Peer discovery became testable before global first-contact discovery was solved.
The GNS baseline audit on August 5, 2026 found an important documentation gap. Discovery and NAT/relay code already existed, along with supporting tests, but the M9 and M10 milestones lacked dedicated acceptance records.
The work at this point included making those boundaries explicit. The date marks the acceptance documentation, not necessarily the first day every underlying feature was implemented.
Keeping peer knowledge bounded
The discovery foundation separated transport NodeID from producer identity. It also introduced framed peer announcements with time-to-live limits, malformed-input rejection and anti-spam checks.
Candidate selection respected configured peer limits. Persisted peer metadata could survive a snapshot restore, allowing a restarting node to retain useful information about previously observed peers.
This was knowledge management with constraints. A peer announcement was not permission to grow the peerbook indefinitely, and a discovered address did not remove the need for connection checks.
The M9 acceptance tests passed for malformed announcements, candidate limits, identity fields and metadata persistence across restart.
Adding a relay foundation
M10 covered NAT traversal hooks and status, external reachable-address hints, relay envelopes, policy checks and usage accounting.
The tests exercised relay delivery and a relay-based hello fallback. They also checked rejection paths: disabled relay operation, missing required stake, exceeded quota and exhausted relay time-to-live.
Those negative cases mattered. A relay interface needed limits and authorization behavior as well as a successful forwarding scenario.
The acceptance record reported PASS for that foundation scope.
What the foundation did not settle
At this stage, NAT traversal still included placeholder behavior. The tests deliberately preserved that classification instead of treating an available status field as proof of production hole punching.
The framed discovery and relay paths were exercised through the repository's HTTP carrier. That was useful implementation evidence, but it did not establish the intended final internal TCP+TLS transport.
Most importantly, these mechanisms still needed candidate knowledge. A completely fresh node with no configured peers, useful cache or other discovery input faced an earlier question: where should its first attempt go?
Handling discovered peers and discovering the first public peer are separate milestones.
Why this stage mattered
The acceptance work made it possible to describe progress precisely. Peer metadata, bounded discovery and controlled relay behavior had concrete tests. Global bootstrap and production NAT remained open.
That distinction shaped the later networking experiments. Instead of assuming that a collection of working components implied a usable public network, the project could ask which missing input or path prevented a fresh node from joining.
The next stages would make those missing pieces visible through dedicated bootstrap and reachability evidence.
Built in Public. Documented as we build.
Top comments (0)