Audition a melody-free bed stem, and the full 8-bar Midnight Terminal theme

[schmug]

Task

Export two pieces of audio that the music integration was built to accept but that were never supplied, then audition each against a real episode. Both are explicitly deferred follow-ups from the original handoffs, not defects.

Context

render.py already supports both shapes — "mode": "bed" (a loopable theme ducked under the bookends) and "mode": "sting" (a short signature that never overlaps speech). See skills/daily-podcast/SKILL.md § Intro/outro music and skills/frontier-commits/SKILL.md § Intro/outro sting. No renderer change is needed for either item below; both are asset work plus a config change.

1. A melody-free bed stem for Pixel Window (daily show)

Today's daily mix retains the full melody under the intro and sign-off narration at duck_db: -18.0, because no bed-only stem was ever produced. The handoff was explicit about not claiming otherwise, and skills/daily-podcast/assets/music/PROVENANCE.json records it.

Export a stem from skills/daily-podcast/assets/music/pixel-window.strudel with the lead line removed (the square and sparkle triangle voices), keeping pads and bass. Audition it against a real episode and compare to the current mix.

2. The full 8-bar Midnight Terminal theme + its bed stem (Frontier Commits)

The bundled midnight-terminal.wav is one bar (2.307688 s), used as a sting. The full composition is 8 bars. Per skills/frontier-commits/assets/music/PROVENANCE.json:

• Export cycles 0 through 8 from midnight-terminal.strudel with Multi Channel Orbits OFF; expected length 18.461538 s.
• For the narration bed, set theme = 0 in that source and export the same range.

With both, Frontier Commits could move from sting to bed. Proposed (not approved) bed settings from the handoff: 104 BPM, one bar 2.307692 s, opening lead one bar, closing tail two bars (4.615385 s).

How to audition without shipping

Copy a past workdir, add the music block to its manifest, and dry-run — this is exactly how the current mixes were auditioned:

python3 skills/daily-podcast/render.py --manifest <workdir>/manifest.json --workdir <workdir> --dry-run

Use absolute paths (a relative --workdir breaks the ffmpeg concat list). If the workdir predates 2026-09-04, its TTS sidecars carry the pre-#206 cache key and will miss once — re-key them rather than paying for a fresh render.

Constraints

• Do not change render.py's music layer; both modes already exist and are tested (tests/test_music.py).
• Do not publish, and do not change either show's live music block until Cory has approved the new sound.
• New assets go under the owning show's skills/<show>/assets/music/ with their hash and provenance recorded, and must clear the 2048 KB check-added-large-files hook.

Acceptance criteria

• Each new stem is bundled with a PROVENANCE.json entry (hash, source, export settings).
• Opening/closing previews produced from a real episode for each, for Cory to approve.
• A recommendation on whether to switch either show, with the measured numbers behind it.

Out of scope

• Sample licensing — tracked separately in Resolve drum-sample licensing for the two bundled music assets #225.
• Any change to slugs, RSS identifiers, ship mode, or schedules.