← RFC

Cross-Referencing the High Sites into a Candidate INFRA Backbone

Status: Real-data analysis. Date: 2026-06-09. Reproduce: python3 tools/backbone_topology.py (after snapshot_nodedb.py + elevated_nodes.py).

UPDATE 2026-06-10 (noon-eval-20260610.md): now 9 sites, 18/36 candidate links, still one connected component; BSRG is the most-connected site (6 links). And the first links are observed, not just geometric: 0b3b↔BSRG (41 km, 26 direct rx, median SNR +5.9 dB), 0b3b↔FIVE (23 km, −3.8 dB), 0b3b↔b76c (32 km, −10 dB) — the keystone bridge to Stone Mountain is real with ~13 dB margin. (Rky↔RFM1 re-resolved to 1.9 km apart — two towers, still a redundant pair.)

The recommended architecture (design-decision.md) puts an INFRA backbone at its center: the elevated coordinator nodes run an aggregated link-state network among themselves and act as the time/slot authority for the clients beneath them. elevated-nodes.md found 8 real high sites by altitude. This doc asks the next question — could those 8 sites actually form a backbone? — by testing which of them can plausibly reach each other.

The Loomwave premise being tested

Elevated-to-elevated links reach far beyond ground-to-ground range. The radio horizon grows with the square root of each antenna's height advantage, so two high sites see each other across distances no ordinary node pair could bridge. We saw this directly: Stone Mountain heard a 123 km link (stone-mountain-hidden-node-hub.md). If the premise holds, the high sites should knit into a connected backbone even though they're scattered across ~180 km.

Method

For each elevated pair we compare great-circle distance against the 4/3-earth radio horizon

d_los (km) = 4.12 · (√h₁ + √h₂)        h = height advantage above local terrain (m)

using above_local (altitude minus the local-neighbor median, from elevated_nodes.py) as the height advantage — the right quantity for a horizon calculation (height above surrounding terrain, not raw ASL). A pair is a candidate backbone link if it clears the horizon geometrically and falls within a long-range LoRa budget cap (LORA_KM = 120 km, anchored to the observed Stone Mountain link). Geometry permitting line-of-sight is necessary but not sufficient — the link budget must close too.

Result — the 8 high sites form a single connected backbone

Candidate INFRA backbone over the high sites

15 of 28 pairs are candidate links, and they form one connected component spanning all 8 sites — no islands. The longest single backbone hop is 116 km; the whole region is reachable in 2–3 elevated hops.

Property	Value
High sites	8
Candidate elevated-to-elevated links	15 of 28 pairs
Connectivity	1 component — all 8 sites connected
Longest single hop	116 km
Most-connected sites (degree)	Rky : 5, RFM1 : 5, BSRG : 5
Least-connected (leaf)	FIVE : 2

Structure that matters for the design: - BSRG (+195 m) is the keystone bridge. Geographically central, it is the only well-connected link between the northern cluster (LMC3 / EDP1 / RFM1·Rky) and the southern cluster (FtBk / 0b3b / FIVE). Lose BSRG and the backbone risks splitting — it is the node whose redundancy matters most. - Rky and RFM1 are co-located (0.0 km apart) — effectively the same tower carrying two radios. That is a built-in redundant relay at a central position (both score degree 5), and a useful real example of the multi-radio INFRA node the architecture anticipates. - 0b3b (Nano Newt, Stone Mountain) is a leaf in the southeast, reaching the backbone via BSRG (41 km) and FtBk (84 km). The hidden-node hub we characterized is itself a backbone edge node — the clients it concentrates would be coordinated through its uplinks to BSRG/FtBk. - The two highest sites, LMC3 (+351 m) and EDP1 (+330 m), anchor the north and reach the core through RFM1·Rky. (EDP1↔BSRG clears the horizon geometrically at 131 km but exceeds the 120 km budget cap — shown dotted — so it is not counted as a usable link; the backbone does not depend on it.)

Why this informs the backbone design

The backbone is physically realizable from existing high ground. With no new towers, 8 sites already on the air cover the region in 2–3 elevated hops. The INFRA link-state network the design assumes isn't hypothetical — the topology is sitting in the node DB.
It tells us where redundancy is structural, not optional. BSRG is a single point of failure for north–south connectivity; the INFRA routing layer should treat such cut vertices specially (and the Rky/RFM1 co-located pair shows the natural fix — dual radios at keystone sites).
It grounds the sim. Instead of scattering INFRA nodes synthetically, the neighborhood scenario can use these 8 real positions, their real height advantages, and this real link graph — including the realistic fact that some "INFRA" sites are mislabeled clients (elevated-nodes.md).
It reinforces the role argument. The backbone is defined by vantage and reach, not by config role: half these sites are role-missed (CLIENT_MUTE / unset). Whoever coordinates the backbone must be chosen by measured elevation and link budget, not a self-declared role field.

Honest caveats

Geometry + budget proxy, not measured links. A candidate link means LOS geometry closes and the separation is within a long-range LoRa budget — it does not prove the link decodes. Terrain obstructions between two peaks, foliage, and real antenna gain/height aren't modeled. The Stone Mountain capture gives one measured anchor (123 km) that makes the 120 km cap credible, but each backbone edge should ultimately be confirmed by a real reception, not just geometry.
above_local is a terrain proxy. Height above the local-neighbor median approximates height above terrain; a site whose neighbors are themselves elevated will be under-credited (and vice versa). The horizon estimate inherits that error.
Self-reported ASL. Altitudes are GPS self-reports; the clear +96–351 m outliers are robust, but individual horizon distances carry the underlying GPS error.
One gateway's view. These 8 sites come from KD4HME's node DB. Snapshotting additional gateways (snapshot_nodedb.py <host>) would likely surface more high sites and densify the backbone graph.

Bottom line: the region's existing high sites already cross-reference into a single connected INFRA backbone — 8 elevated nodes, 15 candidate elevated-to-elevated links, 2–3 hops end to end, with BSRG as the keystone bridge and a naturally redundant dual-radio site at Rky/RFM1. The backbone the architecture assumes is buildable from hardware already on the air.