Roadmap

[bionicbutterfly13]

This pinned issue tracks the public roadmap for linoss-dynamics. The full document lives at docs/roadmap.md.

Where we are: v0.2.0 (released)

• Single-step oscillator dynamics (linoss_step, damped_linoss_step)
• Sequence-level scan (linoss_scan)
• Variable-dt analytic step (damped_oscillator_closed_form)
• Stability and frequency introspection (stability module)
• Bayesian filtering and smoothing (kalman_filter, rts_smoother)
• Optional [probabilistic] extra: matrix-exponential discretization + MLE parameter fitting

v0.3.0 — candidate next-version work

• PLSO (piecewise locally stationary oscillatory) regimes
• Switching state-space models for transient events
• DMD / Koopman-DMD spectral identification
• Koopman-Kalman hybrids for online tracking

v1.0 — unified oscillators + attractor basins

The synthesis goal: combine oscillator dynamics (LinOSS-family) with attractor memory (Hopfield-family) into one runtime substrate.

v1.x — spatio-temporal extensions

BioOSS-class field-shaped state. Separate spatial.* namespace.

Explicit non-goals

• Training-time JAX neural networks (use upstream tk-rusch/linoss)
• Web/runtime services
• Agent frameworks

---

Discussion welcome. PRs targeting v0.3 candidates are particularly valuable.