How many scans each node has appeared in (its seen_count). A tall left bar = lots of one-shot fly-bys; weight on the right = a durable core seen scan after scan.
Like node stability, but keyed by public key instead of ip:port — from peers seen connecting via observe. Deduped by cryptographic identity, so a peer that roams across IPs (NAT / mobile) counts once. Weight on the right = recurring, persistent identities.
New nodes discovered per hour (up) vs nodes last seen / departed per hour (down). The first hour (cold-start backlog) is omitted so the ongoing churn stays readable.
Approx distinct nodes active in each hour (their first–last sighting interval overlaps the hour). The timeline ends at the last observation, not the current clock.
Of all nodes, the % whose observed lifespan (last − first seen) is at least X hours. Steeper = more transient population.
Measured at the end of each scan run. Fills in as you run more scans over time.
From storeprobe: canary records put into the DHT, then re-polled. % retrievable and % of replicas surviving past the ~20-min record TTL measure how reliably the network stores data.
Average # of closest nodes still serving a record vs minutes since it was put. The drop-off near the dashed line is hyperdht's ~20-min record expiry (records vanish unless republished).
Average concurrent presence by hour-of-day (local). Datacenter nodes stay flat; home/dynamic nodes show a day-night cycle. Fills out as the observed span lengthens.