What's new

The first cut of the v1 cloud surface: oversee your fleet and compare runs across it. This is steps one and two of the fleet-learning loop; the model-loop phase remains on the roadmap.

Added

• Fleet registry and oversight — which edges exist, online / last-seen, software version, and their active runs.
• Like-process aggregation — group runs that instantiate the same process-class template into comparable sets across edges.
• Cross-run comparison for A/B scenario analysis within a fleet.

Changed

• Run ingestion now records the process-class template identity on every run, so aggregation has a stable join key.

Fixed

• Last-seen timestamps no longer drift when an edge reconnects after an outage.

Groundwork for fleet learning: a registry for process-class templates, and a simulation runtime that reproduces exactly.

Added

• Process-class template registry and identity on the cloud — the definitions edges instantiate runs from.
• Deterministic, fixed-timestep simulation on the edge: seedable and replayable, so a simulated run reproduces given the same seed.

Changed

• Virtual, replay, and physical sensors now share one array abstraction, so experiments no longer special-case the source of a channel.

Fixed

• Hybrid-mode runs no longer desync the simulated and real clocks under load.