← RFC

Flagging Elevated Nodes by Altitude (role-independent high-site detection)

Status: Real-data tooling + observation. Date: 2026-06-09. Reproduce: python3 tools/snapshot_nodedb.py <host> then python3 tools/elevated_nodes.py.

UPDATE 2026-06-10 (fresh snapshot, noon-eval-20260610.md): now 9 elevated nodes of 113 with altitude — b76c (+80 m, role unset) joins the table below; 5 of 9 are role-missed.

A node's configured role (CLIENT / ROUTER / REPEATER) does not tell you whether it sits on a high site — and high sites are exactly the elevated relays that worsen the hidden-node problem (stone-mountain-hidden-node-hub.md). Meshtastic nodes do report altitude, which lets us flag elevated nodes independently of role.

Where the data is

Altitude is not in the static Meshview export (lat/long only), but it is in the live node DB via the Meshtastic API: of 200 nodes the KD4HME gateway knows, 109 report altitude (snapshot_nodedb.py dumps the DB with it).

Scoring — altitude relative to local neighbors

Meshtastic altitude is ASL, not height-above-terrain, so a raw number is misleading (metro Atlanta's base is ~250–320 m ASL). We instead score each node by altitude − median altitude of its neighbors within 8 km (elevated_nodes.py); a node well above its neighbors is a real high site regardless of terrain.

Result — 8 elevated nodes, and role would have missed half of them

Node	Role	Alt (m ASL)	Above local (m)
LMC3	ROUTER	624	+351
EDP1	CLIENT_MUTE	612	+330
Rky	CLIENT_MUTE	497	+244
RFM1	ROUTER_LATE	487	+234
BSRG	ROUTER	520	+195
FtBk	(unset)	527	+170
0b3b (Nano Newt / Stone Mtn)	ROUTER	420	+128
FIVE	(unset)	400	+96

4 of the 8 are role-missed — EDP1 and Rky are labeled CLIENT_MUTE, FtBk/FIVE have no role, yet all sit 96–330 m above their neighbors. EDP1 (+330 m, a "client") is one of the highest sites in the whole network. This confirms the hypothesis: altitude catches elevated hubs that the role field misses. (Nano Newt on Stone Mountain shows +128 m — clearly elevated, as expected.)

Why it matters

• Sharper hidden-node analysis. The elevated nodes are the collision-domain concentrators. We can now identify all of them (not just known ones like Nano Newt) and ask which nodes each one hears — the real map of hidden-node hubs.
• More realistic sim scenarios. Today the sim places INFRA/elevated nodes synthetically; real elevated-node positions (and the fact that some are mislabeled clients) ground the topology and the role mix.
• A product point for Loomwave: role-based assumptions about who relays/coordinates are unreliable; elevation (and measured link budget) are better signals — which is consistent with making the INFRA/coordinator role about capability and vantage, not just a config label.

Caveats

• Altitude is self-reported (GPS), so individual values carry some error; the relative score and the clear +100–350 m outliers are robust to that.
• The local-median needs ≥3 neighbors within 8 km; sparse-area nodes fall back to their 5 nearest.
• This is one gateway's node DB (KD4HME's view); snapshotting additional gateways' DBs would extend coverage. snapshot_nodedb.py is re-runnable per host.