ScotMesh Repeater Coverage

MeshCore repeater map, refreshed daily

– active – degraded – silent Last updated: –

LoRa flood simulator

Send simulated flood packets from chosen nodes and see where floods collide, using MeshCore's real airtime/relay-delay formulas (compiled from the same Go code as the server).

Showing

Random seed

Sim duration (ms)

Predict-settings trials

🛰️ Validate against real traffic

Replay a real CoreScope packet

Reconstruct what actually happened on the real network: every relay CoreScope observed proof of, alongside what our own model predicts should have happened from the same origin — so a gap between the two points at where a flood likely collided or got blocked.

Surrounding activity window (± seconds)

Searches composite multi-rule policies (dense/sparse, hilltop, articulation-point, MPR-style and more) ranked by actual delivery, not just collision rate.

Max rounds (0 = unlimited)

Give up after N stale rounds (0 = never)

The optimizer stops at whichever of the two limits above it reaches first.

Allow changing flood.max (off by default — riskier)

Experimental (try one at a time)

Each is independent and off by default — land and measure one at a time rather than stacking several, since each changes the search trajectory in a way that makes it hard to tell which change actually helped.

Adaptive trial budget (racing) Late acceptance hill climbing (escape local optima) SPSA warm start Learned contention-score weights

Pushes synthetic offered load (randomly-originated messages, not your own senders above) up through the load levels below and measures delivery at each — the "knee" is the highest level this network still delivers acceptably at, i.e. how many messages/minute it can actually handle.

Load levels (msgs/min)

Edit a node's delay/power/loop-detect/scope/radio settings, then Apply. Changes take effect on the next run.

Set all repeaters to:

Node	Type	Scopes	Allow unscoped	Flood max	Unscoped max	Radio	Tx delay	Direct tx	Rx delay	Tx power	Loop detect	Hash size	Duty cycle	Received

Reconstructed from real CoreScope observations. Second-resolution timestamps mean this reproduces the traffic load with plausible sub-second timing, not the exact instant every packet collided — a settings win is worth confirming across a few seeds (change the seed above and re-run).

This packet: did our simulation deliver where reality did?

Observer	Heard it (real)	Our sim delivered	Verdict

Where did the flood die?

Delivery probability (timing-jittered runs)

CoreScope stamps whole seconds, so within-second ordering — and therefore collision outcomes — is partly arbitrary. This replays the same episode 10× with ±1 s timing jitter and fresh seeds, reporting how often the target actually got through.

Problems — before vs after your changes

Run once to see the baseline, then change settings (or run the optimizer) and run again. "Set current as baseline" pins a reference so the delta always compares against the state you chose.

Problem	Baseline	Now	Change

Every repeater, ranked by contention

All repeaters, worst first. "Score" is what the optimizer ranks on (collisions caused + own collisions + wasted relays + duty cycle). Click a row for the full diagnosis and what to do about it.

Repeater	Score	txdelay	Caused	Collided	Wasted relays	Relayed	Diagnosis

Improvement over time

One row per round. ✓ marks a round whose change was kept; the rest were measured, rejected, and reverted.

Round	Kept	Targeted	Move	Candidates tried	Delivery	Collisions	Contention

What changed, and why

Each row is one repeater the optimizer specifically backed off, in the order it found them — not a class-based rule, a targeted exception on top of the policy search result above.