SplitStaff Glob Detector MVP

Local browser prototype: upload Jamcorder-style MIDI, view a piano roll, detect visual “globs,” flag cascades/chord masses/close-register behavior separation, add a manual chord map, and export an analysis packet for LLM review.

Notes0

Tracks0

TPQ0

Meter4/4

1. Input

Upload MIDI Load demo

No MIDI loaded yet. Files are read locally in this browser.

---

Export analysis JSON

This JSON is the main handoff file: notes, chord map, windows, glob labels, and evidence.

2. Manual chord map

MVP assumption: chord changes are manually entered. Use bars separated by |. Bar numbers at line starts are okay.

1 | Ebmaj7 | Cm7 F7 | Fm7 Bb7 | Ebmaj7 | 5 | Abmaj7 | Am7b5 D7 | Gm7 C7 | Fm7 Bb7 |

3. Detection settings

Window beats Min glob notes

Glob pitch gap, semitones Glob time gap, beats Separate-glob pitch gap, semitones Chord onset tolerance, beats Layer rate ratio Layer duration ratio

What this MVP does

• Detects visually connected note masses: globs.
• Flags two separated globs as likely separate layer/hand candidates.
• Flags one glob as possible one hand, joined hands, or close-register overlap.
• Looks inside one glob for different rhythmic/duration/contour behavior.
• Exports structured evidence for AI analysis.

Piano roll

separate globs behavior-separated chord mass cascade/run ambiguous

Upload a MIDI file or load the demo.

Selected window

Click a colored strip under the piano roll, or click a row below.

Detected windows

Measures	Notes	Globs	Reading	Confidence
No analysis yet.

Export packet preview

No packet yet.