Development Roadmap

This page summarizes where the library is headed. The authoritative, regularly updated plan lives in ROADMAP.md; release history lives in CHANGELOG.md.

Current State (v2.8.0)

  • 8,200+ tests passing; coverage measured honestly (88.8% as CI’s branch-coverage gate sees it, with numba kernels traced; 93.8% locally where the MLX layer is visible), docstring examples run in CI

  • Static estimation suite (v2.8.0): nine of MATLAB’s 11 Static_Estimation localization estimators (TDOA, bistatic range-only, range-rate-only, Doppler-ratio, bearings-only, and the polynomial-measurement Fisher information matrix), built on a Macaulay null-space multivariate polynomial solver; plus the direction-cosine u-v measurement coordinate system and IGRF-14 — all validated against MATLAB fixtures, with four upstream MATLAB defects found, fixed and documented

  • MATLAB refraction suite (v2.7.0): humidity and dew point, astronomical refraction, standard-exponential-model radar refraction (bistatic r-u-v ray tracing, bias approximation, cubature conversions) and speed of sound, every function validated against fixtures captured from the MATLAB source tree

  • Audited: the post-v2.5.0 audit verified the MATLAB parity inventory against the source tree, fixed eleven wrong-answer code defects and ~60 incorrect docstrings, and left permanent gates behind – dead-parameter detection, patch coverage on every PR, executed markdown/rst examples, a complete module-maturity registry checked both ways, and weekly mutation testing over the STABLE modules

  • Typed configs and sessions (Typed Configs and Sessions): msgspec.Struct configs (IMMConfig, GaussianSumConfig, RBPFConfig, SingleTargetConfig, MultiTargetConfig) accepted via a keyword-only config=, and full state snapshot/resume (pytcl.io.save_session/load_session) for six tracker and filter classes, bit-exact resume for the four deterministic ones

  • Validated geophysics: WMM2025 (default), WMM2020, IGRF-14 (with IGRF-13 retained), and WMMHR2025 magnetic models verified against independent references to sub-nT accuracy; UTM projections verified against EPSG to sub-millimeter

  • Real-data validation: recorded ADS-B air traffic scores the tracking chain; vendored Space-Track TLE history scores SGP4/SDP4 self-prediction at 1-28 day horizons

  • GPU acceleration: dual-backend (CuPy for NVIDIA CUDA, MLX for Apple Silicon) batch Kalman and particle filters, both verified on real hardware

  • Diagnostics: opt-in observability (Diagnostics Guide) — silent by default, with gating/association/filter-health/data-file instrumentation

  • Results I/O (Results I/O): CSV/Parquet measurement ingest, to_polars() accessors, msgspec JSON/MessagePack and ASDF serialization, and AIS NMEA decoding

  • Real ship traffic: a recorded AIS capture (299 vessels) scores the tracking chain against self-broadcast speed, alongside the ADS-B and TLE fixtures

  • Track management: SQL and HDF5 persistence with lifecycle management and v1.x migration tools

  • Tooling: uv-managed workflow, ruff (lint + format), ty as the type gate, prek hooks, pinned CI toolchain

Next

  • Measured backlog: the remaining MATLAB parity gaps (MATLAB TCL parity inventory) — the cubature-point library’s dimension-specialized subdirectory files and out-of-scope region types (LCD samples and the general-dimension region-cubature rules have shipped), the refraction suite, localization estimators, and filter variants.

Long term (unscheduled)

RAPIDS and distributed tracking, ROS 2 integration, PHD/CPHD and labeled multi-Bernoulli trackers, and the exploratory items listed under “Long-Term Vision” in ROADMAP.md.

Contributing

See CONTRIBUTING.md for guidelines and the high-impact contribution areas listed in ROADMAP.md.