GitP/How To/Draft an Episode from Session Assets

• Draft an Episode from Session Assets

  • This guide is for the person recording a GitP episode. It describes what to give the AI assistant and what the assistant prepares before publication.

  • Record and leave a note

    • Record the episode in Ableton under ~/Documents/ableton/GitP/<project folder> and export one WAV into that folder.
    • Create a Making/Music/Log entry with a link to the project and exported WAV, plus a few notes about the session.
    • Say each preset's number and name into the commentary mic as you select it.
    • Ask the AI assistant to prepare the episode from that log entry. The exported WAV is the episode recording.

  • What the AI prepares

    • Episode page and audio

      • The AI creates the session episode page using Logseq/Entity/Podcast/Episode, linking the music-making log and its project, recording, and other source pointers.
      • The AI converts the exported WAV to an MP3, uploads it to Backblaze B2 at a stable, readable URL, and verifies that the uploaded file matches the prepared MP3.
      • Follow the established WAV-to-MP3 workflow and podcast encoding best practices, including an appropriate bitrate, to keep episode files manageable. The longer GitP/A/Session/26/10/29-Thu/Asset/Audio/Halloween/mp3 had trouble loading; its file size is a suspected cause.

    • MicroFreak presets

      • Every saved preset referenced by the session gets detailed notes upstream, including presets already filed in the garden and sounds used briefly. Header metadata or an empty Notes section is a starting point.
      • The notes explain the saved export and its provenance, oscillator controls, filter, envelopes, LFO, modulation matrix, and performance context. Supported units and cited sources accompany the settings; unknown conversions and session-time changes remain explicit.
      • Existing archived exports can establish the saved baseline without changing the instrument. A factory match does not establish that the session used those exact settings. An unidentified initialized patch stays a plain description rather than an invented saved preset.
      • When a new capture is needed and authorized, protect unsaved sounds and back up the slot before any rename. Upload prepared exports and add their download links. Missing slot identification leaves the instrument unchanged.

    • Session details

      • The AI reads the captured MIDI, the music-making log, and the MicroFreak and Launchpad manuals to explain what happened in the session.
      • When the log doesn't name the presets, the AI transcribes the commentary track and takes each preset and its start time from the spoken announcement. It asks only when the commentary names none.
      • It correlates recorded MIDI CC messages with the MicroFreak manual and reads the downloaded presets for oscillator type, preset category, and other supported details. The episode page links the sources for those details.
      • The AI distinguishes recorded changes from the settings in a saved preset. It leaves unknown details out of the episode description.

    • MIDI download

      • Export the full session as a standard MIDI file containing its notes and controller changes, with the tempo aligned to the recording.
      • Upload the MIDI and add a direct download link to the episode so listeners can import it into a DAW.
      • MIDI analysis in episode notes describes how controls vary between chapters.

    • Publication PR

      • Run mise run gitpa:preset:audit --episode 'GitP/A/Session/YY/MM/DD-Day' --check from this garden to find missing preset research before publication. The check reports recorded evidence and coverage; the notes still need review for accuracy. --all reports historical gaps without blocking site deployment.
      • The AI fills in the episode's MP3 and preset links in this garden, then mirrors the episode, every referenced preset and relevant asset pages into Gitpa and prepares a publication PR.
      • The PR includes a direct link to the one episode MP3, its episode title, and a short description so the recording is easy to find and approve.

  • Check public media

    • GitP uses the public S3 download endpoint https://s3.us-east-005.backblazeb2.com/logseq-encode-garden/<page-derived filename>. Backblaze supports bucket names in the URL path.
    • Open the asset page’s URL without signing in. An MP3 should return audio/mpeg; artwork should return its image format, such as image/gif.
    • Request a small MP3 range with curl --range 0-1023 --dump-header - --output /dev/null <recording-url>. A seekable recording returns 206 with a Content-Range header.
    • In Gitpa’s preview, play and seek the recording, load the artwork, and follow preset and MIDI download links.
    • If the download host fails DNS resolution, compare its DNS answer through another resolver before changing the asset URL. A DNS failure alone says nothing about bucket access.

  • Listen and publish

    • Listen to the episode MP3 once and give it a go or a no-go.
    • For a go, merge the Gitpa PR. Gitpa's CI publishes the episode after the merge.
    • For a no-go, tell the AI to hold the episode. Any changes to the recording follow that decision.

  • AI execution references

    • The gitp-asset-gathering skill describes the assistant's responsibilities. GitP/How To/Prepare Podcast Metadata covers the existing metadata commands.
    • Session-page sync and RSS generation are available. Automatic preset capture, renaming, and MIDI enrichment still need implementation; they are part of the intended workflow above.