Assessing the data quality of RIPE Atlas 👹
This repo contains scripts/outputs for some data quality metrics we propose for RIPE Atlas. We intend to run these periodically so we can see if/how data quality improves over time.
We track the following types of anchor measurements (and anchors) that require additional investigation:
- Failing measurements to connected anchors: Indicates a potential misconfiguration of the measurement (e.g., targeting an old IP) or the anchor (e.g., no IPv6).
- Failing measurements to nonconnected anchors: This is expected, but since these are ongoing measurements they should probably be stopped.
- Succeeding measurements to nonconnected anchors: This can indicate probes that hallucinate results or anchors that do not communicate with the Atlas backend even though they are online.
- Measurements to unknown anchors: These are anchors without an entry in the /anchors endpoint, which should never happen.
- Disconnected anchors: A simple list of disconnected anchors sorted by disconnection time. Anchors that are disconnected for a long time should be decommissioned.
There are probes that always claim to reach the measurement target. We can identify these using measurement 40072013, which targets a purposefully unannounced prefix. These probes should be fixed or decommissioned, since they produce fake results.