Soundings 0.2.1

0.2.1 — 2026-09-29

Soundings 0.2.1 is a corrective patch for 0.2.0. It makes WebVTT speaker attribution faithful, keeps manual-enrichment drafts safe, and ships the enrichment hardening that was verified for, but missing from, the published 0.2.0 assets.

🗣️ WebVTT faithfulness

• Per-speaker attribution: A caption cue containing several voices (for example <v Alice>… then <v Bob>…) now produces one block per speaker. In 0.2.0 every line in such a cue was attributed to the first speaker.
• Header metadata: Files with Kind: or Language: lines under WEBVTT (common in YouTube captions) are now eligible instead of unreadable.
• Cue separators: A line containing only spaces between cues now separates them; it no longer merges two cues into one block.
• Character references: < now renders as < instead of being decoded twice.

✍️ Manual enrichment

• If publishing a companion note does not succeed (for example because the transcript note changed or the destination appeared), the form stays open with everything you typed. After a source change, Soundings re-checks the note before you review again.
• Source notes are identified only from their exact Soundings frontmatter (type, source, and a supported soundings_version), including notes whose metadata contains ---.
• The companion backlink uses the full vault path, so same-named notes in different folders link correctly.

🔒 Release integrity

• Releases are built by a read-only job and published by a separate job with GitHub artifact attestations for main.js, manifest.json, and styles.css.
• Only bare version tags (for example 0.2.1) publish releases, matching what Obsidian installs.

Notes

• Existing notes are never modified. Only new conversions of multi-voice cues use the new per-speaker layout.