Studies of how MeshCore actually behaves, run on real firmware against a simulated radio and a simulated channel.
Published at https://meshbench.github.io/meshbench-reports/.
MeshBench compiles the real MeshCore application natively, one operating-system process per node, and runs it against a model of an SX1262 and a model of the air. The firmware and its radio driver are unmodified, so the decisions in these reports are the same decisions the code makes on hardware. What is simulated is everything the radio would have told it.
- Testing 1.17's listen-before-talk on a radio that lies — MeshCore 1.17 rewrote listen-before-talk to survive LoRa interrupt flags that get stuck. On a well-behaved radio the change is invisible, so we built a radio that sticks. The two versions then separated by a factor of 43.
Every figure is measured output from MeshBench. The interpretation and the prose are Claude Opus 5's.
Results are a best case throughout: no multipath, bare-earth terrain, no body loss, no oscillator error, and no interference beyond the mesh's own traffic. Every one of those omissions makes real links worse.
The faulty-radio builds used here come from
meshcore-native, which publishes
host builds of MeshCore alongside -faultyirq variants that misbehave the way
real radios do.