summary cards show all unique vocabulary and kanji recorded in the local tracking database; **New This Week** is the only weekly figure and uses a rolling seven-day window. The word and kanji tables load first while those complete totals calculate separately. Top Repeated Words and New Words by Day use complete tracking history rather than the table's browsing page. New-word history is maintained as a permanent daily lexical rollup using the same token-visibility rules as the totals, including normalization of older timestamps stored in either seconds or milliseconds and retroactive corrections when tracked material is removed or reprocessed. On the first launch after an applicable upgrade, that history is version-rebuilt in the background and the chart refreshes when it is ready; if it remains unavailable, polling stops and an inline Retry control appears. The cards and charts also refresh automatically after the word exclusion list changes. The rest of the tab includes cross-title and frequency rank tables with Hide Known / Hide Kana filters, kanji breakdown, word exclusion list, and click-through occurrence drilldown with Mine Word / Mine Sentence / Mine Audio buttons.
 
 ![Stats Vocabulary](/screenshots/stats-vocabulary.png)
 
@@ -125,6 +132,36 @@ Secondary subtitle text (typically English translations) is stored alongside pri
 
 The Vocabulary tab toolbar includes an **Exclusions** button for hiding words from all vocabulary views. Excluded words are stored in the immersion database, with older browser localStorage exclusions imported on first load after upgrade. They can be managed (restored or cleared) from the exclusion modal. Exclusions affect stat cards, charts, the frequency rank table, and the word list.
 
+### Repeated Line Cleanup
+
+Karaoke openings and animated signs are authored as one subtitle event per animation frame, all carrying the same text. Playback reports every one of those frames, so a single OP lyric could be recorded hundreds of times and dominate "Top Repeated Words".
+
+Recording now collapses those runs as they happen, matching what the subtitle sidebar shows:
+
+- When a typeset ASS file stores a clean lyric or sign in a timed authoring comment, or in full-line events surrounding generated fragments, the matching complete line is recorded once. The repeated glyph or clip-animation frames are not recorded. Dialogue spoken while such an animation is on screen records as itself, without the fragment lines beside it.
+- When karaoke styling redraws the same complete lyric across consecutive color or highlight phases, those phases are combined into one line with their full timing. Repeated ordinary dialogue remains separate.
+- When the active subtitle source has been parsed, its cue list has already had duplicate events and animation bursts merged. A line landing inside a surviving cue but after that cue's start is a frame the sidebar merged away, and is not recorded.
+- When no parsed cue covers the live timing, including while a subtitle source is changing or shifted, the strict metadata-free rule applies: a run of identical, contiguous lines each shorter than 0.1s stops being recorded after a few frames. Runs are tracked per line of text, so dual-line karaoke (a kanji and a romaji line frame-flipped together) collapses both lines. Ordinary repeated dialogue, and lines held for a normal beat, always record.
+
+For stats recorded before this, the Vocabulary tab toolbar has a **Duplicates** button:
+
+- Pick how far back to look (7 days, 30 days, 90 days, 1 year, or all time). A narrower window does less work and keeps older history untouched.
+- **Scan** reports the bursts found, the lines they added, and the word and kanji counts they inflated, without writing anything.
+- **Clean Up** applies exactly what the scan reported: each run collapses to its first line (extended to cover the run), and the removed lines' word and kanji occurrences are subtracted from the vocabulary aggregates.
+
+The same thing runs from the terminal:
+
+```bash
+subminer stats cleanup --duplicate-lines --dry-run --lookback-days 30
+subminer stats cleanup --duplicate-lines --lookback-days 30
+```
+
+`--duplicate-lines` (short: `-d`) picks the cleanup mode, so it cannot be combined with `--vocab` or `--lifetime`, and `--dry-run` and `--lookback-days <days>` only apply to it. Omitting `--lookback-days` scans all history; the value must be at least one day.
+
+The cleanup chains runs per line of text, so interleaved dual-line karaoke collapses each of its lines. It also removes the short residue the live rule stores before a run is long enough to recognize: a run one frame short of the usual minimum qualifies when every event is under the strict 0.1s bound.
+
+Runs never cross a session boundary, so rewatching an episode keeps both watches. Session telemetry (watch time, lines seen, tokens seen) and the rollups derived from it are left as recorded: they are cumulative samples taken during playback, and cannot be recomputed for sessions whose raw rows have since been pruned.
+
 ## Retention Defaults
 
 By default, SubMiner keeps all retention tables and raw data (`0` means keep all) while continuing daily/monthly rollup maintenance:
@@ -145,6 +182,7 @@ In practice:
 - Anime and episode pages keep lifetime totals from summary tables while session drill-down still reads retained sessions directly. With the current defaults, both are kept forever.
 - Trends can read the full available history because daily/monthly rollups are also kept forever by default.
 - Vocabulary and kanji totals are cumulative and not bounded by the raw session retention knobs.
+- New-word charts use their own permanent lexical daily rollups, which are not pruned by activity-rollup retention.
 
 ## Storage / Performance Model
 
@@ -314,6 +352,7 @@ Rollup tables:
 
 - `imm_daily_rollups`
 - `imm_monthly_rollups`
+- `imm_lexical_daily_rollups` - permanent first-discovery counts for vocabulary and kanji chart history
 - `imm_rollup_state` - incremental rollup progress bookkeeping
 
 Vocabulary tables:

docs-site/index.md

@@ -88,7 +88,7 @@ features:
 <script setup>
 import { withBase } from 'vitepress';
 
-const demoAssetVersion = '20260223-2';
+const demoAssetVersion = '20260819-1';
 </script>
 
 <div class="landing-shell">

docs-site/installation.md

@@ -215,6 +215,10 @@ Install [`subminer-bin`](https://aur.archlinux.org/packages/subminer-bin) from t
 paru -S subminer-bin
 ```
 
+For the [anime browser](anime-browser.md#the-bridge), optionally add
+`mangatan-extension-server`. SubMiner picks the package up instead of
+downloading its own copy of the bridge, and pacman keeps it updated.
+
 Or manually:
 
 ```bash
@@ -392,7 +396,7 @@ subminer -u
 subminer --update
 ```
 
-SubMiner verifies AppImage, launcher, and Linux support-asset downloads against `SHA256SUMS.txt`. On Linux those support assets include the launcher-managed runtime plugin copy under `SubMiner/plugin/subminer` plus the rofi theme at `SubMiner/themes/subminer.rasi`. If the binary is in a protected path, SubMiner shows the exact command to run rather than elevating itself.
+SubMiner verifies AppImage, launcher, and Linux support-asset downloads against `SHA256SUMS.txt`. On Linux those support assets include the launcher-managed runtime plugin copy under `SubMiner/plugin/subminer`, the rofi theme at `SubMiner/themes/subminer.rasi`, and the scoped Matroska thumbnailer registration under `SubMiner/thumbnailers`. If the binary is in a protected path, SubMiner shows the exact command to run rather than elevating itself.
 
 The tray "Check for Updates" entry installs the new app automatically on Linux, macOS, and Windows. On Linux it replaces the running `.AppImage` in place via `electron-updater` and refreshes the managed support assets from `subminer-assets.tar.gz`; AppImages managed by a system package (for example the AUR `/opt/SubMiner/SubMiner.AppImage`) are skipped so the package manager stays in charge.
 
@@ -404,7 +408,7 @@ SubMiner is an overlay that sits on top of mpv. It connects to mpv through an IP
 
 The `subminer` launcher handles mpv IPC socket setup automatically. If you launch mpv yourself or from another tool, you must pass `--input-ipc-server=/tmp/subminer-socket` (or `\\.\pipe\subminer-socket` on Windows) - without it the overlay starts but subtitles won't appear.
 
-The bundled mpv plugin is injected at runtime automatically - you don't need to install it separately. On Linux, the `subminer` launcher now checks for its managed runtime plugin copy and rofi theme before every mpv-managed launch and installs those support assets from the bundled app automatically if either one is missing. It provides in-player keybindings (the `y` chord) for controlling the overlay from within mpv. See [MPV Plugin](/mpv-plugin) for the full keybinding and configuration reference.
+The bundled mpv plugin is injected at runtime automatically - you don't need to install it separately. On Linux, the `subminer` launcher checks for its managed runtime plugin copy, rofi theme, and scoped thumbnailer registration before every mpv-managed launch and installs those support assets from the bundled app automatically if one is missing. For a rofi picker launch, this check runs before the picker opens. It provides in-player keybindings (the `y` chord) for controlling the overlay from within mpv. See [MPV Plugin](/mpv-plugin) for the full keybinding and configuration reference.
 
 ## Platform Notes
 
@@ -456,18 +460,20 @@ sudo chmod +x /usr/local/bin/subminer
 
 ### Linux Support Assets
 
-SubMiner ships the Linux rofi theme plus the launcher-managed runtime plugin copy in `subminer-assets.tar.gz`:
+SubMiner ships the Linux rofi theme, scoped Matroska thumbnailer registration, and launcher-managed runtime plugin copy in `subminer-assets.tar.gz`:
 
 ```bash
 wget https://github.com/ksyasuda/SubMiner/releases/latest/download/subminer-assets.tar.gz -O /tmp/subminer-assets.tar.gz
 tar -xzf /tmp/subminer-assets.tar.gz -C /tmp
 mkdir -p ~/.local/share/SubMiner/themes
 cp /tmp/assets/themes/subminer.rasi ~/.local/share/SubMiner/themes/subminer.rasi
+mkdir -p ~/.local/share/SubMiner/thumbnailers
+cp /tmp/assets/thumbnailers/subminer-ffmpegthumbnailer.thumbnailer ~/.local/share/SubMiner/thumbnailers/
 mkdir -p ~/.local/share/SubMiner/plugin
 cp -R /tmp/plugin/subminer ~/.local/share/SubMiner/plugin/subminer
 ```
 
-`subminer -u` and the tray updater keep those Linux support assets in sync automatically once the `SubMiner` data dir exists. Normal Linux launcher playback also auto-installs the managed runtime plugin copy and rofi theme from the bundled app if either support asset is missing, so manual extraction is mainly useful for pre-seeding or custom setups.
+`subminer -u` and the tray updater keep those Linux support assets in sync automatically once the `SubMiner` data dir exists. Normal Linux launcher playback also auto-installs all three assets from the bundled app if one is missing, so manual extraction is mainly useful for pre-seeding or custom setups. Rofi receives the SubMiner data path through its process-local `XDG_DATA_DIRS`, so the thumbnailer registration does not change the desktop-wide configuration.
 
 Override the theme path with `SUBMINER_ROFI_THEME=/absolute/path/to/theme.rasi`.
 

docs-site/jellyfin-integration.md

@@ -54,7 +54,7 @@ From then on, pause / resume / seek / stop and audio or subtitle track changes y
 - **Resume works.** If Jellyfin has a saved position for the item, SubMiner seeks there on load.
 - **Direct play first.** When the source allows it and the container is in your direct-play allowlist, SubMiner streams the original file; otherwise it requests a transcoded stream from Jellyfin.
 - **Japanese subtitles are auto-selected,** preferring Jellyfin's default and embedded tracks over external sidecar files when several match.
-- **Subtitle timing is corrected when possible.** SubMiner removes Jellyfin's server-selected subtitle stream from the mpv load URL, suppresses the mpv plugin's one-shot subtitle auto-selection and overlay auto-start for managed Jellyfin loads, stages downloaded subtitle tracks without letting mpv auto-switch between tracks, then selects the Japanese track once after applying any saved or inferred timing delay. When Jellyfin provides both Japanese and English subtitle files, SubMiner compares their cue timelines and applies a global delay if one track is clearly offset. Manual delay shifts you make with SubMiner's adjacent-cue controls are saved per item and subtitle track, then restored the next time you select that track.
+- **Downloaded subtitles keep their original timing.** SubMiner removes Jellyfin's server-selected subtitle stream from the mpv load URL, suppresses the mpv plugin's one-shot subtitle auto-selection and overlay auto-start for managed Jellyfin loads, stages the subtitle files exposed by Jellyfin without letting mpv auto-switch between tracks, resets mpv's subtitle delay to zero, then selects the Japanese track. SubMiner does not compare Japanese and English cue timelines or save an inferred delay.
 
 ## Settings
 

docs-site/jimaku-integration.md

@@ -23,12 +23,12 @@ If no files match the current episode filter, a "Show all files" button lets you
 
 ### Modal Keyboard Shortcuts
 
-| Key | Action |
-| --- | --- |
-| `Enter` (in text field) | Search |
-| `Enter` (in list) | Select entry / download file |
-| `Arrow Up` / `Arrow Down` | Navigate entries or files |
-| `Escape` | Close modal |
+| Key                       | Action                       |
+| ------------------------- | ---------------------------- |
+| `Enter` (in text field)   | Search                       |
+| `Enter` (in list)         | Select entry / download file |
+| `Arrow Up` / `Arrow Down` | Navigate entries or files    |
+| `Escape`                  | Close modal                  |
 
 ## Configuration
 
@@ -41,26 +41,26 @@ Add a `jimaku` section to your `config.jsonc`:
     "apiKeyCommand": "cat ~/.jimaku_key",
     "apiBaseUrl": "https://jimaku.cc",
     "languagePreference": "ja",
-    "maxEntryResults": 10
-  }
+    "maxEntryResults": 10,
+  },
 }
 ```
 
-| Option | Type | Default | Description |
-| --- | --- | --- | --- |
-| `jimaku.apiKey` | `string` | - | Jimaku API key (plaintext). Mutually exclusive with `apiKeyCommand`. |
-| `jimaku.apiKeyCommand` | `string` | - | Shell command that prints the API key to stdout. Useful for secret managers (e.g., `pass jimaku/api-key`). |
-| `jimaku.apiBaseUrl` | `string` | `"https://jimaku.cc"` | Base URL for the Jimaku API. Only change this if using a mirror or local instance. |
-| `jimaku.languagePreference` | `"ja"` \| `"en"` \| `"none"` | `"ja"` | Sort subtitle files by language tag. `"ja"` pushes Japanese-tagged files to the top; `"en"` does the same for English. `"none"` preserves the API order. |
-| `jimaku.maxEntryResults` | `number` | `10` | Maximum number of anime entries returned per search. |
+| Option                      | Type                         | Default               | Description                                                                                                                                              |
+| --------------------------- | ---------------------------- | --------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `jimaku.apiKey`             | `string`                     | -                     | Jimaku API key (plaintext). Mutually exclusive with `apiKeyCommand`.                                                                                     |
+| `jimaku.apiKeyCommand`      | `string`                     | -                     | Shell command that prints the API key to stdout. Useful for secret managers (e.g., `pass jimaku/api-key`).                                               |
+| `jimaku.apiBaseUrl`         | `string`                     | `"https://jimaku.cc"` | Base URL for the Jimaku API. Only change this if using a mirror or local instance.                                                                       |
+| `jimaku.languagePreference` | `"ja"` \| `"en"` \| `"none"` | `"ja"`                | Sort subtitle files by language tag. `"ja"` pushes Japanese-tagged files to the top; `"en"` does the same for English. `"none"` preserves the API order. |
+| `jimaku.maxEntryResults`    | `number`                     | `10`                  | Maximum number of anime entries returned per search.                                                                                                     |
 
 The keyboard shortcut is configured separately under `shortcuts`:
 
 ```jsonc
 {
   "shortcuts": {
-    "openJimaku": "Ctrl+Shift+J"
-  }
+    "openJimaku": "Ctrl+Shift+J",
+  },
 }
 ```
 
@@ -79,6 +79,8 @@ SubMiner extracts media info from the current video path to pre-fill the search
 
 - **Season + episode patterns:** `S01E03`, `1x03`
 - **Episode-only patterns:** `E03`, `EP03`, or dash-separated numbers like `Title - 03 -`
+- **Spelled-out episode labels:** `Episode 4`, `第4話`
+- **Season named at the end of the title:** `… Season 3`, `… 3rd Season`, `… S3`, `… 第3期` - the season goes in the Season field instead of being searched for as part of the series name
 - **Season folders:** a parent directory named `Season 2` or `S2` fills in the season when the filename lacks one
 - **Bracket tags:** `[SubGroup]`, `[1080p]`, `[HEVC]` - stripped before title extraction
 - **Year tags:** `(2024)` - stripped
@@ -87,6 +89,8 @@ SubMiner extracts media info from the current video path to pre-fill the search
 
 If the parser produces a high-confidence result (title + episode both detected), the search runs automatically when the modal opens. Otherwise, you can adjust the fields manually before searching.
 
+Episodes launched from the [anime browser](/anime-browser) skip parsing entirely: Title, Season, and Episode come from the source's own listing, which is why the search runs immediately even though the stream URL says nothing about the episode.
+
 ## Troubleshooting
 
 **"Jimaku API key not set"**

docs-site/launcher-script.md

@@ -34,18 +34,22 @@ subminer -R -r -d ~/Anime              # rofi picker, recursive
 subminer -R /directory                 # rofi picker, directory shortcut
 ```
 
-rofi shows a GUI menu with icon thumbnails when available. SubMiner ships the rofi theme plus the Linux launcher-managed runtime plugin copy in the release assets tarball:
+rofi shows a GUI menu with icon thumbnails when available. SubMiner ships the rofi theme, a scoped `ffmpegthumbnailer` MIME registration, and the Linux launcher-managed runtime plugin copy in the release assets tarball:
 
 ```bash
 wget https://github.com/ksyasuda/SubMiner/releases/latest/download/subminer-assets.tar.gz -O /tmp/subminer-assets.tar.gz
 tar -xzf /tmp/subminer-assets.tar.gz -C /tmp
 mkdir -p ~/.local/share/SubMiner/themes
 cp /tmp/assets/themes/subminer.rasi ~/.local/share/SubMiner/themes/subminer.rasi
+mkdir -p ~/.local/share/SubMiner/thumbnailers
+cp /tmp/assets/thumbnailers/subminer-ffmpegthumbnailer.thumbnailer ~/.local/share/SubMiner/thumbnailers/
 mkdir -p ~/.local/share/SubMiner/plugin
 cp -R /tmp/plugin/subminer ~/.local/share/SubMiner/plugin/subminer
 ```
 
-Once the `SubMiner` data dir exists, `subminer -u` refreshes both assets automatically. Normal Linux launcher playback also checks for the managed runtime plugin copy and rofi theme before mpv launch and installs them from the bundled app automatically if either one is missing.
+Once the `SubMiner` data dir exists, `subminer -u` refreshes these assets automatically. Normal Linux launcher playback checks for all three assets and installs them from the bundled app when one is missing. For `subminer -R`, this repair runs before rofi opens.
+
+When `ffmpegthumbnailer` is installed, SubMiner prepends its own data directory to `XDG_DATA_DIRS` for the rofi process only. This lets rofi recognize the canonical Matroska MIME types used by newer GLib versions without changing the desktop-wide MIME or thumbnailer configuration. An existing registration in your own `$XDG_DATA_HOME/thumbnailers` still takes priority.
 
 The theme is auto-detected from these paths (first match wins):
 
@@ -151,6 +155,7 @@ subminer stats -b                       # start background stats daemon
 | `subminer stats`                           | Start the stats server (opens the dashboard when `stats.autoOpenBrowser` is on)                   |
 | `subminer stats -b` / `-s`                 | Start/reuse or stop the background stats daemon                                                   |
 | `subminer stats cleanup`                   | Backfill vocabulary metadata and prune stale rows (`-v` vocab, `-l` lifetime summaries)           |
+| `subminer stats cleanup -d`                | Collapse repeated lines from typeset subs (`--dry-run`, `--lookback-days <n>`)                    |
 | `subminer stats rebuild` / `backfill`      | Rebuild or backfill rollup data                                                                   |
 | `subminer doctor`                          | Dependency + config + socket diagnostics (`--refresh-known-words` refreshes the known-word cache) |
 | `subminer settings`                        | Open the SubMiner settings window                                                                 |

docs-site/links.test.ts

@@ -21,6 +21,7 @@ function slugify(heading: string): string {
 }
 
 const EXCLUDED_PAGES = new Set(['README.md']);
+const UNLISTED_ROUTES = new Set(['/demos']);
 const PUBLIC_PREFIXES = ['/assets/', '/screenshots/', '/config.example.jsonc', '/favicon'];
 
 function loadPages(): Map<string, string> {
@@ -114,7 +115,7 @@ test('slugify matches the VitePress cases these docs actually rely on', () => {
   expect(slugify('2. Install SubMiner')).toBe('_2-install-subminer');
 });
 
-test('every docs page is reachable from the sidebar', async () => {
+test('every docs page is reachable from the sidebar unless explicitly unlisted', async () => {
   const { default: config } = await import('./.vitepress/config');
   const sidebar = config.themeConfig?.sidebar as Array<{
     items?: Array<{ text: string; link?: string }>;
@@ -127,6 +128,8 @@ test('every docs page is reachable from the sidebar', async () => {
     }
   }
 
-  const orphans = [...pages.keys()].filter((route) => !linked.has(route));
+  const orphans = [...pages.keys()].filter(
+    (route) => !linked.has(route) && !UNLISTED_ROUTES.has(route),
+  );
   expect(orphans).toEqual([]);
 });

docs-site/mining-workflow.md

@@ -41,7 +41,7 @@ If you prefer a hands-on approach (animecards-style), you can copy the current s
    - For multiple lines: press `Ctrl/Cmd+Shift+C`, then a digit `1`–`9` to select how many recent subtitle lines to combine. The combined text is copied to the clipboard.
 3. Press `Ctrl/Cmd+V` to update the last-added card with the clipboard contents plus audio, image, and translation - the same fields auto-update would fill.
 
-Manual clipboard updates always replace generated sentence audio, even when `ankiConnect.behavior.overwriteAudio` is disabled. The word audio field is left unchanged because the word itself does not change in this flow.
+Manual clipboard updates always replace generated sentence audio in `ankiConnect.fields.audio`, even when `ankiConnect.behavior.overwriteAudio` is disabled. Normal word-card updates use the configured sentence and audio fields even when Lapis or Kiku support is enabled.
 
 This is useful when auto-update is disabled or when you want explicit control over which subtitle line gets attached to the card.
 
@@ -72,17 +72,17 @@ After adding a word via Yomitan, press the audio card shortcut (`Ctrl/Cmd+Shift+
 Audio card marking uses the same `ankiConnect.isLapis.sentenceCardModel` note type as sentence cards. See [Anki Integration - Sentence Cards](/anki-integration#sentence-cards-lapis) for setup.
 :::
 
-### Field Grouping (Kiku)
+### Field Grouping (Kiku/Senren)
 
-If you mine the same word from different sentences, SubMiner can merge the cards instead of creating duplicates. This feature is designed for use with [Kiku](https://github.com/youyoumu/kiku) and similar note types that support grouped fields.
+If you mine the same word from different sentences, SubMiner can merge the cards instead of creating duplicates. This feature is designed for use with [Kiku](https://github.com/youyoumu/kiku) and [Senren](https://github.com/BrenoAqua/Senren) note types that support grouped fields (Senren calls it scene switching).
 
 1. You add a word via Yomitan.
 2. SubMiner detects the new card and checks if a card with the same expression already exists.
-3. If a duplicate is found (this requires `ankiConnect.isKiku.fieldGrouping` to be set to `"auto"` or `"manual"`; it defaults to `"disabled"`):
-   - **Auto mode** (`ankiConnect.isKiku.fieldGrouping: "auto"`): Merges automatically. Both sentences, audio clips, and images are combined into the existing card. The duplicate is optionally deleted.
-   - **Manual mode** (`ankiConnect.isKiku.fieldGrouping: "manual"`): A modal appears showing both cards side by side. You choose which card to keep and preview the merged result before confirming.
+3. If a duplicate is found (this requires Kiku or Senren to be enabled with a field grouping mode of `"auto"` or `"manual"`):
+   - **Auto mode**: Merges automatically. Both sentences, audio clips, images, and source info are combined into the existing card. The duplicate is optionally deleted.
+   - **Manual mode**: A modal appears showing both cards side by side. You choose which card to keep and preview the merged result before confirming.
 
-See [Anki Integration - Field Grouping](/anki-integration#field-grouping-kiku) for configuration options, merge behavior, and modal keyboard shortcuts.
+See [Anki Integration - Field Grouping](/anki-integration#field-grouping-kiku-senren) for configuration options, merge behavior, and modal keyboard shortcuts.
 
 ## Overlay Model
 
@@ -108,8 +108,12 @@ The secondary bar is a compact top-strip region in the same overlay window. It s
 - Quick comprehension checks without leaving the mining flow.
 - Auto-populating the translation field on mined cards - when a card is created, SubMiner uses the secondary subtitle text as the translation field value (unless AI translation is configured to override it).
 
+For local media, SubMiner can parse supported embedded secondary tracks into timed cues. For remote URLs and files on network mounts, it uses mpv's live secondary subtitle text instead of scanning the media with ffmpeg.
+
 It is controlled by `secondarySub` configuration and shares its lifecycle with the main overlay window. Cycle which track feeds it with `Shift+J`.
 
+SubMiner collapses duplicate ASS layers in parsed secondary tracks. Exact repeated lines collapse at any length, while distinct simultaneous short lines remain separate. Long dialogue and positioned-sign copies also collapse when they differ only in whitespace or terminal punctuation. Dense multi-row sign layouts, such as translated timetables, are excluded instead of being concatenated into the secondary bar.
+
 ### Display Modes
 
 Both the primary and secondary subtitle bars share the same three visibility modes, and each can be changed independently at runtime:

docs-site/public/config.example.jsonc

@@ -341,6 +341,12 @@
         "__playlist-browser-open"
       ] // Command setting.
     },
+    {
+      "key": "Ctrl+Alt+KeyA", // Key setting.
+      "command": [
+        "__anime-browser-open"
+      ] // Command setting.
+    },
     {
       "key": "Ctrl+Shift+KeyH", // Key setting.
       "command": [
@@ -523,7 +529,7 @@
   // ==========================================
   // AnkiConnect Integration
   // Automatic Anki updates and media generation options.
-  // Hot-reload: ankiConnect.ai.enabled, media.normalizeAudio/mirrorMpvVolume, knownWords, nPlusOne, fields.word/audio/image/sentence/miscInfo, behavior.autoUpdateNewCards, isLapis.sentenceCardModel, isKiku.fieldGrouping, and lapisKiku.wordCardKind update live while SubMiner is running.
+  // Hot-reload: ankiConnect.ai.enabled, media.normalizeAudio/mirrorMpvVolume, knownWords, nPlusOne, fields.word/audio/image/sentence/miscInfo, behavior.autoUpdateNewCards, isLapis.sentenceCardModel, isKiku.fieldGrouping, isSenren.fieldGrouping, and lapisKiku.wordCardKind update live while SubMiner is running.
   // Shared AI provider transport settings are read from top-level ai and typically require restart.
   // Most other AnkiConnect settings still require restart.
   // ==========================================
@@ -606,11 +612,31 @@
       "fieldGrouping": "disabled", // Kiku duplicate-card field grouping mode. Values: auto | manual | disabled
       "deleteDuplicateInAuto": true // When Kiku field grouping is "auto", delete the duplicate source card after grouping completes. Values: true | false
     }, // Is kiku setting.
+    "isSenren": {
+      "enabled": false, // Enable Senren-specific duplicate handling (scene-switching field grouping, including miscInfo grouping). Mutually exclusive with isKiku.enabled. Values: true | false
+      "fieldGrouping": "auto", // Senren duplicate-card field grouping mode (scene switching). Values: auto | manual | disabled
+      "deleteDuplicateInAuto": true // When Senren field grouping is "auto", delete the duplicate source card after grouping completes. Values: true | false
+    }, // Is senren setting.
     "lapisKiku": {
       "wordCardKind": "word-and-sentence" // Card-type flag SubMiner marks on Kiku/Lapis word cards. Only one flag is set at a time; the others are cleared. Requires isKiku.enabled or isLapis.enabled. Values: word-and-sentence | click | sentence | audio | none
     } // Lapis kiku setting.
   }, // Automatic Anki updates and media generation options.
 
+  // ==========================================
+  // Anime Browser
+  // Anime browser sources. SubMiner ships no extension repositories and bundles no sources;
+  // add a repository index URL here (or drop .apk files in the extensions directory) to have any.
+  // Hot-reload: autoOpenJimaku applies to the next episode; other anime changes apply the next time the anime browser opens.
+  // ==========================================
+  "anime": {
+    "autoOpenJimaku": false, // Pause Anime Browser playback and open Jimaku when an episode loads. Playback resumes after a subtitle loads or the modal closes only when auto-open initiated the pause. Values: true | false
+    "extensionsDir": "", // Directory holding Aniyomi extension .apk files. Empty uses <userData>/anime-extensions.
+    "repos": [], // Extension repository index URLs (any https .json index, e.g. https://.../index.min.json). Empty by default; SubMiner ships no repositories.
+    "preferredQuality": "", // Preferred stream quality label, matched as a substring (for example: 1080). Empty uses the source order.
+    "defaultSource": "", // Source the Anime Browser selects when it opens: a source id (<package>:<source>) or "all" for every installed source. Empty selects the first installed source. "Set default" in the Extensions tab writes this value.
+    "bridgeDir": "" // Directory holding an M-Extension-Server bundle (java runtime plus server jar) to run instead of the copy SubMiner downloads. Empty checks the package-manager install (Arch: mangatan-extension-server), then <userData>/anime-bridge.
+  }, // Anime browser sources. SubMiner ships no extension repositories and bundles no sources;
+
   // ==========================================
   // Jimaku
   // Jimaku API configuration and defaults.

docs-site/seo.test.ts

@@ -56,34 +56,43 @@ test('main docs canonical uses /main/ and emits noindex', async () => {
     { rel: 'canonical', href: 'https://docs.subminer.moe/main/' },
   ]);
   expect(head).toContainEqual(['meta', { name: 'robots', content: 'noindex,follow' }]);
+  expect(mainDocsConfig.sitemap).toBeUndefined();
 
   process.env.SUBMINER_DOCS_CHANNEL = previousChannel;
   process.env.SUBMINER_DOCS_BASE = previousBase;
 });
 
-test('latest stable archive canonical points to root equivalent', async () => {
-  const previousChannel = process.env.SUBMINER_DOCS_CHANNEL;
-  const previousBase = process.env.SUBMINER_DOCS_BASE;
-  const previousVersion = process.env.SUBMINER_DOCS_VERSION;
-  const previousLatest = process.env.SUBMINER_DOCS_LATEST_STABLE;
-  process.env.SUBMINER_DOCS_CHANNEL = 'stable-archive';
-  process.env.SUBMINER_DOCS_BASE = '/v/0.14.0/';
-  process.env.SUBMINER_DOCS_VERSION = 'v0.14.0';
-  process.env.SUBMINER_DOCS_LATEST_STABLE = 'v0.14.0';
-  const { default: latestArchiveConfig } = await import('./.vitepress/config?latest-archive');
+test.each([
+  ['latest stable', 'v0.14.0', '/v/0.14.0/', 'https://docs.subminer.moe/v/0.14.0/usage'],
+  ['superseded', 'v0.12.0', '/v/0.12.0/', 'https://docs.subminer.moe/v/0.12.0/usage'],
+])(
+  '%s archive keeps a self-referential canonical and stays out of the index',
+  async (_label, version, base, expectedCanonical) => {
+    const previousChannel = process.env.SUBMINER_DOCS_CHANNEL;
+    const previousBase = process.env.SUBMINER_DOCS_BASE;
+    const previousVersion = process.env.SUBMINER_DOCS_VERSION;
+    const previousLatest = process.env.SUBMINER_DOCS_LATEST_STABLE;
+    process.env.SUBMINER_DOCS_CHANNEL = 'stable-archive';
+    process.env.SUBMINER_DOCS_BASE = base;
+    process.env.SUBMINER_DOCS_VERSION = version;
+    process.env.SUBMINER_DOCS_LATEST_STABLE = 'v0.14.0';
+    try {
+      const { default: archiveConfig } = await import(`./.vitepress/config?archive-${version}`);
 
-  const head = await latestArchiveConfig.transformHead?.(makeTransformContext('usage.md'));
+      const head = await archiveConfig.transformHead?.(makeTransformContext('usage.md'));
 
-  expect(head).toContainEqual([
-    'link',
-    { rel: 'canonical', href: 'https://docs.subminer.moe/usage' },
-  ]);
-
-  process.env.SUBMINER_DOCS_CHANNEL = previousChannel;
-  process.env.SUBMINER_DOCS_BASE = previousBase;
-  process.env.SUBMINER_DOCS_VERSION = previousVersion;
-  process.env.SUBMINER_DOCS_LATEST_STABLE = previousLatest;
-});
+      expect(head).toContainEqual(['link', { rel: 'canonical', href: expectedCanonical }]);
+      expect(head).toContainEqual(['meta', { name: 'robots', content: 'noindex,follow' }]);
+      // A sitemap here would advertise the archive tree we just excluded.
+      expect(archiveConfig.sitemap).toBeUndefined();
+    } finally {
+      process.env.SUBMINER_DOCS_CHANNEL = previousChannel;
+      process.env.SUBMINER_DOCS_BASE = previousBase;
+      process.env.SUBMINER_DOCS_VERSION = previousVersion;
+      process.env.SUBMINER_DOCS_LATEST_STABLE = previousLatest;
+    }
+  },
+);
 
 test('stable archive theme links stay on the selected version', async () => {
   const previousCwd = process.cwd();
@@ -433,3 +442,22 @@ test('docs sitemap excludes duplicate README page from indexable URLs', async ()
 
   expect(transformedItems?.map((item) => item.url)).toEqual(['', 'usage']);
 });
+
+test('docs sitemap dates every URL from the tracked checkout', async () => {
+  const previousRepoDir = process.env.SUBMINER_DOCS_REPO_DIR;
+  // Production builds render from an untracked snapshot, so the date has to come from
+  // the real checkout rather than VitePress's own srcDir git lookup.
+  process.env.SUBMINER_DOCS_REPO_DIR = docsSiteDir;
+  try {
+    const { default: sitemapConfig } = await import('./.vitepress/config?sitemap-lastmod');
+
+    const items = await sitemapConfig.sitemap?.transformItems?.([{ url: '' }, { url: 'usage' }]);
+
+    expect(items).toHaveLength(2);
+    for (const item of items ?? []) {
+      expect(item.lastmod).toMatch(/^\d{4}-\d{2}-\d{2}T/);
+    }
+  } finally {
+    process.env.SUBMINER_DOCS_REPO_DIR = previousRepoDir;
+  }
+});

docs-site/shortcuts.md

@@ -35,7 +35,7 @@ These work when the overlay window has focus.
 | `Ctrl/Cmd+G`       | Trigger field grouping (Kiku merge check)       | `shortcuts.triggerFieldGrouping`        |
 | `Ctrl/Cmd+Shift+A` | Mark last card as audio card                    | `shortcuts.markAudioCard`               |
 
-The multi-line shortcuts open a digit selector with a 3-second timeout (`shortcuts.multiCopyTimeoutMs`). Press `1`–`9` to select how many recent subtitle lines to combine. When the shortcut starts from mpv, SubMiner focuses the visible overlay for that selector instead of reserving the number keys in the mpv plugin.
+The multi-line shortcuts open a digit selector with a 3-second timeout (`shortcuts.multiCopyTimeoutMs`). Press `1`–`9` to select the total number of subtitle lines to combine, ending at the current line and moving backward through the subtitle timeline. The current line counts toward the selected total. When the shortcut starts from mpv, SubMiner focuses the visible overlay for that selector instead of reserving the number keys in the mpv plugin.
 
 ## Overlay Controls
 
@@ -49,6 +49,7 @@ These control playback and subtitle display. They require overlay window focus.
 | `J`                  | Cycle primary subtitle track                               |
 | `Shift+J`            | Cycle secondary subtitle track                             |
 | `Ctrl+Alt+P`         | Open playlist browser for current directory + queue        |
+| `Ctrl+Alt+A`         | Toggle Anime Browser inside the player                      |
 | `ArrowRight`         | Seek forward 5 seconds                                     |
 | `ArrowLeft`          | Seek backward 5 seconds                                    |
 | `ArrowUp`            | Seek forward 60 seconds                                    |
@@ -67,7 +68,7 @@ These control playback and subtitle display. They require overlay window focus.
 | `Right-click`        | Toggle pause (outside subtitle area)                       |
 | `Right-click + drag` | Reposition subtitles (on subtitle area)                    |
 
-The mpv-command rows above (`Space`, `F`, `J`, `Shift+J`, the seek/sub-seek/sub-step/sub-delay keys, replay/play-next, and quit) are merged from the `keybindings` config array and can be remapped or disabled there. `V` and the mouse actions are built-in overlay behaviors and are not part of the `keybindings` array. The playlist browser opens a split overlay modal with sibling video files on the left and the live mpv playlist on the right.
+The mpv-command rows above (`Space`, `F`, `J`, `Shift+J`, the seek/sub-seek/sub-step/sub-delay keys, replay/play-next, and quit) are merged from the `keybindings` config array and can be remapped or disabled there. `V` and the mouse actions are built-in overlay behaviors and are not part of the `keybindings` array. The playlist browser opens a split overlay modal with sibling video files on the left and the live mpv playlist on the right. The Anime Browser shortcut toggles a dedicated overlay modal bounded to the player; the standalone Anime Browser window remains independent, while active playback is shared.
 
 On macOS managed playback, SubMiner disables mpv's menu-bar shortcuts so configured SubMiner shortcuts like `Cmd+Shift+O` reach the mpv plugin instead of opening native mpv menu actions.
 

docs-site/subtitle-sidebar.md

@@ -9,9 +9,11 @@ The sidebar is enabled by default. Set `subtitleSidebar.enabled` to `false` if y
 When SubMiner parses the active subtitle source into a cue list, the sidebar becomes available. Toggle it with the `\` key (configurable via `subtitleSidebar.toggleKey`). While open:
 
 - The active cue is highlighted and kept in view as playback advances (when `autoScroll` is `true`).
-- Clicking any cue seeks mpv to that timestamp.
+- Clicking any cue seeks mpv into that line. For overlapping ASS karaoke, SubMiner moves past the previous line's exit animation when the selected cue has enough time remaining.
 - The sidebar stays synchronized with the overlay - media transitions and subtitle source changes update both simultaneously.
 
+For typeset ASS karaoke and animated signs, SubMiner collapses generated animation frames and repeated full-line color phases before they reach the sidebar. It recovers a clean complete line from a matching timed authoring comment or from full-line events surrounding generated fragments. Ordinary ASS comments, editor notes, alternate lines, repeated dialogue, and separately positioned signs remain distinct.
+
 The sidebar only appears when a parsed cue list is available. External subtitle sources that SubMiner cannot parse (for example, embedded ASS tracks rendered directly by mpv) will not populate the sidebar.
 
 ## Layout Modes

docs-site/troubleshooting.md

@@ -171,7 +171,9 @@ Without FFmpeg, card creation still works but audio and image fields will be emp
 
 **Audio or screenshot generation hangs**
 
-Media generation has a 30-second timeout (60 seconds for animated AVIF). If your video file is on a slow network mount or the codec requires software decoding, generation may time out. Try:
+Audio extraction has a 2-minute timeout. SubMiner also limits FFmpeg probing when mpv provides the selected audio stream, which avoids scanning unrelated subtitle and font-attachment streams in large MKV files. Screenshots retain a 30-second timeout, and animated AVIF uses 60 seconds.
+
+If your video file is on a slow or unresponsive network mount, generation may still time out. Try:
 
 - Using a local copy of the video file.
 - Reducing `ankiConnect.media.imageQuality` or switching from `avif` to `static` image type.
@@ -207,21 +209,26 @@ Resume playback and wait for the next subtitle to appear, then try mining again.
 
 Both **alass** and **ffsubsync** are optional external dependencies. Subtitle syncing requires at least one of them to be installed.
 
-**"Configured alass executable not found"**
+Subsync writes to the application log under the `subsync` scope, so the full command failure — exit code, stderr, resolved file paths — is recorded there as well as on the OSD.
+
+**"Could not find alass" / "Configured alass executable not found"**
 
 Install alass or configure the path:
 
+- **Homebrew**: `brew install alass`
 - **Arch Linux (AUR)**: `paru -S alass`
 - **Cargo**: `cargo install alass-cli`
 - Set the path: `subsync.alass_path` in your config.
 
-**"Configured ffsubsync executable not found"**
+Leaving the option empty searches `PATH` plus the usual install prefixes, and accepts either `alass` or `alass-cli`. Set the option explicitly when the binary lives somewhere else. The second message means the configured path itself does not exist — SubMiner never silently substitutes a different binary for one you named.
+
+**"Could not find ffsubsync" / "Configured ffsubsync executable not found"**
 
 Install ffsubsync or configure the path:
 
 - **Arch Linux (AUR)**: `paru -S python-ffsubsync`
 - **pip**: `pip install ffsubsync`
-- Must be on `PATH` or configured via `subsync.ffsubsync_path` in your config.
+- Must be discoverable, or configured via `subsync.ffsubsync_path` in your config.
 
 **"alass synchronization failed" / "ffsubsync synchronization failed"**
 
@@ -232,6 +239,38 @@ If subtitle sync fails (the error message is prefixed with the engine name):
 - Try running the sync tool manually to see detailed error output.
 - ffsubsync requires local files and cannot handle remote media streams (e.g., streaming URLs).
 
+**Syncing subtitles on a stream (Aniyomi extensions, Jellyfin)**
+
+Subtitle tracks mpv loaded from a URL are downloaded to a temporary file first, reusing mpv's own request headers, so they work as either the sync target or the alass reference. Downloading a Japanese track from Jimaku or TsukiHime while a stream is playing also works — it becomes the primary track and therefore the sync target.
+
+Internal subtitle tracks of a stream still go through `ffmpeg`, which has to reach the origin itself. If that fails, prefer an external track or a Jimaku download as the reference.
+
+Streams usually serve WebVTT, which alass cannot parse — it picks its parser from the file extension and treats a `.vtt` file as a video, failing with "no audio stream in file". SubMiner rewrites both the target and the reference as SRT for alass, so this is handled automatically; the retimed track mpv loads is that SRT.
+
+## Anime Browser
+
+See the [anime browser guide](/anime-browser) for how sources and the bridge work.
+
+**Nothing to search**
+
+SubMiner ships no extension repositories and bundles no sources. Until you add a repository index URL in the **Extensions** tab (or drop Aniyomi `.apk` files into the extensions directory) the browser has nothing to query.
+
+**"No pinned checksum for …"**
+
+The bridge bundle is verified against a SHA-256 that a maintainer has checked by hand. Bundles exist for macOS (arm64, x64), Linux (x64), and Windows (x64), but only macOS arm64 and Linux x64 are pinned so far; the rest stop with this message rather than running an unverified download. Other platforms are unsupported outright.
+
+**A source returns nothing, or asks for a login**
+
+Most extensions need configuration first - a server address, credentials, a preferred quality. Open the **Source settings** tab and fill them in; each save is handed straight back to the extension. Sources that need an Android WebView (typically for Cloudflare challenges) cannot work here, because the bridge has none.
+
+**"Playback failed" with an mpv error**
+
+"Playing" is only reported once mpv actually configures a video output, so this is a real failure rather than a silent one: a dead host, an expired stream URL, or an undecodable stream. Resolve the episode again; if it keeps failing, try another source or quality entry.
+
+**The browser starts failing every request**
+
+A bridge that dies (killed, crashed, stopped mid-operation) is named in the status bar and restarted on the next request. Anything already playing ends with it, since stream URLs point at its loopback proxy.
+
 ## TsukiHime
 
 **"xz binary not found"**

docs-site/tsukihime-integration.md

@@ -16,11 +16,11 @@ Unlike Jimaku, TsukiHime needs no account or API key. The only requirement is th
 
 The integration runs through an in-overlay modal opened with `Ctrl+Shift+T` by default. The modal has two tabs that filter the subtitle tracks of the selected release by role: the first follows `secondarySub.secondarySubLanguages` (English when unset), and the second is always **Japanese**, the currently supported primary subtitle language. Tracks with no language tag stay visible on the secondary tab.
 
-When you open the modal, SubMiner parses the current video filename to extract a title and episode number (same parser as Jimaku - `S01E03`, `1x03`, `E03`, and dash-separated numbers all work). If the filename yields a high-confidence match, SubMiner auto-searches immediately.
+When you open the modal, SubMiner parses the current video filename to extract a title, season, and episode number (same parser as Jimaku - `S01E03`, `1x03`, `E03`, and dash-separated numbers all work). If the filename yields a high-confidence match, SubMiner auto-searches immediately. Episodes launched from the [anime browser](/anime-browser) fill all three fields from the source's own listing instead of from a filename.
 
 From there:
 
-1. **Search** - SubMiner queries TsukiHime with `<title> <episode>`. Results appear as a list of releases (e.g. `[SubsPlease] ... - 28 (1080p)`), each showing size, file count, and the subtitle languages the release carries.
+1. **Search** - SubMiner queries TsukiHime with `<title> <season> <episode>`. Season 1 is left out on purpose: releases of a first season almost never carry `S01` in their name, so including it would match nothing. Later seasons are included, which is what makes their releases findable at all. Results appear as a list of releases (e.g. `[SubsPlease] ... - 28 (1080p)`), each showing size, file count, and the subtitle languages the release carries.
 2. **Browse releases** - Select a release to list the text subtitle tracks extracted from its files. English tracks sort first; image-based tracks (PGS/VobSub) are filtered out.
 3. **Download** - Selecting a track downloads the xz-compressed subtitle from TsukiHime's storage, decompresses it, saves it next to the video (or a temp directory for remote/streamed media), and loads it into mpv. Japanese tracks are selected as mpv's **primary** subtitle. Tracks from the configured secondary tab are assigned to mpv's **secondary** subtitle slot without replacing the primary. The filename carries the track's language - `<video basename>.en.<ext>` for English, `.ja` for Japanese, and so on - so mpv and media servers detect the language correctly.
 

docs-site/usage.md

@@ -70,6 +70,7 @@ subminer https://youtu.be/...     # Play a YouTube URL
 subminer stats                    # Open the immersion stats dashboard
 subminer doctor                   # Check dependencies, config, and the mpv socket
 subminer settings                 # Open the SubMiner settings window
+subminer anime                    # Open the anime browser window
 subminer app --setup              # Re-open first-run setup
 subminer -u                       # Check for updates
 ```
@@ -95,6 +96,8 @@ subminer texthooker               # Texthooker-only mode (-o also opens the brow
 subminer stats -b                 # Start/reuse the background stats daemon
 subminer stats -s                 # Stop the background stats daemon
 subminer stats cleanup            # Backfill vocabulary metadata, prune stale rows
+subminer stats cleanup -d --dry-run  # Preview cleanup of repeated typeset subtitle lines
+subminer stats cleanup -d --lookback-days 30  # Clean only lines recorded in the last 30 days
 subminer stats rebuild            # Rebuild rollup data
 subminer doctor --refresh-known-words  # Refresh the known-word cache
 subminer logs -e                  # Export a sanitized log ZIP and print its path
@@ -107,6 +110,8 @@ subminer app --stop               # Stop the background app
 subminer --version                # Print the launcher's version
 ```
 
+`stats cleanup` runs one mode per invocation: `-v`/`--vocab` (the default), `-l`/`--lifetime`, or `-d`/`--duplicate-lines`; explicitly selected modes cannot be combined. `--dry-run` and `--lookback-days <days>` apply to `--duplicate-lines` only and are rejected without it; `--lookback-days` must be at least one day, and leaving it off scans all history.
+
 Jellyfin, cross-machine sync, and character-dictionary commands have their own sections: [Jellyfin](/jellyfin-integration), [Sync Between Machines](/launcher-script#sync-between-machines), and [Character Dictionary](/character-dictionary).
 
 </details>
@@ -131,13 +136,14 @@ SubMiner.AppImage --open-tsukihime        # Open TsukiHime subtitle search
 SubMiner.AppImage --yomitan               # Open Yomitan settings
 SubMiner.AppImage --settings              # Open the SubMiner settings window
 SubMiner.AppImage --jellyfin              # Open the Jellyfin setup window
+SubMiner.AppImage --anime                 # Open the anime browser window
 SubMiner.AppImage --dictionary            # Generate a character dictionary ZIP
 SubMiner.AppImage --start --dev           # Enable app/dev mode
 SubMiner.AppImage --start --log-level debug  # Verbose logging without dev mode
 SubMiner.AppImage --help                  # Show all options
 ```
 
-The remaining flags are internal or scripting-only surfaces: the `--jellyfin-*` family (login, library listing, item playback, cast announce), `--sync-cli` (the app's headless sync entrypoint that `subminer sync` proxies to), `--dictionary-candidates` / `--dictionary-select`, and `--playback-feedback <text>`. Run `SubMiner.AppImage --help` for the complete list. The previous `--open-animetosho` flag is still accepted as a deprecated alias for `--open-tsukihime`.
+The remaining flags are internal or scripting-only surfaces: the `--jellyfin-*` family (login, library listing, item playback, cast announce), `--sync-cli` (the app's headless sync entrypoint that `subminer sync` proxies to), the `--stats-cleanup-*` family that `subminer stats cleanup` forwards (`--stats-cleanup-vocab`, `--stats-cleanup-lifetime`, `--stats-cleanup-duplicate-lines`, and its `--stats-cleanup-dry-run` / `--stats-cleanup-lookback-days <days>` modifiers), `--dictionary-candidates` / `--dictionary-select`, and `--playback-feedback <text>`. Run `SubMiner.AppImage --help` for the complete list. The previous `--open-animetosho` flag is still accepted as a deprecated alias for `--open-tsukihime`.
 
 </details>
 

docs/RELEASING.md

@@ -58,12 +58,15 @@
    `latest*.yml` and `*.blockmap` files under `release/`.
 5. Commit the prerelease prep (package.json version bump + the generated
    `release/prerelease-notes.md`). CI does not regenerate notes — it uses the
-   committed file — so review it before committing. If you add more
-   `changes/*.md` fragments for a later beta/RC, rerun
-   `bun run changelog:prerelease-notes --version <version>`; the generator uses
-   the existing prerelease notes as the baseline only when their hidden
-   `prerelease-base-version` marker matches the current base version, and asks
-   Claude to merge only the new fragment material. Do not run
+   committed file — so review it before committing. Rerun
+   `bun run changelog:prerelease-notes --version <version>` for every later
+   beta/RC, even if no fragments changed: the notes carry a hidden
+   `prerelease-version` marker and CI rejects the tag when the marker does not
+   match it (verify locally with
+   `bun run changelog:check-prerelease-notes --version <version>`). The
+   generator reuses the existing notes as the cumulative baseline when their
+   marker (or legacy `prerelease-base-version` marker) matches the current base
+   version, and asks Claude to merge only the new fragment material. Do not run
    `bun run changelog:build`.
 6. Tag the commit: `git tag v<version>`.
 7. Push commit + tag.
@@ -78,6 +81,8 @@ Notes:
 - Pass `--date` explicitly when you want the release stamped with the local cut date; otherwise the generator uses the current ISO date, which can roll over to the next UTC day late at night.
 - `changelog:check` now rejects tag/package version mismatches.
 - `changelog:prerelease-notes` also rejects tag/package version mismatches and writes `release/prerelease-notes.md` without mutating tracked changelog files. When that file already exists, the generator includes it in the Claude prompt so later beta/RC notes reuse the reviewed text instead of starting over.
+- From the second prerelease of a base version onward, the notes open with a `## Changes since <previous tag>` section above the cumulative `## Highlights`. The generator locates the newest preceding beta/RC tag for the same base version (semver order: all betas before all RCs), diffs `changes/*.md` between that tag and the working tree, and asks Claude to describe only the behavioral beta-to-beta differences — added fragments as new changes, modified fragments by their before/after difference (editorial-only edits are dropped), deleted fragments as removed/reverted changes. If no fragments changed (for example a packaging-only rebuild), the section states that explicitly without a Claude call. The delta section carries no separate contributor attribution; `## What's Changed` stays cumulative like `## Highlights`.
+- `changelog:check-prerelease-notes --version <version>` verifies the committed notes' `prerelease-version` marker matches the version being tagged; the prerelease workflow runs it and fails the release on stale notes.
 - `changelog:build` generates `CHANGELOG.md` + `release/release-notes.md` (both polished by `claude -p`) and removes the released `changes/*.md` fragments. The CHANGELOG keeps internal notes inside a `<details><summary>Internal changes</summary>` collapse; the release notes drop them entirely.
 - `release/release-notes.md` (and `release/prerelease-notes.md`) include GitHub-style attribution after `## Highlights`: a `## What's Changed` list crediting each released fragment as `by @<author> in #<pr>`, plus a `## New Contributors` section for first-time authors. Attribution is resolved per fragment via `git log` (the commit that added the fragment) + `gh api .../commits/<sha>/pulls`, with one `gh` search per author for the first-contribution check. It needs `gh` installed and authenticated; if `gh` is unavailable or a lookup fails, the generator warns and emits notes without the attribution sections rather than failing. The CHANGELOG itself stays attribution-free.
 - The release workflow no longer auto-runs `changelog:build`. If pending `changes/*.md` fragments are present on a tag-based run, CI exits with a clear `::error::` pointing at the local fix. Run `bun run changelog:build --version <version>` locally, commit the polished output, then tag.

docs/architecture/2026-03-15-renderer-performance-design.md

@@ -64,24 +64,31 @@ External subtitle files only (SRT, VTT, ASS). Embedded subtitle tracks are out o
 A cue parser extracts both timing and text content from subtitle files for prefetching.
 
 **Parsed cue structure:**
-
 ```typescript
 interface SubtitleCue {
   startTime: number; // seconds
   endTime: number; // seconds
   text: string; // plain text, decoded from the source format
+  source?: 'canonical-ass'; // recovered authored text for generated ASS animation
+  animationStartTime?: number; // full generated-frame envelope; entrance/exit frames
+  animationEndTime?: number; //   run past the authored timing, live matching uses this
 }
 ```
 
 **Supported formats:**
-
 - SRT/VTT: Regex-based parsing of timing lines + text content between timing blocks.
-- ASS: Parse `[Events]` section, extract `Dialogue:` lines, read the field order from the `Format:` row, and take everything after the Text field index as the text (Text can itself contain commas).
+- ASS: Parse the `[Events]` section, read the field order from the `Format:` row, and extract timed `Dialogue:` lines. Timed `Comment:` lines are normally ignored, but can supply canonical authored text when they match a nearby generated animation from the same style and actor. Text can itself contain commas.
 
 **ASS decoding.** The parser is where ASS text is decoded, once, via `assToPlainText()` in `src/core/services/ass-text.ts`. That decoder mirrors mpv's `ass_to_plaintext` so a cue read from a file reads identically to the same line arriving live on `sub-text`: `{...}` override blocks are markup, `\pN … \p0` vector drawing runs are dropped rather than shown as text, `\N`/`\n`/`\h` are the only escapes (`\{`, `\}` and `\\` are not), and an unclosed `{` is rendered verbatim. Every layer downstream — renderer, timing tracker, tokenizer, tokenization cache keys — receives plain text and uses `normalizePlainSubtitleText()` for whitespace only, so nothing decodes the same string twice and one authored line always maps to one cache key.
 
 **Duplicate collapsing.** Typeset scripts emit one `Dialogue:` event per animation frame, plus layered copies of the same line. The parser collapses identical text over an identical span unconditionally, and collapses contiguous same-text runs of at least three events when the run looks like an animation. For ASS that means shared style and actor plus authoring evidence: a temporal tag (`\t`, `\move`, `\k`/`\kf`/`\ko`/`\K`, or anything wrapped in `\t(...)`), an animated `Effect` column (`Karaoke`, `Banner`, `Scroll`), or override values that change across the run. Static tags shared by every event (`\pos`, an identical `\clip`) are not evidence. SRT/VTT carry no such metadata, so there collapsing needs at least five contiguous events all under 0.1s — the frame timing left behind by ASS-to-SRT conversion. The parser keeps this authoring metadata (style, actor, layer, `Effect`, parsed override commands, source order) private; `parseSubtitleCues()` returns only `SubtitleCue`.
 
+ASS scripts can also redraw one complete lyric for two or more long color/highlight phases. Those flush-timed phases collapse separately from short animation frames when they share text, style, actor, and layer and carry direct animation evidence, such as temporal tags or changing non-spatial overrides. Spatial command changes do not prove a phase, so separately positioned signs remain distinct.
+
+**Canonical animation recovery.** Some ASS producers keep the readable lyric or sign as a timed `Comment:` and generate hundreds of `Dialogue:` frames containing repeated glyphs or changing clip regions. Others retain the complete line as brief `Dialogue:` events around the generated fragments. A complete event is promoted only when nearby dialogue from the same style and actor forms a proven animation cluster and reconstructs its entire text in source order. The generated frames are then replaced by one cue marked `source: 'canonical-ass'`. This source marker lets the live primary-subtitle path prefer the clean authored text and timing for display, sidebar history, immersion recording, and mining, while unmatched editor notes and alternative translations remain ignored. Secondary selection advances to an entering canonical cue at its generated animation start when the preceding authored cue ends before the new authored span. Unrelated simultaneous cues that continue through the new span remain visible.
+
+**Font texture cleanup.** A clipped repeated-glyph run or frequent changes to secondary alpha marks a texture seed. Clipped runs do not need a font override because some signs build their masks from ordinary `l` glyphs. The parser removes short clipped pieces that share a no-font seed's style and timing, or pieces that share a font seed's style, timing, and font even when the actor changes. It also removes positioned text layers with at least `E0` global alpha when they overlap a seed in the same style. Opaque authored sign text stays publishable when the texture switches fonts or actors around it.
+
 #### Prefetch Service Lifecycle
 
 1. **Activation trigger:** When a subtitle track is activated (or changes), check if it's external via MPV's `track-list` property. If `external === true`, read the file via `external-filename` using the existing `loadSubtitleSourceText` infrastructure.
@@ -158,7 +165,6 @@ tokens (already have frequencyRank values from parser-level applyFrequencyRanks)
 ### Dependency Analysis
 
 All annotations either depend on MeCab POS data or benefit from running after it:
-
 - **Known word marking:** Needs base tokens (surface/headword). No POS dependency, but no reason to run separately.
 - **Frequency filtering:** Uses `pos1Exclusions` and `pos2Exclusions` to clear frequency ranks on excluded tokens (particles, noise). Depends on MeCab POS data.
 - **JLPT marking:** Uses `shouldIgnoreJlptForMecabPos1` to filter. Depends on MeCab POS data.
@@ -175,14 +181,18 @@ function annotateTokens(tokens, deps, options): MergedToken[] {
 
   // Single pass: known word + frequency filtering + JLPT computed together
   const annotated = tokens.map((token) => {
-    const isKnown = nPlusOneEnabled ? token.isKnown || computeIsKnown(token, deps) : false;
+    const isKnown = nPlusOneEnabled
+      ? token.isKnown || computeIsKnown(token, deps)
+      : false;
 
     // Filter frequency rank using POS exclusions (rank values already set at parser level)
     const frequencyRank = frequencyEnabled
       ? filterFrequencyRank(token, pos1Exclusions, pos2Exclusions)
       : undefined;
 
-    const jlptLevel = jlptEnabled ? computeJlptLevel(token, deps.getJlptLevel) : undefined;
+    const jlptLevel = jlptEnabled
+      ? computeJlptLevel(token, deps.getJlptLevel)
+      : undefined;
 
     return { ...token, isKnown, frequencyRank, jlptLevel };
   });
@@ -223,7 +233,6 @@ Replace `document.createElement('span')` calls in the renderer with `templateSpa
 ### Current Behavior
 
 In `renderWithTokens` (`subtitle-render.ts`), each render cycle:
-
 1. Clears DOM with `innerHTML = ''`
 2. Creates a `DocumentFragment`
 3. Calls `document.createElement('span')` for each token (~10-15 per subtitle)
@@ -259,30 +268,27 @@ Full recycling (collecting old nodes, clearing attributes, reusing them) require
 
 ## Combined Impact Summary
 
-| Scenario                          | Before     | After      | Improvement |
-| --------------------------------- | ---------- | ---------- | ----------- |
-| Normal playback (prefetch-warmed) | ~200-320ms | ~30-50ms   | ~80-85%     |
-| Cache hit (repeated subtitle)     | ~72ms      | ~55-65ms   | ~10-20%     |
-| Cache miss (immediate seek)       | ~200-320ms | ~150-260ms | ~20-25%     |
+| Scenario | Before | After | Improvement |
+|----------|--------|-------|-------------|
+| Normal playback (prefetch-warmed) | ~200-320ms | ~30-50ms | ~80-85% |
+| Cache hit (repeated subtitle) | ~72ms | ~55-65ms | ~10-20% |
+| Cache miss (immediate seek) | ~200-320ms | ~150-260ms | ~20-25% |
 
 ---
 
 ## Files Summary
 
 ### New Files
-
 - `src/core/services/subtitle-prefetch.ts`
 - `src/core/services/subtitle-cue-parser.ts`
 
 ### Modified Files
-
 - `src/core/services/subtitle-processing-controller.ts` (expose `preCacheTokenization`)
 - `src/core/services/tokenizer/annotation-stage.ts` (batched single-pass)
 - `src/renderer/subtitle-render.ts` (template cloneNode)
 - `src/main.ts` (wire up prefetch service)
 
 ### Test Files
-
 - New tests for subtitle cue parser (SRT, VTT, ASS formats)
 - New tests for subtitle prefetch service (priority window, seek, pause/resume)
 - Updated tests for annotation stage (same behavior, new implementation)

docs/architecture/domains.md

@@ -3,7 +3,7 @@
 # Domain Ownership
 
 Status: active
-Last verified: 2026-07-15
+Last verified: 2026-08-02
 Owner: Kyle Yasuda
 Read when: you need to find the owner module for a behavior or test surface
 
@@ -25,9 +25,20 @@ Read when: you need to find the owner module for a behavior or test surface
 - Anki workflow: `src/anki-integration/`, `src/core/services/anki-jimaku*.ts`
 - Immersion tracking: `src/core/services/immersion-tracker/`
   Includes stats storage/query schema such as `imm_videos`, `imm_media_art`, and `imm_youtube_videos` for per-video and YouTube-specific library metadata.
-  `delete-maintenance-scheduler.ts` coalesces and serializes stats deletes; expensive deletion and summary rebuilds run in `delete-maintenance-worker-thread.ts` while the tracker queues playback writes. Each batch uses one transaction, lexical update, rollup refresh, and lifetime rebuild.
+  Library-entry identity aliases and merge recommendations are persisted alongside this schema; the stats HTTP and SPA layers only expose and present those domain decisions.
+  `delete-maintenance-scheduler.ts` coalesces and serializes stats deletes; the expensive work runs in `delete-maintenance-worker-thread.ts` while the tracker queues playback writes. Each batch uses one transaction, lexical update, rollup refresh, and incremental lifetime subtraction (`planLifetimeRemovals`/`applyLifetimeRemovals` in `lifetime.ts`). Merges, moves, AniList reassignments, and `stats cleanup -l` use `repairLifetimeSummariesFromMedia` (recompute from the per-video media ledger). The full lifetime rebuild survives only as the empty-table bootstrap; anywhere else it would collapse lifetime totals to the session retention window.
 - AniList tracking + character dictionary: `src/core/services/anilist/`, `src/main/runtime/composers/anilist-*`, `src/main/character-dictionary-runtime.ts`, `src/main/character-dictionary-runtime/`
 - Jellyfin integration: `src/core/services/jellyfin*.ts`, `src/main/runtime/composers/jellyfin-*`
+- Anime browser: extension bridge client, sidecar, and stream handling in `src/anime-bridge/`;
+  the loopback stream proxy separates request transport/retry from response transformation in
+  `stream-strip-transport.ts` and `stream-strip-response.ts`;
+  browser window UI in `src/animeui/` (preload `src/preload-animeui.ts`); runtime wiring in
+  `src/main/runtime/anime-browser-application-runtime.ts`, `src/main/runtime/anime-browser-runtime.ts`,
+  `src/main/runtime/anime-browser-ipc-handlers.ts`, `src/main/runtime/anime-browser-sessions.ts`,
+  `src/main/runtime/anime-bridge-installer.ts`, and `src/main/runtime/stream-playback-metadata.ts`.
+  The play queue resolves episodes on click and appends them to mpv's real playlist
+  (`src/main/runtime/anime-browser-queue.ts`), then observes media-path changes to
+  attach prepared external tracks and update the browser queue state.
 - Window trackers: `src/window-trackers/`
 - Stats HTTP app: `src/core/services/stats-server.ts`, with route groups and shared route support
   in `src/core/services/stats-server/`
@@ -45,6 +56,7 @@ Read when: you need to find the owner module for a behavior or test surface
 - Settings UI contracts: `src/types/settings.ts`
 - Session-binding contracts: `src/types/session-bindings.ts`
 - Stats HTTP wire contracts: `src/types/stats-wire.ts`, `src/types/stats-http-contract.ts`
+- Anime browser contracts: `src/types/anime-browser.ts`, bridge wire types in `src/anime-bridge/types.ts`
 - Compatibility-only barrel: `src/types.ts`
 
 ## Ownership Heuristics

docs/architecture/stats-trends-data-flow.md

@@ -23,7 +23,9 @@ Trend charts now consume one chart-oriented backend payload from `/api/stats/tre
   - lookup rate trends
   - watch-time by day-of-week/hour
 - vocabulary-backed:
-  - new-words trend
+  - new-words trend reads permanent daily lexical rollups
+  - rollup rows count only vocabulary-visible tokens and normalize mixed legacy timestamp units
+  - a persisted rollup version invalidates stale materializations and triggers an atomic background rebuild
 
 ## Metric Semantics
 

docs/architecture/subtitle-overlay-priming.md

@@ -3,7 +3,7 @@
 # Subtitle Overlay Priming
 
 Status: active
-Last verified: 2026-08-04
+Last verified: 2026-08-19
 Owner: Kyle Yasuda
 Read when: debugging subtitle state or blank Linux/X11 overlay windows when the visible overlay is shown or recreated
 
@@ -69,14 +69,97 @@ coming and prefetching would otherwise idle for the rest of the cue.
 
 ## Live Cue Delivery
 
+- Primary live text first resolves recovered canonical ASS animations. Otherwise, when
+  every live mpv line matches an active parsed cue, it uses the parsed cue text so exact
+  full-span style layers appear once instead of repeating for fill, border, blur, shadow,
+  or equivalent whitespace variants. Any unmatched live line keeps the complete live
+  stack, preserving dialogue or signs that overlap a lyric.
 - A tokenization cache miss emits the plain cue synchronously. Tokenization remains serialized so
   live work does not contend for Yomitan state.
+- The initial `time-pos`, explicit renderer seeks, and later seek-like jumps reprocess mpv's
+  current raw `sub-text` after the new playback time is stored. Explicit intent matters because
+  adjacent subtitle jumps can be shorter than the general seek-distance threshold. This corrects
+  ASS cleanup when mpv delivered the destination subtitle before the destination timestamp.
+- Renderer `sub-seek` commands use the active parsed cue list when available. Simultaneous cues
+  share one boundary, overlapping lyrics advance from the latest active boundary, and mpv's native
+  command remains the fallback when no parsed destination exists. This prevents generated karaoke
+  frames from consuming next/previous subtitle presses.
+- Subtitle sidebar selections seek past the preceding sanitized cue's overlapping exit span when
+  the selected cue has enough time remaining. This keeps direct row selection on the requested
+  karaoke line while clamping the seek inside that cue.
+- If startup paints raw text before embedded ASS parsing finishes, parsed cue arrival may replace
+  that provisional line. The one-prime-per-media guard still suppresses identical repeats.
 - If a newer cue arrives while an older line is still tokenizing, the newer plain cue or empty
   clear payload is emitted immediately. The older tokenization result is dropped before it can
   replace the current cue.
 - The current cue upgrades in place when its tokens and annotations are ready. This can reflow text
   or character images, but cue visibility does not wait for that work.
 
+## Secondary Subtitle Flow
+
+- `secondary-sub-text` remains the immediate fallback, so unreadable subtitle sources, remote URLs,
+  and still-extracting embedded tracks appear without waiting for file resolution. Embedded-track
+  extraction runs for local and network-mounted files alike (demuxing reads the whole container,
+  about 10 seconds per GB on gigabit, under a generous timeout); only true remote URLs skip it,
+  having no on-disk container to demux.
+- The live fallback also suppresses per-glyph typesetting walls: when many simultaneous
+  one-glyph lines are present (generated karaoke lettering flattened into live text), those
+  lines and their short syllable companions are dropped while concurrent dialogue lines stay.
+  This keeps the overlay clean while extraction is still in flight and for sources that never
+  produce parsed cues.
+- Parsed secondary text and the live fallback remove exact repeated lines at any length. A
+  flattened-line identity also removes long dialogue/sign repetitions that differ only in
+  whitespace or terminal punctuation, while distinct simultaneous short lines remain separate.
+- `secondary-subtitle-track.ts` resolves `secondary-sid` against mpv's track list. External tracks
+  are read directly; supported embedded text tracks are extracted through the same ffmpeg-backed
+  source resolver used by primary subtitle prefetching.
+- The selected source is parsed with `parseSubtitleCues()`, including metadata-aware ASS duplicate
+  and animation collapse. Playback `time-pos` selects the active parsed cue after applying
+  `secondary-sub-delay`.
+- Fragment reconstruction marks tall multi-row positioned parts as a grid only when they read
+  like tiling: a couple of texts repeated across many fragments, the same text re-shown at one
+  spot over time (countdown/animation frames), or scattered single glyphs. Secondary text omits
+  those grids instead of flattening a translated table or schedule into one synthetic line.
+  Wrapped lyric rows, CC-style dialogue blocks, and reconstructed single-line karaoke remain
+  eligible for display.
+- The resolved text is stored in `mpvClient.currentSecondarySubText` before it is broadcast. The
+  overlay, mining, timing tracker, and immersion statistics therefore consume the same secondary
+  text when a readable source is available.
+- Simultaneous parsed cues use whitespace-insensitive identity, so ASS layers that vary only
+  between ordinary, hard, or ideographic spaces appear once.
+- Simultaneous ASS lines are flattened in top-to-bottom positioned order, falling back to their
+  authored source order when no usable position exists.
+- Half-size kana positioned directly above a same-timed kanji caption is treated as ASS
+  furigana. The parser omits it from published cues but retains hidden matching metadata so
+  mpv's raw live text can be reconciled without displaying or mining the reading. The
+  timing tracker (clipboard copy, recent-line mining) and immersion recorders run the same
+  reconciliation on the `sub-start`/`sub-end` sample, so they record what the overlay shows.
+- Broadcast-caption rows that spell one utterance across several same-timed positioned events
+  (same style, layer, and vertical band, stacked at most two text rows apart) are joined into
+  one cue with a single line break, so `preserveLineBreaks` treats them like an authored `\N`,
+  and the recorders above see the whole sentence. A row continues the one above it when that row
+  is a bare speaker label, ends without terminal punctuation, or leaves a ≪…≫ / ⸨…⸩ span open; a
+  lower row that opens its own label or span always starts a new cue, which keeps two speakers
+  sharing the screen on separate lines. The pass runs only on scripts that read as broadcast
+  captions (a meaningful share of events carry speaker labels or ≪…≫ / ⸨…⸩ spans) and only on
+  rows containing Japanese, because fansub typesetting stacks positioned rows for signs, chat
+  bubbles, and headlines where that punctuation convention does not hold.
+- Fragment-only ASS karaoke is reconstructed per style before publication. Explicit spaces
+  survive concatenation. Latin fragment typesetting with no literal spaces also recovers word
+  boundaries represented only by materially larger horizontal `\pos` or `\move` gaps within that
+  line. Unpositioned fragments stay compact instead of gaining guessed spaces between syllables.
+  Short runs qualify only when overlapping positioned events also show changing overrides or
+  repeated layer copies; an English or romaji style name alone never turns ordinary dialogue into
+  a lyric.
+- Recovered canonical ASS text remains active for the generated animation envelope. For
+  reconstructed lyric styles, the longest-lived active line wins over brief entrance and exit
+  fragments from the same style.
+- Media and `secondary-sid` changes clear the previous parsed state before refreshing the source;
+  track-list changes refresh without discarding an unchanged source. Observed
+  `secondary-sub-delay` changes retime the active parsed cue without rereading the file. If loading,
+  extraction, or parsing fails, the controller returns to live mpv text and the renderer's
+  conservative short stack heuristic remains the final display fallback.
+
 ## Emitted State
 
 - `emitSubtitle(payload)` maps to `emitSubtitlePayload(payload)`. Overlay windows and annotation
@@ -84,8 +167,8 @@ coming and prefetching would otherwise idle for the rest of the cue.
 - The basic subtitle websocket receives the immediate plain cue only. Because its serialized
   payload discards annotations, the later upgrade would be an identical duplicate and is skipped
   when text and cue timing match.
-- Secondary priming reads mpv `secondary-sub-text`, stores it in
-  `mpvClient.currentSecondarySubText`, and broadcasts `secondary-subtitle:set` to overlay windows.
+- Secondary priming reads mpv `secondary-sub-text` and routes it through the secondary track
+  controller. A parsed active cue replaces the live text when the selected source is readable.
 - If secondary `requestProperty` fails, the primary flow stays complete and only a debug line is
   written.
 
@@ -129,7 +212,11 @@ coming and prefetching would otherwise idle for the rest of the cue.
   path, empty or stale bounding shapes produced invisible or clipped subtitles even though the
   overlay window remained mapped above mpv.
 - Pointer pass-through should continue to use `setIgnoreMouseEvents(true, { forward: true })` and
-  the Linux cursor-poll fallback, not bounding-shape clipping.
+  the Linux cursor-poll fallback, not bounding-shape clipping. Note that on Windows click-through
+  must go through `applyOverlayClickThrough()` (`src/core/services/overlay-click-through.ts`),
+  which omits `forward: true` there: Electron implements forwarding with a global low-level mouse
+  hook that lags mouse input system-wide whenever the main thread stalls; the Windows cursor poll
+  handles overlay wake-up instead.
 - Visible-overlay show/reset marks Linux pointer passthrough state dirty even when the logical
   interaction state is already inactive. The next cursor-poll tick must still reapply
   `setIgnoreMouseEvents(true, { forward: true })`; otherwise a newly shown Electron overlay can keep

docs/knowledge-base/catalog.md

@@ -3,7 +3,7 @@
 # Documentation Catalog
 
 Status: active  
-Last verified: 2026-05-23  
+Last verified: 2026-08-13
 Owner: Kyle Yasuda  
 Read when: finding internal docs or checking verification status
 
@@ -11,16 +11,16 @@ Read when: finding internal docs or checking verification status
 | --- | --- | --- | --- | --- |
 | KB home | `docs/README.md` | active | 2026-05-23 | internal entrypoint |
 | Architecture index | `docs/architecture/README.md` | active | 2026-05-23 | top-level runtime map |
-| Domain ownership | `docs/architecture/domains.md` | active | 2026-05-23 | runtime and feature ownership |
+| Domain ownership | `docs/architecture/domains.md` | active | 2026-08-02 | runtime and feature ownership |
 | Layering rules | `docs/architecture/layering.md` | active | 2026-05-23 | dependency direction and smells |
 | Subtitle overlay priming | `docs/architecture/subtitle-overlay-priming.md` | active | 2026-06-01 | visible-overlay subtitle startup flow |
 | KB rules | `docs/knowledge-base/README.md` | active | 2026-05-23 | maintenance policy |
 | Core beliefs | `docs/knowledge-base/core-beliefs.md` | active | 2026-03-13 | agent-first principles |
 | Quality scorecard | `docs/knowledge-base/quality.md` | active | 2026-03-13 | quality grades and gaps |
-| Workflow index | `docs/workflow/README.md` | active | 2026-05-23 | execution map |
+| Workflow index | `docs/workflow/README.md` | active | 2026-08-13 | execution map |
 | Planning guide | `docs/workflow/planning.md` | active | 2026-05-23 | lightweight vs execution plans |
-| Agent plugins | `docs/workflow/agent-plugins.md` | active | 2026-05-23 | repo-local agent workflow plugin ownership |
-| Verification guide | `docs/workflow/verification.md` | active | 2026-05-23 | maintained verification lanes |
+| Agent skills | `docs/workflow/agent-skills.md` | active | 2026-08-23 | repo-local workflow skill ownership |
+| Verification guide | `docs/workflow/verification.md` | active | 2026-08-13 | maintained verification lanes |
 | Release guide | `docs/RELEASING.md` | active | 2026-05-23 | release checklist |
 
 ## Update Rules

docs/superpowers/specs/2026-04-09-library-summary-replaces-per-day-design.md

@@ -1,184 +0,0 @@
-# Library Summary Replaces Per-Day Trends — Design
-
-**Status:** Draft
-**Date:** 2026-04-09
-**Scope:** `stats/` frontend, `src/core/services/immersion-tracker/query-trends.ts` backend
-
-## Problem
-
-The "Library — Per Day" section on the stats Trends tab (`stats/src/components/trends/TrendsTab.tsx:224-254`) renders six stacked-area charts — Videos, Watch Time, Cards, Words, Lookups, and Lookups/100w, each broken down per title per day.
-
-In practice these charts are not useful:
-
-- Most titles only have activity on one or two days in a window, so they render as isolated bumps on a noisy baseline.
-- Stacking 7+ titles with mostly-zero days makes individual lines hard to follow.
-- The top "Activity" and "Period Trends" sections already answer "what am I doing per day" globally.
-- The "Library — Cumulative" section directly below already answers "which titles am I progressing through" with less noise.
-
-The per-day section occupies significant vertical space without carrying its weight, and the user has confirmed it should be replaced.
-
-## Goal
-
-Replace the six per-day stacked charts with a single "Library — Summary" section that surfaces per-title aggregate statistics over the selected date range. The new view should make it trivially easy to answer: "For the selected window, which titles am I spending time on, how much mining output have they produced, and how efficient is my lookup rate on each?"
-
-## Non-goals
-
-- Changing the "Library — Cumulative" section (stays as-is).
-- Changing the "Activity", "Period Trends", or "Patterns" sections.
-- Adding a new API endpoint — the existing dashboard endpoint is extended in place.
-- Renaming internal `anime*` data-model identifiers (`animeId`, `imm_anime`, etc.). Those stay per the convention established in `c5e778d7`; only new fields/types/user-visible strings use generic "title"/"library" wording.
-- Supporting a true all-time library view on the Trends tab. If that's ever wanted, it belongs on a different tab.
-
-## Solution Overview
-
-Delete the "Library — Per Day" section. In its place, add "Library — Summary", composed of:
-
-1. A horizontal-bar leaderboard chart of watch time per title (top 10, descending).
-2. A sortable table of every title with activity in the selected window, with columns: Title, Watch Time, Videos, Sessions, Cards, Words, Lookups, Lookups/100w, Date Range.
-
-Both controls are scoped to the top-of-page date range selector. The existing shared Anime Visibility filter continues to work — it now gates Summary + Cumulative instead of Per-Day + Cumulative.
-
-## Backend
-
-### New type
-
-Add to `stats/src/types/stats.ts` and the backend query module:
-
-```ts
-type LibrarySummaryRow = {
-  title: string;              // display title — anime series, YouTube video title, etc.
-  watchTimeMin: number;       // sum(total_active_min) across the window
-  videos: number;             // distinct video_id count
-  sessions: number;           // session count from imm_sessions
-  cards: number;              // sum(total_cards)
-  words: number;              // sum(total_tokens_seen)
-  lookups: number;            // sum(lookup_count) from imm_sessions
-  lookupsPerHundred: number | null; // lookups / words * 100, null when words == 0
-  firstWatched: number;       // min(rollup_day) as epoch day, within the window
-  lastWatched: number;        // max(rollup_day) as epoch day, within the window
-};
-```
-
-### Query changes in `src/core/services/immersion-tracker/query-trends.ts`
-
-- Add `librarySummary: LibrarySummaryRow[]` to `TrendsDashboardQueryResult`.
-- Populate it from a single aggregating query over `imm_daily_rollups` joined to `imm_videos` → `imm_anime`, filtered by `rollup_day` within the selected window. Session count and lookup count come from `imm_sessions` aggregated by `video_id` and then grouped by the parent library entry. Use a single query (or at most two joined/unioned) — no N+1.
-- `imm_anime` is the generic library-grouping table; anime series, YouTube videos, and yt-dlp imports all land there. The internal table name stays `imm_anime`; only the new field uses generic naming.
-- Return rows pre-sorted by `watchTimeMin` descending so the leaderboard is zero-cost and the table default sort matches.
-- Emit `lookupsPerHundred: null` when `words == 0`.
-
-### Removed from API response
-
-Drop the entire `animePerDay` field from `TrendsDashboardQueryResult` (both backend in `src/core/services/immersion-tracker/query-trends.ts` and frontend in `stats/src/types/stats.ts`).
-
-Internally, the existing helpers (`buildPerAnimeFromDailyRollups`, `buildEpisodesPerAnimeFromDailyRollups`) are still used as intermediates to build `animeCumulative.*` via `buildCumulativePerAnime`. Keep those helpers — just scope their output to local variables inside `getTrendsDashboard` instead of exposing them on the response. The `buildPerAnimeFromSessions` call for lookups and the `buildLookupsPerHundredPerAnime` helper become unused and can be deleted.
-
-Before removing `animePerDay` from the frontend type, verify no other file under `stats/src/` references it. Based on current inspection, only `TrendsTab.tsx` and `stats/src/types/stats.ts` touch it.
-
-## Frontend
-
-### New component: `stats/src/components/trends/LibrarySummarySection.tsx`
-
-Owns the header, leaderboard chart, visibility-filtered data, and the table. Keeps `TrendsTab.tsx` from growing. Component props: `{ rows: LibrarySummaryRow[]; hiddenTitles: ReadonlySet<string>; windowStart: Date; windowEnd: Date }`.
-
-Internal state: `useState<{ column: ColumnId; direction: 'asc' | 'desc' }>` for sort, defaulting to `{ column: 'watchTimeMin', direction: 'desc' }`.
-
-### Layout
-
-Replaces `TrendsTab.tsx:224-254`:
-
-```
-[SectionHeader: "Library — Summary"]
-[AnimeVisibilityFilter — unchanged, shared with Cumulative below]
-[Card, col-span-full: Leaderboard — horizontal bar chart, ~260px tall]
-[Card, col-span-full: Sortable table, auto height up to ~480px with internal scroll]
-```
-
-Both cards use the existing chart/card wrapper styling.
-
-### Leaderboard chart
-
-- Recharts horizontal bar chart (matches the rest of the page — existing charts use `recharts`, not ECharts).
-- Top 10 titles by watch time. If fewer titles have activity, render what's there.
-- Y-axis: title (category), truncated with ellipsis at container width; full title visible in the Recharts tooltip.
-- X-axis: minutes (number).
-- Use `layout="vertical"` with `YAxis dataKey="title" type="category"` and `XAxis type="number"`.
-- Single series color: `#8aadf4` (matching the existing Watch Time color).
-- Reuse `CHART_DEFAULTS`, `CHART_THEME`, `TOOLTIP_CONTENT_STYLE` from `stats/src/lib/chart-theme.ts` so theming matches the rest of the dashboard.
-- Chart order is fixed at watch-time desc regardless of table sort — the leaderboard's meaning is fixed.
-
-### Table
-
-- Plain HTML `<table>` with Tailwind classes. No new deps.
-- Columns, in order:
-  1. **Title** — left-aligned, sticky, truncated with ellipsis, full title on hover.
-  2. **Watch Time** — formatted `Xh Ym` when ≥60 min, else `Xm`.
-  3. **Videos** — integer.
-  4. **Sessions** — integer.
-  5. **Cards** — integer.
-  6. **Words** — integer.
-  7. **Lookups** — integer.
-  8. **Lookups/100w** — one decimal place, `—` when null.
-  9. **Date Range** — `Mon D → Mon D` using the title's `firstWatched` / `lastWatched` within the window.
-- Click a column header to sort; click again to reverse. Visual arrow on the active column.
-- Numeric columns right-aligned.
-- Null `lookupsPerHundred` sorts as the lowest value in both directions (consistent with "no data").
-- Row hover highlight; no row click action (read-only view).
-- Empty state: "No library activity in the selected window."
-
-### Visibility filter integration
-
-Hiding a title via `AnimeVisibilityFilter` removes it from both the leaderboard and the table. The filter's set of available titles is built from the union of titles that appear in `librarySummary` and the existing `animeCumulative.*` arrays (matches current behavior in `buildAnimeVisibilityOptions`).
-
-### `TrendsTab.tsx` changes
-
-- Remove the `filteredEpisodesPerAnime`, `filteredWatchTimePerAnime`, `filteredCardsPerAnime`, `filteredWordsPerAnime`, `filteredLookupsPerAnime`, `filteredLookupsPerHundredPerAnime` locals.
-- Remove the six `<StackedTrendChart>` calls in the "Library — Per Day" section.
-- Remove the `<SectionHeader>Library — Per Day</SectionHeader>` and the `<AnimeVisibilityFilter>` from that position.
-- Insert `<SectionHeader>Library — Summary</SectionHeader>` + `<AnimeVisibilityFilter>` + `<LibrarySummarySection>` in the same place.
-- Update `buildAnimeVisibilityOptions` input to use `librarySummary` titles instead of the six dropped `animePerDay.*` arrays.
-
-## Data flow
-
-1. `useTrends(range, groupBy)` calls `/api/stats/trends/dashboard`.
-2. Response now includes `librarySummary` (sorted by watch time desc).
-3. `TrendsTab` holds the shared `hiddenAnime` set (unchanged).
-4. `LibrarySummarySection` receives `librarySummary` + `hiddenAnime`, filters out hidden rows, renders the leaderboard from the top-10 slice of the filtered list, renders the table from the filtered list with local sort state applied.
-5. Date-range selector changes trigger a new fetch; `groupBy` toggle does not affect the summary section (it's always window-total).
-
-## Edge cases
-
-- **No activity in window:** Section renders header + empty-state card. Leaderboard card hidden. Visibility filter hidden.
-- **One title only:** Leaderboard renders a single bar; table renders one row. No special-casing.
-- **Title with zero words but non-zero lookups:** `lookupsPerHundred` is `null`, rendered as `—`. Sort treats null as lowest.
-- **Title with zero cards/lookups/words but non-zero watch time:** Normal zero rendering, still shown.
-- **Very long titles:** Ellipsis in chart y-axis labels and table title column; full title in `title` attribute / ECharts tooltip.
-- **Mixed sources (anime + YouTube):** No special case — both land in `imm_anime` and are grouped uniformly.
-
-## Testing
-
-### Backend (`query-trends.ts`)
-
-New unit tests, following the existing pattern:
-
-1. Empty window returns `librarySummary: []`.
-2. Single title with a few rollups: all aggregates are correct; `firstWatched`/`lastWatched` match the bounding days within the window.
-3. Multiple titles: rows returned sorted by watch time desc.
-4. Mixed sources (anime-style + YouTube-style entries in `imm_anime`): both appear in the summary with their own aggregates.
-5. Title with `words == 0`: `lookupsPerHundred` is `null`.
-6. Date range excludes some rollups: excluded rollups are not counted; `firstWatched`/`lastWatched` reflect only within-window activity.
-7. `sessions` and `lookups` come from `imm_sessions`, not `imm_daily_rollups`, and are correctly attributed to the parent library entry.
-
-### Frontend
-
-- Existing Trends tab smoke test should continue to pass after wiring.
-- Optional: a targeted render test for `LibrarySummarySection` (empty state, single title, sort toggle, visibility filter interaction). Not required for merge if the smoke test exercises the happy path.
-
-## Release / docs
-
-- One fragment in `changes/*.md` summarizing the replacement.
-- No user-facing docs (`docs-site/`) changes unless the per-day section was documented there — verify during implementation.
-
-## Open items
-
-None.

docs/superpowers/specs/2026-04-09-stats-dashboard-feedback-pass-design.md

@@ -1,347 +0,0 @@
-# Stats Dashboard Feedback Pass — Design
-
-Date: 2026-04-09
-Scope: Stats dashboard UX follow-ups from user feedback (items 1–7).
-Delivery: **Single PR**, broken into logically scoped commits.
-
-## Goals
-
-Address seven concrete pieces of feedback against the Statistics menu:
-
-1. Library — collapse episodes behind a per-series dropdown.
-2. Sessions — roll up multiple sessions of the same episode within a day.
-3. Trends — add a 365d range option.
-4. Library — delete an episode (video) from its detail view.
-5. Vocabulary — tighten spacing between word and reading in the Top 50 table.
-6. Episode detail — hide cards whose Anki notes have been deleted.
-7. Trend/watch charts — add gridlines, fix tick legibility, unify theming.
-
-Out of scope for this pass: English-token ingestion cleanup and Overview stat-card drill-downs (feedback items 8 and 9). Those require a larger design decision and a migration respectively.
-
-## Files touched (inventory)
-
-Dashboard (`stats/src/`):
-- `components/library/LibraryTab.tsx` — collapsible groups (item 1).
-- `components/library/MediaDetailView.tsx`, `components/library/MediaHeader.tsx` — delete-episode action (item 4).
-- `components/sessions/SessionsTab.tsx`, `components/library/MediaSessionList.tsx` — episode rollup (item 2).
-- `components/trends/DateRangeSelector.tsx`, `hooks/useTrends.ts`, `lib/api-client.ts`, `lib/api-client.test.ts` — 365d (item 3).
-- `components/vocabulary/FrequencyRankTable.tsx` — word/reading column collapse (item 5).
-- `components/anime/EpisodeDetail.tsx` — filter deleted Anki cards (item 6).
-- `components/trends/TrendChart.tsx`, `components/trends/StackedTrendChart.tsx`, `components/overview/WatchTimeChart.tsx`, `lib/chart-theme.ts` — chart clarity (item 7).
-- New file: `stats/src/lib/session-grouping.ts` + `session-grouping.test.ts`.
-
-Backend (`src/core/services/`):
-- `immersion-tracker/query-trends.ts` — extend `TrendRange` and `TREND_DAY_LIMITS` (item 3).
-- `immersion-tracker/__tests__/query.test.ts` — 365d coverage (item 3).
-- `stats-server.ts` — passthrough if range validation lives here (check before editing).
-- `__tests__/stats-server.test.ts` — 365d coverage (item 3).
-
-## Commit plan
-
-One PR, one feature per commit. Order picks low-risk mechanical changes first so failures in later commits don't block merging of earlier ones.
-
-1. `feat(stats): add 365d range to trends dashboard` (item 3)
-2. `fix(stats): tighten word/reading column in Top 50 table` (item 5)
-3. `fix(stats): hide cards deleted from Anki in episode detail` (item 6)
-4. `feat(stats): delete episode from library detail view` (item 4)
-5. `feat(stats): collapsible series groups in library` (item 1)
-6. `feat(stats): roll up same-episode sessions within a day` (item 2)
-7. `feat(stats): gridlines and unified theme for trend charts` (item 7)
-
-Each commit must pass `bun run typecheck`, `bun run test:fast`, and any change-specific checks listed below.
-
----
-
-## Item 1 — Library collapsible series groups
-
-### Current behavior
-
-`LibraryTab.tsx` groups media via `groupMediaLibraryItems` and always renders the full grid of `MediaCard`s beneath each group header.
-
-### Target behavior
-
-Each group header becomes clickable. Groups with `items.length > 1` default to **collapsed**; single-video groups stay expanded (collapsing them would be visual noise).
-
-### Implementation
-
-- State: `const [collapsedGroups, setCollapsedGroups] = useState<Set<string>>(...)`. Initialize from `grouped` where `items.length > 1`.
-- Toggle helper: `toggleGroup(key: string)` adds/removes from the set.
-- Group header: wrap in a `<button>` with `aria-expanded` and a chevron icon (`▶`/`▼`). Keep the existing cover + title + subtitle layout inside the button.
-- Children grid is conditionally rendered on `!collapsedGroups.has(group.key)`.
-- Header summary (`N videos · duration · cards`) stays visible in both states so collapsed groups remain informative.
-
-### Tests
-
-- New `LibraryTab.test.tsx` (if not already present — check first) covering:
-  - Multi-video group renders collapsed on first mount.
-  - Single-video group renders expanded on first mount.
-  - Clicking the header toggles visibility.
-  - Header summary is visible in both states.
-
----
-
-## Item 2 — Sessions episode rollup within a day
-
-### Current behavior
-
-`SessionsTab.tsx:10-24` groups sessions by day label only (`formatSessionDayLabel(startedAtMs)`). Multiple sessions of the same episode on the same day show as independent rows. `MediaSessionList.tsx` has the same problem inside the library detail view.
-
-### Target behavior
-
-Within each day, sessions with the same `videoId` collapse into one parent row showing combined totals. A chevron reveals the individual sessions. Single-session buckets render flat (no pointless nesting).
-
-### Implementation
-
-- New helper in `stats/src/lib/session-grouping.ts`:
-  ```ts
-  export interface SessionBucket {
-    key: string;              // videoId as string, or `s-${sessionId}` for singletons
-    videoId: number | null;
-    sessions: SessionSummary[];
-    totalActiveMs: number;
-    totalCardsMined: number;
-    representativeSession: SessionSummary; // most recent, for header display
-  }
-  export function groupSessionsByVideo(sessions: SessionSummary[]): SessionBucket[];
-  ```
-  Sessions missing a `videoId` become singleton buckets.
-
-- `SessionsTab.tsx`: after day grouping, pipe each `daySessions` through `groupSessionsByVideo`. Render each bucket:
-  - `sessions.length === 1`: existing `SessionRow` behavior, unchanged.
-  - `sessions.length >= 2`: render a **bucket row** that looks like `SessionRow` but shows combined totals and session count (e.g. `3 sessions · 1h 24m · 12 cards`). Chevron state stored in a second `Set<string>` on bucket key. Expanded buckets render the child `SessionRow`s indented (`pl-8`) beneath the header.
-- `MediaSessionList.tsx`: within the media detail view, a single video's sessions are all the same `videoId` by definition — grouping here is by day only, and within a day multiple sessions render nested under a day header. Re-use the same visual pattern; factor the bucket row into a shared `SessionBucketRow` component.
-
-### Delete semantics
-
-- Deleting a bucket header offers "Delete all N sessions in this group" (reuse `confirmDayGroupDelete` pattern with a bucket-specific message, or add `confirmBucketDelete`).
-- Deleting an individual session from inside an expanded bucket keeps the existing single-delete flow.
-
-### Tests
-
-- `session-grouping.test.ts`:
-  - Empty input → empty output.
-  - All unique videos → N singleton buckets.
-  - Two sessions same videoId → one bucket with correct totals and representative (most recent start time).
-  - Missing videoId → singleton bucket keyed by sessionId.
-- `SessionsTab.test.tsx` (extend or add) verifying the rendered bucket rows expand/collapse and delete hooks fire with the right ID set.
-
----
-
-## Item 3 — 365d trends range
-
-### Backend
-
-`src/core/services/immersion-tracker/query-trends.ts`:
-- `type TrendRange = '7d' | '30d' | '90d' | '365d' | 'all';`
-- Add `'365d': 365` to `TREND_DAY_LIMITS`.
-- `getTrendDayLimit` picks up the new key automatically because of the `Exclude<TrendRange, 'all'>` generic.
-
-`src/core/services/stats-server.ts`:
-- Search for any hardcoded range validation (e.g. allow-list in the trends route handler) and extend it.
-
-### Frontend
-
-- `hooks/useTrends.ts`: widen the `TimeRange` union.
-- `components/trends/DateRangeSelector.tsx`: add `'365d'` to the options list. Display label stays as `365d`.
-- `lib/api-client.ts` / `api-client.test.ts`: if the client validates ranges, add `365d`.
-
-### Tests
-
-- `query.test.ts`: extend the existing range table to cover `365d` returning 365 days of data.
-- `stats-server.test.ts`: ensure the route accepts `range=365d`.
-- `api-client.test.ts`: ensure the client emits the new range.
-
-### Change-specific checks
-
-- `bun run test:config` is not required here (no schema/defaults change).
-- Run `bun run typecheck` + `bun run test:fast`.
-
----
-
-## Item 4 — Delete episode from library detail
-
-### Current behavior
-
-`MediaDetailView.tsx` provides session-level delete only. The backend `deleteVideo` exists (`query-maintenance.ts:509`), the API is exposed at `stats-server.ts:559`, and `api-client.deleteVideo` is already wired (`stats/src/lib/api-client.ts:146`). `EpisodeList.tsx:46` already uses it from the anime tab.
-
-### Target behavior
-
-A "Delete Episode" action in `MediaHeader` (top-right, small, `text-ctp-red`), gated by `confirmEpisodeDelete(title)`. On success, call `onBack()` and make sure the parent `LibraryTab` refetches.
-
-### Implementation
-
-- Add an `onDeleteEpisode?: () => void` prop to `MediaHeader` and render the button only if provided.
-- In `MediaDetailView`:
-  - New handler `handleDeleteEpisode` that calls `apiClient.deleteVideo(videoId)`, then `onBack()`.
-  - Reuse `confirmEpisodeDelete` from `stats/src/lib/delete-confirm.ts`.
-- In `LibraryTab`:
-  - `useMediaLibrary` returns fresh data on mount. The simplest fix: pass a `refresh` function from the hook (extend the hook if it doesn't already expose one) and call it when the detail view signals back.
-  - Alternative: force a remount by incrementing a `libraryVersion` key on the library list. Prefer `refresh` for clarity.
-
-### Tests
-
-- Extend the existing `MediaDetailView.test.tsx`: mock `apiClient.deleteVideo`, click the new button, confirm `onBack` fires after success.
-- `useMediaLibrary.test.ts`: if we add a `refresh` method, cover it.
-
----
-
-## Item 5 — Vocabulary word/reading column collapse
-
-### Current behavior
-
-`FrequencyRankTable.tsx:110-144` uses a 5-column table: `Rank | Word | Reading | POS | Seen`. Word and Reading are auto-sized, producing a large gap.
-
-### Target behavior
-
-Merge Word + Reading into a single column titled "Word". Reading sits immediately after the headword in a muted, smaller style.
-
-### Implementation
-
-- Drop the `<th>Reading</th>` header and cell.
-- Word cell becomes:
-  ```tsx
-  <td className="py-1.5 pr-3">
-    <span className="text-ctp-text font-medium">{w.headword}</span>
-    {reading && (
-      <span className="text-ctp-subtext0 text-xs ml-1.5">
-        【{reading}】
-      </span>
-    )}
-  </td>
-  ```
-  where `reading = fullReading(w.headword, w.reading)` and differs from `headword`.
-- Keep `fullReading` import from `reading-utils`.
-
-### Tests
-
-- Extend `FrequencyRankTable.test.tsx` (if present — otherwise add a focused test) to assert:
-  - Headword renders.
-  - Reading renders when different from headword.
-  - Reading does not render when equal to headword.
-
----
-
-## Item 6 — Hide Anki-deleted cards in Cards Mined
-
-### Current behavior
-
-`EpisodeDetail.tsx:109-147` iterates `cardEvents`, fetches note info via `ankiNotesInfo(allNoteIds)`, and for each `noteId` renders a row even if no matching `info` came back — the user sees an empty word with an "Open in Anki" button that leads nowhere.
-
-### Target behavior
-
-After `ankiNotesInfo` resolves:
-- Drop `noteId`s that are not in the resolved map.
-- Drop `cardEvents` whose `noteIds` list was non-empty but is now empty after filtering.
-- Card events with a positive `cardsDelta` but no `noteIds` (legacy rollup path) still render as `+N cards` — we have no way to cross-reference them, so leave them alone.
-
-### Implementation
-
-- Compute `filteredCardEvents` as a `useMemo` depending on `data.cardEvents` and `noteInfos`.
-- Iterate `filteredCardEvents` instead of `cardEvents` in the render.
-- Surface a subtle note (optional, muted) "N cards hidden (deleted from Anki)" at the end of the list if any were filtered — helps the user understand why counts here diverge from session totals. Final decision on the note can be made at PR review; default: **show it**.
-
-### Tests
-
-- Add a test in `EpisodeDetail.test.tsx` (add the file if not present) that stubs `ankiNotesInfo` to return only a subset of notes and verifies the missing ones are not rendered.
-
-### Other call sites
-
-- Grep so far shows `ankiNotesInfo` is only used in `EpisodeDetail.tsx`. Re-verify before landing the commit; if another call site appears, apply the same filter.
-
----
-
-## Item 7 — Trend/watch chart clarity pass
-
-### Current behavior
-
-`TrendChart.tsx`, `StackedTrendChart.tsx`, and `WatchTimeChart.tsx` render Recharts components with:
-- No `CartesianGrid` → no horizontal reference lines.
-- 9px axis ticks → borderline unreadable.
-- Height 120 → cramped.
-- Tooltip uses raw labels (`04/04` etc.).
-- No shared theme object; each chart redefines colors and tooltip styles inline.
-
-`stats/src/lib/chart-theme.ts` already exists and currently exports a single `CHART_THEME` constant with tick/tooltip colors and `barFill`. It will be extended, not replaced, to preserve existing consumers.
-
-### Target behavior
-
-All three charts share a theme, have horizontal gridlines, readable ticks, and sensible tooltips.
-
-### Implementation
-
-Extend `stats/src/lib/chart-theme.ts` with the additional shared defaults (keeping the existing `CHART_THEME` export intact so current consumers don't break):
-```ts
-export const CHART_THEME = {
-  tick: '#a5adcb',
-  tooltipBg: '#363a4f',
-  tooltipBorder: '#494d64',
-  tooltipText: '#cad3f5',
-  tooltipLabel: '#b8c0e0',
-  barFill: '#8aadf4',
-  grid: '#494d64',
-  axisLine: '#494d64',
-} as const;
-
-export const CHART_DEFAULTS = {
-  height: 160,
-  tickFontSize: 11,
-  margin: { top: 8, right: 8, bottom: 0, left: 0 },
-  grid: { strokeDasharray: '3 3', vertical: false },
-} as const;
-
-export const TOOLTIP_CONTENT_STYLE = {
-  background: CHART_THEME.tooltipBg,
-  border: `1px solid ${CHART_THEME.tooltipBorder}`,
-  borderRadius: 6,
-  color: CHART_THEME.tooltipText,
-  fontSize: 12,
-};
-```
-
-Apply to each chart:
-- Import `CartesianGrid` from recharts.
-- Insert `<CartesianGrid stroke={CHART_THEME.grid} {...CHART_DEFAULTS.grid} />` inside each chart container.
-- `<XAxis tick={{ fontSize: CHART_DEFAULTS.tickFontSize, fill: CHART_THEME.tick }} />` and equivalent `YAxis`.
-- `YAxis` gains `axisLine={{ stroke: CHART_THEME.axisLine }}`.
-- `ResponsiveContainer` height changes from 120 → `CHART_DEFAULTS.height`.
-- `Tooltip` `contentStyle` uses `TOOLTIP_CONTENT_STYLE`, and charts pass a `labelFormatter` when the label is a date key (e.g. show `Fri Apr 4`).
-
-### Unit formatters
-
-- `TrendChart` already accepts a `formatter` prop — extend usage sites to pass unit-aware formatters where they aren't already (`formatDuration`, `formatNumber`, etc.).
-
-### Tests
-
-- `chart-theme.test.ts` (if present — otherwise add a trivial snapshot to keep the shape stable).
-- `TrendChart` snapshot/render tests: no regression, gridline element present.
-
----
-
-## Verification gate
-
-Before requesting code review, run:
-
-```
-bun run typecheck
-bun run test:fast
-bun run test:env
-bun run test:runtime:compat   # dist-sensitive check for the charts
-bun run build
-bun run test:smoke:dist
-```
-
-No docs-site changes are planned in this spec; if `docs-site/` ends up touched (e.g. screenshots), also run `bun run docs:test` and `bun run docs:build`.
-
-No config schema changes → `bun run test:config` and `bun run generate:config-example` are not required.
-
-## Risks and open questions
-
-- **MediaDetailView refresh**: `useMediaLibrary` may not expose a `refresh` function. If it doesn't, the simplest path is adding one; the alternative (keying a remount) works but is harder to test. Decide during implementation.
-- **Session bucket delete UX**: "Delete all N sessions in this group" is powerful. The copy must make it clear the underlying sessions are being removed, not just the grouping. Reuse `confirmBucketDelete` wording from existing confirm helpers if possible.
-- **Anki-deleted-cards hidden notice**: Showing a subtle "N cards hidden" footer is a call that can be made at PR review.
-- **Bucket delete helper**: `confirmBucketDelete` does not currently exist in `delete-confirm.ts`. Implementation either adds it or reuses `confirmDayGroupDelete` with bucket-specific wording — decide during the session-rollup commit.
-
-## Changelog entry
-
-User-visible PR → needs a fragment under `changes/*.md`. Suggested title:
-`Stats dashboard: collapsible series, session rollups, 365d trends, chart polish, episode delete.`

docs/workflow/README.md

@@ -3,7 +3,7 @@
 # Workflow
 
 Status: active  
-Last verified: 2026-05-23  
+Last verified: 2026-08-13
 Owner: Kyle Yasuda  
 Read when: planning or executing nontrivial work in this repo
 
@@ -13,7 +13,7 @@ This section is the internal workflow map for contributors and agents.
 
 - [Planning](./planning.md) - when to write a lightweight plan vs a full execution plan
 - [Verification](./verification.md) - maintained test/build lanes and handoff gate
-- [Agent Plugins](./agent-plugins.md) - repo-local plugin ownership for agent workflow skills
+- [Agent Skills](./agent-skills.md) - repo-local workflow skill ownership
 - [Release Guide](../RELEASING.md) - tagged release workflow
 
 ## Default Flow

docs/workflow/agent-plugins.md

@@ -1,27 +0,0 @@
-<!-- read_when: using or modifying repo-local agent plugins -->
-
-# Agent Plugins
-
-Status: active
-Last verified: 2026-05-23
-Owner: Kyle Yasuda
-Read when: packaging or migrating repo-local agent workflow skills into plugins
-
-## SubMiner Workflow Plugin
-
-- Canonical plugin path: `plugins/subminer-workflow/`
-- Marketplace catalog: `.agents/plugins/marketplace.json`
-- Canonical skill sources:
-  - `plugins/subminer-workflow/skills/subminer-scrum-master/`
-  - `plugins/subminer-workflow/skills/subminer-change-verification/`
-
-## Migration Rule
-
-- Plugin-owned skills are the source of truth.
-- `.agents/skills/subminer-*` remain only as compatibility shims.
-- Existing script entrypoints under `.agents/skills/subminer-change-verification/scripts/` stay as wrappers so historical commands do not break.
-
-## Verification
-
-- For plugin/docs-only changes, start with `bun run test:docs:kb`.
-- Use the plugin-owned verifier when the change crosses from docs into scripts or workflow logic.

docs/workflow/agent-skills.md

@@ -0,0 +1,34 @@
+<!-- read_when: using or modifying repo-local agent skills -->
+
+# Agent Skills
+
+Status: active
+Last verified: 2026-08-23
+Owner: Kyle Yasuda
+Read when: using, adding, or changing a repo-local agent workflow skill
+
+## Canonical Skills
+
+- `.agents/skills/subminer-change-verification/`
+  - Selects the cheapest sufficient repo-native verification lane.
+  - Defers command ownership to `package.json` and `docs/workflow/verification.md`.
+- `.agents/skills/subminer-release/`
+  - Prepares, cuts, publishes, or repairs stable and prerelease releases.
+  - Defers release procedure and policy to `docs/RELEASING.md`.
+
+Repo-local workflows stay as standalone skills. Do not add plugin packaging, marketplace metadata, or compatibility shims unless the workflow is intentionally being distributed beyond this repository.
+
+## Rules
+
+- Keep each skill focused on one repeatable repository task.
+- Prefer instructions over helper scripts unless deterministic tooling provides clear value beyond existing package commands.
+- Keep trigger descriptions narrow enough to avoid invoking skills for unrelated requests.
+- Update this page and the documentation catalog when skill ownership changes.
+
+## Verification
+
+For skill or internal workflow documentation changes, run:
+
+```bash
+bun run test:docs:kb
+```

docs/workflow/verification.md

@@ -3,14 +3,14 @@
 # Verification
 
 Status: active  
-Last verified: 2026-07-06  
+Last verified: 2026-08-13
 Owner: Kyle Yasuda  
 Read when: selecting the right verification lane for a change
 
 ## Lane Infrastructure
 
 - Lane membership is defined once in `scripts/test-lanes.ts` and discovered by
-  directory — new test files join their lane automatically; never hand-list test
+  directory, so new test files join their lane automatically; never hand-list test
   files in `package.json`.
 - `scripts/run-test-lane.mjs` runs each test file in its own `bun test` process
   (per-file isolation with a wall timeout) so a hanging test or leaked global in
@@ -43,8 +43,8 @@ bun run docs:build
 
 ## Cheap-First Lane Selection
 
-- Docs-only boundary/content changes: `bun run docs:test`, `bun run docs:build`
-- Internal KB / `AGENTS.md` changes: `bun run test:docs:kb`
+- User-facing `docs-site/` changes: `bun run docs:test`, `bun run docs:build`
+- Internal KB, `AGENTS.md`, or `.agents/skills/**` changes: `bun run test:docs:kb`
 - Config/schema/defaults: `bun run test:config`, then `bun run generate:config-example` if template/defaults changed
 - Launcher/plugin: `bun run test:launcher` or `bun run test:env`
 - Runtime-compat / compiled behavior: `bun run test:runtime:compat`

launcher/commands/app-command.ts

@@ -16,6 +16,10 @@ type AppCommandDeps = {
     appPath: string,
     logLevel: LauncherCommandContext['args']['logLevel'],
   ) => void;
+  launchAnimeBrowserDetached: (
+    appPath: string,
+    logLevel: LauncherCommandContext['args']['logLevel'],
+  ) => void;
 };
 
 const defaultAppCommandDeps: AppCommandDeps = {
@@ -23,6 +27,8 @@ const defaultAppCommandDeps: AppCommandDeps = {
   launchSyncUiDetached: (appPath, logLevel) =>
     launchAppCommandDetached(appPath, ['--sync-window'], logLevel, 'sync-ui'),
   launchAppBackgroundDetached,
+  launchAnimeBrowserDetached: (appPath, logLevel) =>
+    launchAppCommandDetached(appPath, ['--anime'], logLevel, 'anime'),
 };
 
 export function runAppPassthroughCommand(
@@ -37,6 +43,11 @@ export function runAppPassthroughCommand(
     deps.runAppCommandWithInherit(appPath, ['--settings']);
     return true;
   }
+  if (args.animeBrowser) {
+    // Detached: the browser window is long-lived and owns the bridge process.
+    deps.launchAnimeBrowserDetached(appPath, args.logLevel);
+    return true;
+  }
   if (args.syncUi) {
     deps.launchSyncUiDetached(appPath, args.logLevel);
     return true;

launcher/commands/command-modules.test.ts

@@ -207,6 +207,7 @@ test('app command starts default macOS background app detached from launcher', (
       calls.push('attached');
     },
     launchSyncUiDetached: () => calls.push('sync-ui'),
+    launchAnimeBrowserDetached: () => {},
     launchAppBackgroundDetached: (appPath, logLevel) => {
       calls.push(`detached:${appPath}:${logLevel}`);
     },
@@ -227,6 +228,7 @@ test('app command starts default Linux background app detached from launcher', (
       calls.push('attached');
     },
     launchSyncUiDetached: () => calls.push('sync-ui'),
+    launchAnimeBrowserDetached: () => {},
     launchAppBackgroundDetached: (appPath, logLevel) => {
       calls.push(`detached:${appPath}:${logLevel}`);
     },
@@ -248,6 +250,7 @@ test('app command keeps explicit passthrough args attached', () => {
       forwarded.push(appArgs);
     },
     launchSyncUiDetached: () => detached.push('sync-ui'),
+    launchAnimeBrowserDetached: () => {},
     launchAppBackgroundDetached: () => {
       detached.push('detached');
     },
@@ -266,6 +269,7 @@ test('sync UI command launches the app detached from the terminal', () => {
   const handled = runAppPassthroughCommand(context, {
     runAppCommandWithInherit: () => calls.push('piped'),
     launchSyncUiDetached: (appPath, logLevel) => calls.push(`sync-ui:${appPath}:${logLevel}`),
+    launchAnimeBrowserDetached: () => calls.push('anime'),
     launchAppBackgroundDetached: () => calls.push('detached'),
   });
 

launcher/commands/history-command.ts

@@ -23,6 +23,7 @@ import {
   type HistorySeriesEntry,
 } from '../history.js';
 import type { Args } from '../types.js';
+import { ensureLinuxRuntimePluginAvailable } from '../runtime-plugin-preflight.js';
 import type { LauncherCommandContext } from './context.js';
 
 export type HistorySessionAction = 'previous' | 'replay' | 'next' | 'browse' | 'quit';
@@ -333,6 +334,13 @@ export async function runHistoryCommand(
   const { args, scriptPath } = context;
 
   checkPickerDependencies(args);
+  if (args.useRofi) {
+    await ensureLinuxRuntimePluginAvailable({
+      appPath: context.appPath ?? undefined,
+      scriptPath,
+      logLevel: args.logLevel,
+    });
+  }
   const themePath = args.useRofi ? findRofiTheme(scriptPath) : null;
 
   const dbPath = resolveImmersionDbPath();

launcher/commands/jellyfin-command.ts

@@ -2,6 +2,7 @@ import { fail } from '../log.js';
 import { runAppCommandWithInherit } from '../mpv.js';
 import { commandExists } from '../util.js';
 import { runJellyfinPlayMenu } from '../jellyfin.js';
+import { ensureLinuxRuntimePluginAvailable } from '../runtime-plugin-preflight.js';
 import { shouldForwardLogLevel } from '../types.js';
 import type { LauncherCommandContext } from './context.js';
 
@@ -64,6 +65,13 @@ export async function runJellyfinCommand(context: LauncherCommandContext): Promi
     if (args.useRofi && !commandExists('rofi')) {
       fail('rofi not found. Install rofi or omit -R for fzf.');
     }
+    if (args.useRofi) {
+      await ensureLinuxRuntimePluginAvailable({
+        appPath,
+        scriptPath,
+        logLevel: args.logLevel,
+      });
+    }
     await runJellyfinPlayMenu(appPath, args, scriptPath, mpvSocketPath);
     return true;
   }

launcher/commands/playback-command.test.ts

@@ -63,6 +63,7 @@ function createContext(): LauncherCommandContext {
       logsExport: false,
       version: false,
       settings: false,
+      animeBrowser: false,
       configPath: false,
       configShow: false,
       mpvIdle: false,
@@ -496,3 +497,39 @@ test('playback command ensures Linux runtime plugin before mpv launch', async ()
 
   assert.deepEqual(calls, ['plugin', 'startMpv']);
 });
+
+test('rofi playback repairs support assets before opening the picker', async () => {
+  const context = createContext();
+  context.args = {
+    ...context.args,
+    target: '',
+    targetKind: '',
+    useRofi: true,
+  };
+  const calls: string[] = [];
+
+  await runPlaybackCommandWithDeps(context, {
+    ensurePlaybackSetupReady: async () => {},
+    ensureRuntimePluginReady: async () => {
+      calls.push('assets');
+    },
+    chooseTarget: async () => {
+      calls.push('picker');
+      return { target: '/tmp/movie.mkv', kind: 'file' };
+    },
+    checkPickerDependencies: () => {},
+    checkDependencies: () => {},
+    registerCleanup: () => {},
+    startMpv: async () => {
+      calls.push('startMpv');
+    },
+    waitForUnixSocketReady: async () => true,
+    startOverlay: async () => {},
+    launchAppCommandDetached: () => {},
+    log: () => {},
+    cleanupPlaybackSession: async () => {},
+    getMpvProc: () => null,
+  });
+
+  assert.deepEqual(calls, ['assets', 'picker', 'startMpv']);
+});

launcher/commands/playback-command.ts

@@ -157,6 +157,7 @@ export async function runPlaybackCommand(context: LauncherCommandContext): Promi
       });
     },
     chooseTarget,
+    checkPickerDependencies,
     checkDependencies,
     registerCleanup,
     startMpv,
@@ -177,6 +178,7 @@ type PlaybackCommandDeps = {
     args: Args,
     scriptPath: string,
   ) => Promise<{ target: string; kind: 'file' | 'url' } | null>;
+  checkPickerDependencies?: (args: Args) => void;
   checkDependencies: (args: Args) => void;
   registerCleanup: (context: LauncherCommandContext) => void;
   startMpv: typeof startMpv;
@@ -201,7 +203,18 @@ export async function runPlaybackCommandWithDeps(
   await deps.ensurePlaybackSetupReady(context);
 
   if (!args.target) {
-    checkPickerDependencies(args);
+    (deps.checkPickerDependencies ?? checkPickerDependencies)(args);
+  }
+
+  let runtimeAssetsReady = false;
+  const ensureRuntimeAssetsReady = async (): Promise<void> => {
+    if (runtimeAssetsReady) return;
+    await deps.ensureRuntimePluginReady(context);
+    runtimeAssetsReady = true;
+  };
+
+  if (!args.target && args.useRofi) {
+    await ensureRuntimeAssetsReady();
   }
 
   const targetChoice = await deps.chooseTarget(args, scriptPath);
@@ -266,7 +279,7 @@ export async function runPlaybackCommandWithDeps(
     );
   }
 
-  await deps.ensureRuntimePluginReady(context);
+  await ensureRuntimeAssetsReady();
 
   await deps.startMpv(
     selectedTarget.target,

launcher/commands/stats-command.ts

@@ -157,6 +157,15 @@ export async function runStatsCommand(
     if (args.statsCleanupLifetime) {
       forwarded.push('--stats-cleanup-lifetime');
     }
+    if (args.statsCleanupDuplicateLines) {
+      forwarded.push('--stats-cleanup-duplicate-lines');
+    }
+    if (args.statsCleanupDryRun) {
+      forwarded.push('--stats-cleanup-dry-run');
+    }
+    if (args.statsCleanupLookbackDays) {
+      forwarded.push('--stats-cleanup-lookback-days', String(args.statsCleanupLookbackDays));
+    }
     if (shouldForwardLogLevel(args.logLevel)) {
       forwarded.push('--log-level', args.logLevel);
     }

launcher/commands/update-command.test.ts

@@ -36,6 +36,11 @@ test('runUpdateCommand updates directly on Linux without launching Electron', as
         launcher: { status: 'updated' },
         supportAssets: [
           { status: 'updated', component: 'theme', message: 'Installed theme.' },
+          {
+            status: 'updated',
+            component: 'thumbnailer',
+            message: 'Installed rofi thumbnailer.',
+          },
           { status: 'skipped', component: 'plugin', message: 'Plugin already up to date.' },
         ],
       };
@@ -52,6 +57,7 @@ test('runUpdateCommand updates directly on Linux without launching Electron', as
     'info:AppImage update: updated',
     'info:Launcher update: updated',
     'info:Support assets (theme) update: updated - Installed theme.',
+    'info:Support assets (thumbnailer) update: updated - Installed rofi thumbnailer.',
     'info:Support assets (plugin) update: skipped - Plugin already up to date.',
   ]);
 });

launcher/commands/update-command.ts

@@ -21,7 +21,10 @@ import {
   parseSha256Sums,
   type FetchLike,
 } from '../../src/main/runtime/update/release-assets.js';
-import { updateSupportAssetsFromRelease } from '../../src/main/runtime/update/support-assets.js';
+import {
+  updateSupportAssetsFromRelease,
+  type SupportAssetsUpdateResult,
+} from '../../src/main/runtime/update/support-assets.js';
 
 type UpdateCommandResponse = {
   ok: boolean;
@@ -36,15 +39,14 @@ type DirectReleaseUpdateRequest = {
   channel: UpdateChannel;
 };
 
+type DirectSupportAssetsUpdateResult = Omit<SupportAssetsUpdateResult, 'status'> & {
+  status: string;
+};
+
 type DirectReleaseUpdateResult = {
   appImage: { status: string; command?: string; message?: string };
   launcher: { status: string; command?: string; message?: string };
-  supportAssets: Array<{
-    status: string;
-    component?: 'theme' | 'plugin';
-    command?: string;
-    message?: string;
-  }>;
+  supportAssets: DirectSupportAssetsUpdateResult[];
 };
 
 type UpdateCommandDeps = {
@@ -129,12 +131,7 @@ function readUpdateChannel(root: Record<string, unknown> | null): UpdateChannel
 
 function logUpdateResult(
   label: string,
-  result: {
-    status: string;
-    component?: 'theme' | 'plugin';
-    command?: string;
-    message?: string;
-  },
+  result: DirectSupportAssetsUpdateResult,
   configuredLogLevel: NonNullable<LauncherCommandContext['args']['logLevel']>,
   deps: Pick<UpdateCommandDeps, 'log'>,
 ): void {

launcher/config/args-normalizer.test.ts

@@ -120,6 +120,7 @@ test('applyInvocationsToArgs maps config and jellyfin invocation state', () => {
       logLevel: 'warn',
     },
     settingsInvocation: null,
+    animeInvocation: null,
     mpvInvocation: null,
     appInvocation: null,
     dictionaryTriggered: false,
@@ -134,6 +135,9 @@ test('applyInvocationsToArgs maps config and jellyfin invocation state', () => {
     statsCleanup: false,
     statsCleanupVocab: false,
     statsCleanupLifetime: false,
+    statsCleanupDuplicateLines: false,
+    statsCleanupDryRun: false,
+    statsCleanupLookbackDays: null,
     statsLogLevel: null,
     syncTriggered: false,
     syncCliTokens: [],
@@ -171,6 +175,7 @@ test('applyInvocationsToArgs maps settings invocation to settings window', () =>
     settingsInvocation: {
       logLevel: undefined,
     },
+    animeInvocation: null,
     mpvInvocation: null,
     appInvocation: null,
     dictionaryTriggered: false,
@@ -185,6 +190,9 @@ test('applyInvocationsToArgs maps settings invocation to settings window', () =>
     statsCleanup: false,
     statsCleanupVocab: false,
     statsCleanupLifetime: false,
+    statsCleanupDuplicateLines: false,
+    statsCleanupDryRun: false,
+    statsCleanupLookbackDays: null,
     statsLogLevel: null,
     syncTriggered: false,
     syncCliTokens: [],
@@ -215,6 +223,7 @@ test('applyInvocationsToArgs fails when config invocation has no action', () =>
         action: undefined,
       },
       settingsInvocation: null,
+      animeInvocation: null,
       mpvInvocation: null,
       appInvocation: null,
       dictionaryTriggered: false,
@@ -229,6 +238,9 @@ test('applyInvocationsToArgs fails when config invocation has no action', () =>
       statsCleanup: false,
       statsCleanupVocab: false,
       statsCleanupLifetime: false,
+      statsCleanupDuplicateLines: false,
+      statsCleanupDryRun: false,
+      statsCleanupLookbackDays: null,
       statsLogLevel: null,
       syncTriggered: false,
       syncCliTokens: [],
@@ -257,6 +269,7 @@ test('applyInvocationsToArgs maps texthooker browser-open request', () => {
     jellyfinInvocation: null,
     configInvocation: null,
     settingsInvocation: null,
+    animeInvocation: null,
     mpvInvocation: null,
     appInvocation: null,
     dictionaryTriggered: false,
@@ -271,6 +284,9 @@ test('applyInvocationsToArgs maps texthooker browser-open request', () => {
     statsCleanup: false,
     statsCleanupVocab: false,
     statsCleanupLifetime: false,
+    statsCleanupDuplicateLines: false,
+    statsCleanupDryRun: false,
+    statsCleanupLookbackDays: null,
     statsLogLevel: null,
     syncTriggered: false,
     syncCliTokens: [],

launcher/config/args-normalizer.ts

@@ -162,12 +162,15 @@ export function createDefaultArgs(
     statsCleanup: false,
     statsCleanupVocab: false,
     statsCleanupLifetime: false,
+    statsCleanupDuplicateLines: false,
+    statsCleanupDryRun: false,
     doctor: false,
     doctorRefreshKnownWords: false,
     logsExport: false,
     version: false,
     update: false,
     settings: false,
+    animeBrowser: false,
     configPath: false,
     configShow: false,
     mpvIdle: false,
@@ -258,6 +261,11 @@ export function applyInvocationsToArgs(parsed: Args, invocations: CliInvocations
   if (invocations.statsCleanup) parsed.statsCleanup = true;
   if (invocations.statsCleanupVocab) parsed.statsCleanupVocab = true;
   if (invocations.statsCleanupLifetime) parsed.statsCleanupLifetime = true;
+  if (invocations.statsCleanupDuplicateLines) parsed.statsCleanupDuplicateLines = true;
+  if (invocations.statsCleanupDryRun) parsed.statsCleanupDryRun = true;
+  if (invocations.statsCleanupLookbackDays !== null) {
+    parsed.statsCleanupLookbackDays = invocations.statsCleanupLookbackDays;
+  }
   if (invocations.dictionaryTarget) {
     parsed.dictionaryTarget = parseDictionaryTarget(invocations.dictionaryTarget);
   } else if (
@@ -348,6 +356,12 @@ export function applyInvocationsToArgs(parsed: Args, invocations: CliInvocations
       );
   }
 
+  if (invocations.animeInvocation) {
+    if (invocations.animeInvocation.logLevel) {
+      parsed.logLevel = parseLogLevel(invocations.animeInvocation.logLevel);
+    }
+    parsed.animeBrowser = true;
+  }
   if (invocations.settingsInvocation) {
     if (invocations.settingsInvocation.logLevel) {
       parsed.logLevel = parseLogLevel(invocations.settingsInvocation.logLevel);

launcher/config/cli-parser-builder.ts

@@ -23,6 +23,7 @@ export interface CliInvocations {
   jellyfinInvocation: JellyfinInvocation | null;
   configInvocation: CommandActionInvocation | null;
   settingsInvocation: CommandActionInvocation | null;
+  animeInvocation: CommandActionInvocation | null;
   mpvInvocation: CommandActionInvocation | null;
   appInvocation: { appArgs: string[] } | null;
   dictionaryTriggered: boolean;
@@ -37,6 +38,9 @@ export interface CliInvocations {
   statsCleanup: boolean;
   statsCleanupVocab: boolean;
   statsCleanupLifetime: boolean;
+  statsCleanupDuplicateLines: boolean;
+  statsCleanupDryRun: boolean;
+  statsCleanupLookbackDays: number | null;
   statsLogLevel: string | null;
   syncTriggered: boolean;
   syncCliTokens: string[];
@@ -53,6 +57,16 @@ export interface CliInvocations {
   texthookerOpenBrowser: boolean;
 }
 
+/** `--lookback-days` narrows the duplicate-line cleanup; fractions are floored. */
+function parseStatsLookbackDays(value: unknown): number | null {
+  if (typeof value !== 'string' && typeof value !== 'number') return null;
+  const days = Number(value);
+  if (!Number.isFinite(days) || days < 1) {
+    throw new Error('Stats --lookback-days must be at least one day.');
+  }
+  return Math.floor(days);
+}
+
 function applyRootOptions(program: Command): void {
   program
     .option(
@@ -102,6 +116,7 @@ function getTopLevelCommand(argv: string[]): { name: string; index: number } | n
     'doctor',
     'config',
     'settings',
+    'anime',
     'mpv',
     'logs',
     'dictionary',
@@ -155,6 +170,7 @@ export function parseCliPrograms(
   let jellyfinInvocation: JellyfinInvocation | null = null;
   let configInvocation: CommandActionInvocation | null = null;
   let settingsInvocation: CommandActionInvocation | null = null;
+  let animeInvocation: CommandActionInvocation | null = null;
   let mpvInvocation: CommandActionInvocation | null = null;
   let appInvocation: { appArgs: string[] } | null = null;
   let dictionaryTriggered = false;
@@ -169,6 +185,9 @@ export function parseCliPrograms(
   let statsCleanup = false;
   let statsCleanupVocab = false;
   let statsCleanupLifetime = false;
+  let statsCleanupDuplicateLines = false;
+  let statsCleanupDryRun = false;
+  let statsCleanupLookbackDays: number | null = null;
   let statsLogLevel: string | null = null;
   let syncTriggered = false;
   let syncCliTokens: string[] = [];
@@ -269,6 +288,9 @@ export function parseCliPrograms(
     .option('-s, --stop', 'Stop the background stats server')
     .option('-v, --vocab', 'Clean vocabulary rows in the stats database')
     .option('-l, --lifetime', 'Rebuild lifetime summary rows from retained data')
+    .option('-d, --duplicate-lines', 'Collapse repeated subtitle lines from typeset animations')
+    .option('--dry-run', 'Report what a cleanup would remove without changing anything')
+    .option('--lookback-days <days>', 'Only clean lines recorded in the last N days')
     .option('--log-level <level>', 'Log level')
     .action((action: string | undefined, options: Record<string, unknown>) => {
       statsTriggered = true;
@@ -289,13 +311,35 @@ export function parseCliPrograms(
       if (normalizedAction && (statsBackground || statsStop)) {
         throw new Error('Stats background and stop flags cannot be combined with stats actions.');
       }
-      if (normalizedAction !== 'cleanup' && (options.vocab === true || options.lifetime === true)) {
-        throw new Error('Stats --vocab and --lifetime flags require the cleanup action.');
+      if (
+        normalizedAction !== 'cleanup' &&
+        (options.vocab === true || options.lifetime === true || options.duplicateLines === true)
+      ) {
+        throw new Error(
+          'Stats --vocab, --lifetime and --duplicate-lines flags require the cleanup action.',
+        );
+      }
+      if (
+        options.duplicateLines !== true &&
+        (options.dryRun === true || options.lookbackDays !== undefined)
+      ) {
+        throw new Error('Stats --dry-run and --lookback-days require --duplicate-lines.');
       }
       if (normalizedAction === 'cleanup') {
         statsCleanup = true;
         statsCleanupLifetime = options.lifetime === true;
-        statsCleanupVocab = statsCleanupLifetime ? false : options.vocab !== false;
+        statsCleanupDuplicateLines = options.duplicateLines === true;
+        const explicitModeCount = [options.vocab, options.lifetime, options.duplicateLines].filter(
+          (value) => value === true,
+        ).length;
+        if (explicitModeCount > 1) {
+          throw new Error('Stats cleanup runs one mode at a time.');
+        }
+        // Vocabulary cleanup stays the default so `stats cleanup` keeps its old meaning.
+        statsCleanupVocab =
+          statsCleanupLifetime || statsCleanupDuplicateLines ? false : options.vocab !== false;
+        statsCleanupDryRun = options.dryRun === true;
+        statsCleanupLookbackDays = parseStatsLookbackDays(options.lookbackDays);
       } else if (normalizedAction === 'rebuild' || normalizedAction === 'backfill') {
         statsCleanup = true;
         statsCleanupLifetime = true;
@@ -415,6 +459,16 @@ export function parseCliPrograms(
       };
     });
 
+  commandProgram
+    .command('anime')
+    .description('Open the anime browser window')
+    .option('--log-level <level>', 'Log level')
+    .action((options: Record<string, unknown>) => {
+      animeInvocation = {
+        logLevel: typeof options.logLevel === 'string' ? options.logLevel : undefined,
+      };
+    });
+
   commandProgram
     .command('mpv')
     .description('MPV helpers')
@@ -469,6 +523,7 @@ export function parseCliPrograms(
       jellyfinInvocation,
       configInvocation,
       settingsInvocation,
+      animeInvocation,
       mpvInvocation,
       appInvocation,
       dictionaryTriggered,
@@ -483,6 +538,9 @@ export function parseCliPrograms(
       statsCleanup,
       statsCleanupVocab,
       statsCleanupLifetime,
+      statsCleanupDuplicateLines,
+      statsCleanupDryRun,
+      statsCleanupLookbackDays,
       statsLogLevel,
       syncTriggered,
       syncCliTokens,

launcher/jellyfin.test.ts

@@ -57,6 +57,7 @@ function createArgs(): Args {
     logsExport: false,
     version: false,
     settings: false,
+    animeBrowser: false,
     configPath: false,
     configShow: false,
     mpvIdle: false,

launcher/main.test.ts

@@ -73,20 +73,21 @@ function makeTestEnv(homeDir: string, xdgConfigHome: string): NodeJS.ProcessEnv
   };
 }
 
-// On Linux the playback path runs `ensureLinuxRuntimePluginAvailable`, which —
-// when the runtime plugin/theme are missing — spawns the app with
-// `--ensure-linux-runtime-plugin-assets` and polls up to 30s
+// On Linux the playback path runs `ensureLinuxRuntimePluginAvailable`, which
+// spawns the app with `--ensure-linux-runtime-plugin-assets` when managed
+// support assets are missing and polls up to 30s
 // (RESPONSE_TIMEOUT_MS) for an install response. A fake app that just exits
 // never writes that response, so the launcher hangs and the test times out on
 // Linux CI (the preflight is a no-op on macOS/Windows). This shell prelude makes
-// the fake app install the managed plugin/theme and write the response, matching
+// the fake app install the managed support assets and write the response, matching
 // launcher/smoke.e2e.test.ts. Prepend it to each fake app that reaches playback.
 const RUNTIME_PLUGIN_PREFLIGHT_SH = `if [ "$1" = "--ensure-linux-runtime-plugin-assets" ]; then
   data="\${XDG_DATA_HOME:-$HOME/.local/share}/SubMiner"
-  mkdir -p "$data/plugin/subminer" "$data/themes"
+  mkdir -p "$data/plugin/subminer" "$data/themes" "$data/thumbnailers"
   printf -- '-- test plugin\\n' > "$data/plugin/subminer/main.lua"
   printf 'test=true\\n' > "$data/plugin/subminer.conf"
   printf '/* test theme */\\n' > "$data/themes/subminer.rasi"
+  printf '[Thumbnailer Entry]\\n' > "$data/thumbnailers/subminer-ffmpegthumbnailer.thumbnailer"
   if [ "$2" = "--ensure-linux-runtime-plugin-assets-response-path" ] && [ -n "$3" ]; then
     mkdir -p "$(dirname "$3")"
     printf '{"ok":true,"status":"installed","path":"%s"}' "$data/plugin/subminer/main.lua" > "$3"

launcher/mpv.test.ts

@@ -627,6 +627,7 @@ function makeArgs(overrides: Partial<Args> = {}): Args {
     logsExport: false,
     version: false,
     settings: false,
+    animeBrowser: false,
     configPath: false,
     configShow: false,
     mpvIdle: false,

launcher/parse-args.test.ts

@@ -232,6 +232,75 @@ test('parseArgs maps lifetime stats cleanup flag', () => {
   assert.equal(parsed.statsCleanupLifetime, true);
 });
 
+test('parseArgs maps duplicate-line stats cleanup flags', () => {
+  const parsed = parseArgs(
+    ['stats', 'cleanup', '--duplicate-lines', '--dry-run', '--lookback-days', '30'],
+    'subminer',
+    {},
+  );
+
+  assert.equal(parsed.statsCleanup, true);
+  assert.equal(parsed.statsCleanupVocab, false);
+  assert.equal(parsed.statsCleanupDuplicateLines, true);
+  assert.equal(parsed.statsCleanupDryRun, true);
+  assert.equal(parsed.statsCleanupLookbackDays, 30);
+
+  const fractional = parseArgs(
+    ['stats', 'cleanup', '--duplicate-lines', '--lookback-days', '1.5'],
+    'subminer',
+    {},
+  );
+  assert.equal(fractional.statsCleanupLookbackDays, 1);
+});
+
+test('parseArgs rejects duplicate-line flags without the duplicate-lines mode', () => {
+  const error = withProcessExitIntercept(() => {
+    parseArgs(['stats', 'cleanup', '--dry-run'], 'subminer', {});
+  });
+
+  assert.equal(error.code, 1);
+  assert.match(error.stderr, /--dry-run and --lookback-days require --duplicate-lines/);
+});
+
+test('parseArgs rejects an empty lookback value outside duplicate-line cleanup', () => {
+  const error = withProcessExitIntercept(() => {
+    parseArgs(['stats', '--lookback-days', ''], 'subminer', {});
+  });
+
+  assert.equal(error.code, 1);
+  assert.match(error.stderr, /--dry-run and --lookback-days require --duplicate-lines/);
+});
+
+test('parseArgs rejects combining explicit cleanup modes', () => {
+  for (const modes of [
+    ['--lifetime', '--duplicate-lines'],
+    ['--vocab', '--duplicate-lines'],
+    ['--vocab', '--lifetime'],
+  ]) {
+    const error = withProcessExitIntercept(() => {
+      parseArgs(['stats', 'cleanup', ...modes], 'subminer', {});
+    });
+
+    assert.equal(error.code, 1);
+    assert.match(error.stderr, /Stats cleanup runs one mode at a time/);
+  }
+});
+
+test('parseArgs rejects unusable lookback windows', () => {
+  for (const value of ['0', '0.5', '-5', 'soon']) {
+    const error = withProcessExitIntercept(() => {
+      parseArgs(
+        ['stats', 'cleanup', '--duplicate-lines', '--lookback-days', value],
+        'subminer',
+        {},
+      );
+    });
+
+    assert.equal(error.code, 1);
+    assert.match(error.stderr, /--lookback-days must be at least one day/);
+  }
+});
+
 test('parseArgs rejects cleanup-only stats flags without cleanup action', () => {
   const error = withProcessExitIntercept(() => {
     parseArgs(['stats', '--vocab'], 'subminer', {});
@@ -239,7 +308,10 @@ test('parseArgs rejects cleanup-only stats flags without cleanup action', () =>
 
   assert.equal(error.code, 1);
   assert.match(error.message, /exit:1/);
-  assert.match(error.stderr, /Stats --vocab and --lifetime flags require the cleanup action/);
+  assert.match(
+    error.stderr,
+    /Stats --vocab, --lifetime and --duplicate-lines flags require the cleanup action/,
+  );
 });
 
 test('parseArgs maps stats rebuild action to cleanup lifetime mode', () => {

launcher/picker.test.ts

@@ -3,7 +3,12 @@ import assert from 'node:assert/strict';
 import fs from 'node:fs';
 import path from 'node:path';
 import os from 'node:os';
-import { findRofiTheme, formatRofiPrompt } from './picker';
+import {
+  findRofiTheme,
+  findRofiThumbnailerDataRoot,
+  formatRofiPrompt,
+  prependXdgDataDir,
+} from './picker';
 
 // ── formatRofiPrompt: spacing between prompt and input field ──────────────────
 
@@ -23,6 +28,7 @@ test('formatRofiPrompt leaves an empty prompt empty', () => {
 // ── findRofiTheme: Linux packaged path discovery ──────────────────────────────
 
 const ROFI_THEME_FILE = 'subminer.rasi';
+const ROFI_THUMBNAILER_FILE = 'subminer-ffmpegthumbnailer.thumbnailer';
 
 function makeFile(filePath: string): void {
   fs.mkdirSync(path.dirname(filePath), { recursive: true });
@@ -121,3 +127,42 @@ test('findRofiTheme resolves ~/.local/share/SubMiner/themes/subminer.rasi when X
     fs.rmSync(baseDir, { recursive: true, force: true });
   }
 });
+
+test('findRofiThumbnailerDataRoot resolves the managed XDG data root', () => {
+  const xdgDataHome = fs.mkdtempSync(path.join(os.tmpdir(), 'subminer-test-xdg-'));
+  const originalXdgDataHome = process.env.XDG_DATA_HOME;
+  try {
+    process.env.XDG_DATA_HOME = xdgDataHome;
+    const dataRoot = path.join(xdgDataHome, 'SubMiner');
+    makeFile(path.join(dataRoot, 'thumbnailers', ROFI_THUMBNAILER_FILE));
+
+    const result = withPlatform('linux', () => findRofiThumbnailerDataRoot('/usr/bin/subminer'));
+    assert.equal(result, dataRoot);
+  } finally {
+    if (originalXdgDataHome === undefined) {
+      delete process.env.XDG_DATA_HOME;
+    } else {
+      process.env.XDG_DATA_HOME = originalXdgDataHome;
+    }
+    fs.rmSync(xdgDataHome, { recursive: true, force: true });
+  }
+});
+
+test('findRofiThumbnailerDataRoot is Linux-only', () => {
+  assert.equal(
+    withPlatform('darwin', () => findRofiThumbnailerDataRoot('/usr/bin/subminer')),
+    null,
+  );
+});
+
+test('prependXdgDataDir preserves existing roots and avoids duplicates', () => {
+  const root = '/tmp/subminer-data';
+  assert.equal(
+    prependXdgDataDir(root, `/opt/share${path.delimiter}${root}${path.delimiter}/usr/share`),
+    `${root}${path.delimiter}/opt/share${path.delimiter}/usr/share`,
+  );
+  assert.equal(
+    prependXdgDataDir(root),
+    `${root}${path.delimiter}/usr/local/share${path.delimiter}/usr/share`,
+  );
+});

launcher/picker.ts

@@ -159,6 +159,9 @@ interface RofiIconEntry {
   iconPath?: string;
 }
 
+const ROFI_THUMBNAILER_FILE = 'subminer-ffmpegthumbnailer.thumbnailer';
+const DEFAULT_XDG_DATA_DIRS = ['/usr/local/share', '/usr/share'];
+
 function showRofiIconMenu(
   entries: RofiIconEntry[],
   prompt: string,
@@ -389,6 +392,47 @@ export function findRofiTheme(scriptPath: string): string | null {
   return null;
 }
 
+export function findRofiThumbnailerDataRoot(scriptPath: string): string | null {
+  if (process.platform !== 'linux') return null;
+
+  const scriptDir = path.dirname(realpathMaybe(scriptPath));
+  const xdgDataHome = process.env.XDG_DATA_HOME || path.join(os.homedir(), '.local/share');
+  const roots = [
+    path.join(xdgDataHome, 'SubMiner'),
+    path.posix.join('/usr/local/share/SubMiner'),
+    path.posix.join('/usr/share/SubMiner'),
+    path.join(scriptDir, 'assets'),
+    path.join(scriptDir, '..', 'assets'),
+  ];
+
+  for (const root of roots) {
+    if (fs.existsSync(path.join(root, 'thumbnailers', ROFI_THUMBNAILER_FILE))) {
+      return root;
+    }
+  }
+
+  return null;
+}
+
+export function prependXdgDataDir(dataRoot: string, currentValue?: string): string {
+  const currentDirs = currentValue
+    ? currentValue.split(path.delimiter).filter(Boolean)
+    : DEFAULT_XDG_DATA_DIRS;
+  return [dataRoot, ...currentDirs.filter((candidate) => candidate !== dataRoot)].join(
+    path.delimiter,
+  );
+}
+
+function buildRofiThumbnailEnvironment(scriptPath: string): NodeJS.ProcessEnv {
+  if (!commandExists('ffmpegthumbnailer')) return process.env;
+  const dataRoot = findRofiThumbnailerDataRoot(scriptPath);
+  if (!dataRoot) return process.env;
+  return {
+    ...process.env,
+    XDG_DATA_DIRS: prependXdgDataDir(dataRoot, process.env.XDG_DATA_DIRS),
+  };
+}
+
 export function showRofiMenu(
   videos: string[],
   dir: string,
@@ -420,6 +464,7 @@ export function showRofiMenu(
   const result = spawnSync('rofi', args, {
     input: buildRofiMenu(videos, dir, recursive),
     encoding: 'utf8',
+    env: buildRofiThumbnailEnvironment(scriptPath),
     stdio: ['pipe', 'pipe', 'ignore'],
   });
   if (result.error) {