A monthly summary of new functionality delivered across the WP6 digital twins (blue, red and grey). Each entry links to that month’s full changelog, with deep links into the live dashboards.
| Month | Highlights |
|---|---|
| October / November 2026 | In review: LetsGrow for Red — open PR #31 syncs Neurath’s climate computer (pipes, vents, screens, lamps, irrigation, weather station; 51 sensors, ~14 M rows of history) the way Blue syncs Yookr. Red’s wires, set straight — three wires now installed and behaving; Red reads its wires from upstream’s sensor map, and July–October readings misfiled under the wrong wire are documented. Pages that stay up through an outage — fitted DLI and climate models stay in memory when the object store is down, the climate forecast stops re-downloading its model on each request (~600 ms), and a TimescaleDB restart no longer causes errors. Plus faster home pages, separate e2e databases with a single pre-push check, and Dash0 auto-instrumentation. |
| September / October 2026 | Red’s light model, measured and honest — lamp light is now detected from the two PAR sensors instead of inferred from model error (winter lamp light had been understated by 53%), a new lamp calendar shows it hour by hour, and scored on unseen days the weather-to-plant-light chain went from R² 0.655 to 0.926 and now beats “same as yesterday”. Climate per height — a weather → greenhouse → per-wire forecast of temperature, humidity, CO₂ and light, a wire-uniformity comparison, and a crop-cycles waterfall linking each fruit sample to the climate it grew in. Deploys without downtime — exports, models and uploads moved to object storage, so both dashboards roll with no disks attached; models persist with a nightly refit and a config fingerprint; and an upload clean-up data-loss bug fixed. Plus Blue harvest per plant and per pick date, and 2026’s pooled Mix_D/Mix_E samples. |
| July / August 2026 | Production moved to the new cluster — both twins evacuated off the old Kubernetes cluster on 15 July, databases dumped, restored and verified (Blue’s 5.34 M readings plus the irreplaceable yearly manual data), production URLs flipped, old cluster kept as a one-step rollback — and six storage-related faults found and fixed along the way. Source-aware sync health — the Sync Status page now separates our pipeline is failing from the upstream source has gone quiet, with a 7-day reliability score, failing-since and a per-run sparkline. End-to-end tracing — every request and scheduled job emits OpenTelemetry traces, with logs stamped by trace ID. Plus second & third harvest thresholds on Blue’s GDD tracker, and fixes for a login-provider outage that crash-looped the dashboards and 23 days of silently failing nightly exports — and a community fix making the twins render correctly on non-UTF-8 Windows systems. |
| June / July 2026 | Soil-condition forecasting for Blue — a 7-day look-ahead on soil moisture & temperature per treatment, retrained on every deploy. Manual data in context — fertigation events overlaid on soil charts, a manual-measurement monitor, and mixed sensor/manual views. GDD on OpenMeteo modeled weather — a full calendar year (from Jan 2024) now analysable, with growth constants set to recognised values. Blue fully operational on the SPoHF Datalake — the direct Yookr API fallback retired, removing ~686 lines of code and 4.1 M duplicate rows (43.5% of the readings table). Plus per-axis chart types (line/scatter/boxplot) and a templates/asset refactor. |
| May / June 2026 | Manual data sources for Blue — insect-trap CSV and yearly long_data measurements on a new shared, twin-agnostic upload capability. Multi-height sensor views for Red — a Multi Height section mapping PAR & DLI across sensor heights. Plus server-side chart aggregation, min/max range bands, and coverage-by-source status pages. |