RELEASE HISTORY
Platform changelog
Material platform changes, reliability incidents and corrections, including claims we have retracted. Published so that what changed, and when, is traceable.
Release history
Platform changelog
Material changes to the platform, including reliability incidents and data-quality fixes. Published so that behavior changes are traceable rather than silent.
| Date | Change |
|---|---|
| 2026-07 | Performance. Media on public pages reduced by ~94% (3.6 MB → 225 KB) by moving imagery to WebP and re-encoding video; explicit image dimensions added to eliminate layout shift. |
| 2026-07 | Transparency. Published live status, security policy, data provenance and limitations pages. |
| 2026-07 | Reliability. Resolved a long-running memory-growth issue in the analysis service: per-zone baselines are now bounded by an LRU cache with on-disk persistence. Resident memory fell from ~2.35 GB to ~1.4 GB and the periodic restart cycle stopped. |
| 2026-07 | Visualisation. Added a 3D airspace instrument view (radar dome, altitude histogram, situational globe) rendered client-side. |
| 2026-07 | Interoperability. Public MCP endpoint documented so AI assistants can query live airspace and platform documentation directly. See AI access. |
| 2026-06 | Open data. Released the ADS-B Anomaly Benchmark (CC-BY 4.0, 38 documented columns) with a reproducible baseline. See research. |
| 2026-06 | Detection. Anomaly stack rebuilt on established libraries (River, scikit-learn, pykalman, OpenAP, stumpy, PyOD) with a consensus rule requiring at least two independent detectors to agree. An earlier reinforcement-learning prototype was evaluated and retired; claims referencing it were removed from the site. |
| 2026-06 | Reliability. Analysis-service restarts made invisible to users via a last-known-good cache; database write-ahead log growth bounded. |
| 2026-05 | Security. Session-token handling hardened and default-credential paths removed. |
Policy
How we version and disclose
- Breaking API changes are announced here before they take effect wherever practical. See the API documentation.
- Retracted claims. When something we previously stated turns out to be wrong or unsupported, we remove it and record the removal, as with the retired reinforcement-learning prototype above.
- Incidents that affected data quality or availability are listed rather than quietly patched.
FAQ
Frequently asked questions
How often is AeroScope updated?
Actively. The changelog records material platform changes, reliability incidents and data-quality fixes, with several substantive releases during 2026 covering detection, performance, transparency and open data.
Does AeroScope correct or retract claims?
Yes. When a previous claim proves wrong or unsupported it is removed and the removal is recorded, for example the reinforcement-learning prototype that was evaluated, retired, and stripped from the site's descriptions.