AbstractMusic Backlog

This folder is the engineering memory for AbstractMusic.

The backlog is intentionally code-first: before implementing or changing an item, inspect the current code and docs because this project is moving quickly and older notes may be stale.

Layout

• overview.md: status, priority order, and work ledger.
• planned/: committed work that should be implemented.
• proposed/: useful ideas that need more evidence before implementation.
• completed/: finished work with completion reports.
• deprecated/: retired work with deprecation reports.
• recurrent/: checklist tasks that should run after relevant changes.

Rules

• Keep planned items standalone.
• Preserve the model/provider abstraction; do not hard-code one model into the public API.
• Keep heavy local inference dependencies optional where possible.
• Do not commit model weights, generated WAVs, Python bytecode, or cache artifacts.
• Prefer permissive, commercially usable model licenses for default providers.
• Never silently ignore generation inputs such as lyrics, duration, seed, model choice, fallback, or truncation.
• Validate with fast unit tests and at least one real model smoke path before calling provider work complete.