Track Review
Automated critique of your track, with fixes you can hear before you accept them.
Track Review analyses what you have written and tells you what is working against you — notes that clash, parts that fight for the same space, arrangements that never change. Crucially, each finding comes with fixes you can hear before deciding, rather than advice you have to implement yourself.
Two layers
Review is built in two layers, and understanding the split explains what you can trust.
Layer 1 — the analysis
A deterministic engine reads your track and produces findings. Each finding is tied to a specific beat, carries a severity, states the musical principle behind it, and comes with up to four suggested fixes.
This layer involves no AI. It analyses the actual notes in your document, so a finding is always about something genuinely present in your track.
Layer 2 — the coach
An optional AI layer that reasons only over the findings the engine produced. It names what your loop seems to be trying to do and explains which fix serves that intent.
The coach is constrained to the verified findings — anything it produces that is not grounded in them is discarded rather than shown. It cannot invent a problem your track does not have.
Auditioning fixes
Clicking a suggested fix applies it immediately, as a real edit, so you hear the result in context. Findings then recompute against the changed track.
When several fixes address the same problem, they appear as a pick-one group alongside an Original option, so you can cycle through the alternatives and compare. When you have decided, Keep this one commits it — or Keep the original discards the experiment.
Tip: Audition with the track playing. A fix that reads as obviously correct often sounds wrong in context, which is the entire reason the feature works this way.
Counterpoint findings
Where two parts clash, the review reports the specific collision, and its fixes re-pitch the offending notes — usually the most useful category of finding in a track with more than a couple of melodic parts.
Requirements
Layer 1 works for everyone signed in and requires no API key. The coach needs a server-side Anthropic API key configured for the deployment; without one it silently stays out of the way and you still get the full deterministic review.
Where to go next
- Learnings — the record of the fixes you kept.