Known limits

An honest list, so you can tell “battuta won’t do this” from “battuta can’t do this yet”.

By design — these will not change

battuta does not engrave. There are no spacing controls, no beam-angle handles, no dragging of notes or slurs, no page-layout panel. Verovio decides how notation looks; battuta decides what the notation is. If a rendering looks wrong, the fix is in the MEI or in Verovio, never in a nudged position.

Edit view is not a proof. Each measure is engraved on its own, so spacing is justified per measure rather than across the page, and spans crossing a tile edge (slurs, ties, hairpins, pedal) are drawn as broken continuations — the system-break convention, occurring more often. Page view is the ground truth.

battuta is not a DAW. Playback exists to audition edits and to verify the form.

Full MEI coverage is not the target. battuta understands Common Western notation as produced by Verovio-compatible tools. Anything else in your file is preserved verbatim rather than understood — safe, but not editable in the UI.

Not built yet

Missing	Notes
Autosave, session restore	Nothing recovers unsaved work after a crash
Close confirmation	Closing a starred tab or the window discards unsaved edits silently
File watching	External changes to an open file go unnoticed; your next save overwrites them
Splice-at-caret and overlay-as-new-layer paste	Only replace-measures is implemented
Split view	Cross-file work happens through tabs
MusicXML import/export	MEI only
SVG / PDF / print output	Page view renders, but there is no export
Selection playback, metronome, per-staff mute	The page-view player is all-or-nothing
Dynamics affecting playback volume	Notated, not performed
Lyrics, figured bass, percussion maps	Deliberately deferred
Part extraction, transposed parts	Deliberately deferred
Collaborative editing, plugins	Out of scope

Rough edges

Verovio occasionally draws an exuberant curve on a slur continuation stub in edit view. Cosmetic, upstream, and absent in page view.
The macOS .dmg is unsigned, so Gatekeeper warns on first open. See Install.
Windows has had no test round. The bundles are built by CI, but nobody has put them through their paces.
Very exotic repeat structures fall back to unexpanded playback rather than a guess.

Coming next

The two research phases in the project’s plan, in order:

Reference layers for transcription. Import MIDI (and the JSON sidecar a transcription pipeline emits) as reference tracks, drawn as translucent piano-roll bars behind the notation, aligned to each measure’s own pitch axis so a transcription error is visible at a glance. Per-track solo, mute and colour, plus selection playback against the reference grid.

OMR correction mode. Read MEI <facsimile>/<zone> data (or a sidecar from the OMR pipeline), show the source scan strip above each measure, click through between scan region and notation, tint elements by model confidence, and jump the caret through the flagged ones. Corrections can be written back as annotations so they feed the upstream model’s training data.