OpenContinuity operates from OpenEnvironment Data Plane.
OpenEnvironment collects provider observations for the selected operating location using documented schedules and observed report arrivals. OpenContinuity trains and evaluates challengers from fixed snapshots of that history, and provides labeled forecasts when observations become unavailable.
Authoritative local environmental store
Checking provider timing…
Learning a provider’s publication timing is not an outage. A successful fetch confirms delivery; observation timestamps determine data freshness.
Trust boundaries
One authentic store.OpenAtmosphere, OpenOcean, OpenSpace, and OpenContinuity consume the OpenEnvironment record.
Snapshots are verified before training.Manifest hash, immutable marker, domain, tier, truth mode, quality, and record sequence must agree.
Predictions never train the model.Synthetic, replayed, virtual, and model-predicted records are rejected.
Qualified is not promoted.Baseline, champion, and interval-coverage gates produce a recommendation. An operator promotes.
Train forecasts for the selected operating location.
Training uses a fixed snapshot of measured observations from OpenEnvironment. Results show the usable sample count, forecast error, and evaluation outcome. You decide whether to activate a qualified model.
Review and activate trained models.
The initial champion is the conservative physics baseline. Every learned model retains parent, snapshot, cutoff, validation metrics, SHA-256, and rollback lineage.
From provider observations to forecasts.
Collection cadence, model inference cadence, and retraining cadence remain independent.