list|tag:x field with emitTag silently drops the authored tagsresolveSequence's emit branch walks the nodes a field collected and skips anything that is not a list (packages/runes/src/lib/resolver.ts:194):
Resolve this body against a greedy field matching list|tag:track, with an itemModel and emitTag: 'track':
- Song A - Song B {% track name="Song C" /%} - Song D
Four track tags, in document order: the three list-derived ones and the authored Song C. Both syntaxes are the same thing after conversion, which is the point of emitTag.
names: [ 'Song A', 'Song B', 'Song D' ]
Song C is gone. No error, no warning, no build diagnostic.
This is what stops playlist from using the mechanism that already exists.
cast (plugins/business/src/tags/cast.ts:92-116) gets the dual-syntax composition almost free — a members field matching list with emitTag: 'cast-member', a separate items field matching tag, and a three-line merge because both arrive as the same kind of node:
const allMembers = [...asNodes(resolved.members), ...asNodes(resolved.items)];
playlist cannot do that. It declares match: 'list|tag:track' on one field without emitTag, then spends roughly 55 lines (plugins/media/src/tags/playlist.ts:316-353) reconciling two representations by hand — a listCursor consuming tracksData by list length, a transform-time adoptNestedTrack to retro-fit the authored tags into the parent's conventions, and a defensive tail for "a shape the resolver collected but this loop did not recognise".
Its own content model records the cost:
tag:trackalongsidelistso the composition the docs promise works. Greedy because the two forms may alternate, and greedy collection is consecutive: prose between two tracks ends the run and pushes the rest tobody.
major because the failure is silent content loss, the same class as BUG-028. A rune that refused the mixed declaration would be a smaller problem than one that quietly eats half of it.cast, audio and plot use emitTag on list-only fields, and playlist uses a mixed field without emitTag. The bug is latent, and it is latent because the obvious combination does not work — so it has shaped the code around it rather than showing up as a defect.tagNodes instead of continue-ing past it. It is already a tag, so it needs no conversion — only to keep its place. That yields a homogeneous, document-ordered array, which is strictly better than cast's concatenation (which appends all explicit tags after all list-derived ones and so loses interleaving).adoptNestedTrack exists because playlist found they were not, and that difference should be removed rather than re-encoded.emitTag. It cannot and should not, but the question exposed this.figure dropping non-media children; the same silent-content-loss classemitTag, template and the whole itemModel grammar, so neither half of the dual-syntax contract is visible