# Source Notes

# Source and compliance notes

MoonTrajectory is an independent MoonBit implementation for trajectory storage and replay sampling.

## What is included

- A core trajectory model
- A circular replay buffer
- A prioritized replay buffer
- A deterministic sampler for tests and demo code

## What is not included

- No copied code from upstream reinforcement-learning libraries
- No vendored third-party implementation files
- No generated code whose origin is unclear

## Design scope

The library owns the trajectory/data layer and training-batch utilities. It intentionally does not include a neural-network implementation, environment runtime, automatic differentiation engine, or device backend.

## Third-party material

- No source files are copied from another project.
- No vendored dependency or generated source is committed.
- Test fixtures are deterministic values authored in this repository; they are not downloaded datasets.
- The only runtime dependency is the MoonBit core math package declared in `moon.pkg`.
- Future imported algorithms or datasets must record upstream URL, exact license, copyright notice, modification scope, and redistribution permission.

## Future directions

- n-step transitions
- sequence sampling
- persistence / serialization
- tree-based prioritized sampling
- batch statistics for training loops