B8 - Collaborative route gossip

[M0LTE]

Idea

Today every node discovers the mesh independently. B6.1 Phase 2's peers query is a small step toward sharing - when I successfully probe you I ask "who do you know?" and seed your neighbours as via:<you> candidates I then probe myself, so node existence propagates one-hop-at-a-time across the probe graph. What's missing is route-state sharing: "I successfully reached X via Y at hop-count N, save yourself the discovery cost." That's the AODV → DSDV/RIP step deliberately deferred so far.

The trade-off is airtime. Passive-flood (B5 default) is "learn from what you see anyway, transmit nothing extra"; real route advertisement means periodic gossip that eats into the B7 discovery budget even when nothing changed. Cheapest middle ground: piggyback a small reachability TLV on the existing peers response (or beacon), so a single conversation that already had to happen carries route info as a free side-effect.

Axes worth thinking about

• When to gossip. Tie to B7 strategies - a new Overnight route-gossip window lets nodes do the heavier "here's my full reachability table" exchange during local quiet hours. Default could be: piggyback every conversation cheaply year-round, plus one fuller exchange overnight.
• Bearer choice. Topology info is broadcast-shaped:
  • HF NVIS broadcast - fits B6.2 solicit infrastructure; one node transmits its known-routes table on the beacon channel and every receiver in the NVIS footprint absorbs it. Asymmetric reach is the catch.
  • QO-100 satellite - geostationary, continental footprint, near-zero variance. Opt-in for QO-100-equipped operators; bridges otherwise-isolated terrestrial clusters.
• Fountain codes (see also F2.5 - Fountain codes for broadcast / HF / multi-path #138). DAPPSv1 is acked point-to-point so fountain codes were rejected for messages, but broadcast topology gossip is exactly the unacked-broadcast shape they were built for. Same library covers both.
• Loop / staleness defence. Distance-vector classics: split horizon, poisoned reverse, hop-count cap, periodic re-advertisement, sequence numbers per advertisement.
• Trust. Amateur regs require call-sign ID so the originator is known; whether to trust the reachability claim is separate. Cheapest: "treat advertised routes as candidates, re-probe before believing." Same shape as B6.1 Phase 2's via:<callsign> candidate handling.
• Negative info. "I lost the link to X" is just as load-bearing as "I gained it." Cheapest carrier: per-destination sequence numbers + periodic re-ad - if a destination disappears from the latest table, receivers age it out after N missed cycles.

Wire shape (sketch only)

Extend peers with optional route <callsign> hops=<n> via=<via-callsign> seq=<n> lines, terminated as today by end. Forward-compat - older nodes ignore unknown line types.

Per-destination row in a new routedigest table with (Destination, ViaCallsign, Hops, AdvertisedBy, AdvertisedAt, SeqNo) so the resolver can prefer learned-from-gossip routes when no direct probe row exists, treating them as second-class evidence (lower confidence than a probe-confirmed reach).

Status

Sketch. Worth doing once B8's airtime cost is cheaper than the discovery cost it replaces - i.e., once we have enough nodes that the per-pair probe matrix is meaningfully wasteful. With the current operator population (author + a handful) it's premature.

Migrated from plan.md Phase B8 section (sketch) and the matching scratchpad bullet.