[dben]

Stacked on #15. Retarget to main once that merges.

Per-show types

• Intro, Outro and Keep on the timeline are now the podcast's own type for a block, stored in a new podcast_types table. They apply to that podcast's episodes only; other shows treat the block as it was found.
• These are overrides, not copies of the block. A copy would match twice in every scan and count twice against the 40-per-hash cap on common hashes.
• Remove clears an override, and the block goes back to how it was found.
• Kept sections stay on the timeline in grey and aren't cut.
• Sections with an override, and blocks marked by hand, show a Custom (this show) pill.
• When a block is split into pieces, its overrides carry over to the pieces.

Not a repeat

• Still applies to every show (the block's own ignored type).
• Each report now lists where such blocks play (not_repeats), using an extra scan when the report is built. The shared match table is unchanged, so bumper tagging, nested-block dropping and splitting work as before.
• The timeline leaves these matches out. When an episode has any, a collapsed Not a repeat list under the sections shows them, with Undo. Undoing re-cuts every episode that lists the block.

Blocks marked by hand

• "Catch it in this show's other episodes too" saves a Local, Intro or Outro block of this show only (segments.by_hand).
• These blocks can be deleted from their card. Found blocks can't be deleted; mark them Keep or Not a repeat instead.

Re-cuts

• A per-show change re-cuts only that show's ready episodes where the block plays.
• Not a repeat, and undoing it, re-cut every show's.

Upgrade

• On first open, the store adds by_hand and podcast_types. Hand-typed intros, outros and Local blocks go back to automatic, as agreed. False positives stay.
• Existing reports keep their old cuts until each episode is next processed.

CLI

• podblock type still sets a block's type for every show.
• report --podcast and breaks apply the podcast's own types.

Testing

• All 194 tests pass; ruff is clean.
• New tests cover:
  • Keep, Intro and Remove staying within one show, using a second test show that shares the jingle and ad.
  • Not a repeat applying across shows, being listed in the report, and Undo.
  • Hand-made blocks: scoped to their show, deletable, and found blocks refused for deletion.
  • Store overrides, and carrying them to split pieces.
  • The upgrade reset.
• Checked the timeline in headless Chromium, on desktop and at phone width.

README.md, WEBAPP.md and CLI.md are updated.

---

Generated by Claude Code