Where this actually works
Nobody publishes where direct WebRTC connects — not per country, not per carrier, not per browser pair. This is the attempt: every row is one measured attempt, reviewed by the person who made it before it was published.
This page connects to a relay to read the public database — that is the whole of what it does, and it writes nothing. The demo itself still connects to nothing until you press publish.
Starting…
What this cannot tell you yet
A dataset that overstates its coverage is worth less than one that says plainly what is missing. Each line below is a question nobody can answer from these rows — and an invitation, because the way to close one is to run it.
What is here
By network
One line per country and carrier. The two IPv6 columns are a pair on purpose: without the number of runs that offered IPv6 at all, "no reflexive IPv6" cannot be told apart from "IPv6 was never tried here".
Every row
One attempt per row, because aggregating can be done afterwards and disaggregating cannot. The filters below offer only values that are actually in the data.
What was not mirrored
Anyone may write to this log, so a reader filters rather than a person gatekeeping. A filter whose removals cannot be inspected is an opinion with a URL, so here they are.
How to read this
Every row was measured locally, reviewed by hand and then published deliberately.
The database:
—