v1.3.0

Highlights

• ChapKit integration for chap eval: Auto-detect and evaluate models hosted via ChapKit services, with full CLI support for remote model evaluation.
• Conda runner: Run models in conda-based environments, expanding beyond Docker-only workflows.
• Winkler score metric: New prediction interval evaluation metric for more nuanced model assessment.

Features

• Add KMeans clustering to SeasonalityComparisonPlot
• Add minimal weekly debug dataset for REST API testing
• Add build metadata, server time, and docker flag to /system/info
• Add GHCR-based compose file and publish worker image
• Add force-restart Makefile target for docker compose
• Add uv2 versions for monthly and weekly AR models with pinned deps

Bug Fixes

• Normalize period formats to dashed style in convert request
• Handle integer time_period values in TimePeriod.parse()
• Validate empty provided_data in _read_dataset
• Fix ChapKit CLI eval crashes from missing model_information and geo serialization
• Add None check for pycountry lookup in geometry.py
• Set platform to linux/amd64 for GHCR compose services
• Use load_redis() instead of hardcoded Redis hostname in celery_tasks
• Add netcdf4 dependency for xarray NetCDF file support
• Use TimePeriod.parse() for date parsing in plot utilities
• Propagate stride parameter in backtest function

Refactoring

• Simplify seasonal correlation plot to only show max
• Remove deprecated Pydantic/FastAPI patterns
• Remove disabled model template endpoints
• Move common endpoints to root app, reorganize API tags
• Remove chap serve and chap init CLI commands
• Remove seed, legacy router, debug router, and dead code
• Enable additional ruff lint rules

Chore

• Dependency upgrades (pyarrow, ruff, optuna, uvicorn, cryptography, etc.)