The anchoring mesh

Every anchor measures every other anchor. The result is a dense, near-all-pairs matrix over well-connected, well-located vantage points — the closest thing the platform has to a controlled experiment, and the basis for most comparative latency work.

The mesh is refreshed daily. Mesh views are staff-gated.

What "the mesh" actually is

For each pair of active anchors, in each address family, for each measurement type, there should be a measurement running with the source anchor participating in the target anchor's measurement. A single cell is therefore (target, source, address family, type) — a pair is not one thing but several, and they can fail independently.

Atlas holds the matrix as one record per (mesh measurement → probe) pair, carrying the raw result envelope: ping RTT and loss, traceroute path, HTTP status, source address, firmware, timestamp.

Mesh health

Mesh health asserts one invariant — every eligible source anchor participates in every running mesh measurement of every other active anchor — and reports the cells where it doesn't. Each gap is reported from both directions, because the two answer different questions:

• Inbound coverage gap, on anchor B: some other anchor is missing from at least one of B's measurements. This is the view you want when asking why is B's coverage incomplete?
• Outbound coverage gap, on anchor A: A is missing from at least one other anchor's measurement. This is the view you want when asking why isn't A contributing?

The same underlying gap appears in both, described from each end. Gaps are reported per cell, so "B is fine on ping v4 but missing on traceroute v6" is expressible and visible rather than collapsed into a single healthy/unhealthy flag.

Mesh matrix

The matrix reads the same data for content rather than coverage: latency between any two anchors. Use it for questions like "how far is this anchor from everything in that region", or for spotting a pair whose RTT changed.

A missing cell in the matrix means no usable result, which is a coverage question — check mesh health to find out why.

Mesh topology

Topology renders the mesh as a graph. Be careful reading geography into it: the layout is driven by measured relationships, not by where anchors physically sit. Two anchors adjacent in the graph are close in network terms, which is sometimes but not always the same as close on a map.

Atlas also uses mesh measurements as one input when deciding whether an anchor really is in the country it claims — latency puts an upper bound on distance, which is enough to contradict an implausible claim even though it can never confirm a precise location. See Geofeeds and geolocation.

Mesh quality

Separately from coverage, a daily derivation mines the result content over time — per-link statistics rather than per-cell presence. That is what feeds the comparative and historical parts of the matrix and detail views. It is pure analysis over data already collected, so it adds no load on RIPE.

See the Anchors guide.