Reference Matrix
March 22, 2026 ยท View on GitHub
This file records what to learn from the existing repos without inheriting their structure.
metrust-py
Reference repo: metrust-py
Best aspect:
- scientific parity against
metpy.calc
Fastest / strongest area:
- verified calculation workflows and parity regression corpus
Carry forward:
- formula choices
- constants
- external benchmark corpus
- parity tests and validation philosophy
Do not carry forward:
- Python-first compatibility shell as the core architecture
rustmet
Reference repo: rustmet
Best aspect:
- broad operational weather platform shape
Fastest / strongest area:
- fetch + GRIB decode + serving pipeline organization
Carry forward:
- crate boundary ideas
- operational scope
- CLI/server concepts
Do not carry forward:
- duplicated foundational crates across repos
- mixed core and app concerns in one ownership model
rustdar
Reference repo: rustdar
Best aspect:
- radar-specific algorithms and workstation UX ideas
Fastest / strongest area:
- Level II rendering, meso/TVS logic, color table handling, radar-centric products
Carry forward:
- detection logic
- radar rendering concepts
- color-table-as-data-transform support
Do not carry forward:
- UI-first app boundary as the reusable core
rusbie
Reference repo: rusbie
Best aspect:
- Herbie-like data access ergonomics
Fastest / strongest area:
- source probing and parallel byte-range download ideas
Carry forward:
- model/source template logic
- cache and byte-range planning ideas
Do not carry forward:
- Python drop-in compatibility as the core design center
cfrust
Reference repo: cfrust
Best aspect:
- GRIB/CF decoding coverage and explicit compatibility testing
Fastest / strongest area:
- pure Rust decoder and fixture-driven validation
Carry forward:
- decoder logic
- packing/grid coverage expectations
- compatibility fixtures
Do not carry forward:
- cfgrib-compatible Python API as the core boundary
rustplots
Reference repo: rustplots
Best aspect:
- deterministic meteorological rendering and large visual comparison corpus
Fastest / strongest area:
- static plot generation and native render primitives
Carry forward:
- rendering primitives
- visual comparison suite
- headless output concepts
Do not carry forward:
- Python declarative plotting layer as the production core
rustmet-train
Reference repo: rustmet-train
Best aspect:
- dataset assembly ideas and benchmark generation workflows
Fastest / strongest area:
- practical pipeline ideas for turning weather events into training records
Carry forward:
- manifest ideas
- panel/channel concepts
- study and benchmark structure
Do not carry forward:
- TorNet-driven project shape as the architecture driver
- Python-heavy orchestration as the long-term implementation