View this page on MeshCore Docs.

LoRa OTA through one relay, 29–30 October 2026

An isolated one-hop LoRa transfer completed using ESP-NOW only for laboratory control. The production menu also completed a fresh installation with LoRa administration and ESP-NOW disabled on every participant. The shared-channel capture limits below distinguish these two checks.

Fixture

Board	Role	LoRa frequency
Heltec V4.3	USB source	909.5 MHz
Heltec V4.2	Relay	Primary 909.5 MHz; radio2 908.5 MHz
Heltec V3	Target	Saved 907.5 MHz; temporary 908.5 MHz
T1000-E and Mesh Pocket	Passive RF observers	Input and output channels

Both transfer channels used 500 kHz bandwidth, SF5, CR5 and preamble 128. The relay had repeating and crossing enabled, with ESP-NOW disabled. Source and target kept radio2 and crossing off. Their existing ESP-NOW link carried native ACL administration; bridge filters excluded OTA payloads in both directions.

Completed trials

Trial	Result
Full package	Intentionally stopped at 58/1,047 accepted blocks; no installation
213,348-byte differential package	104/104 blocks complete at 834.831 seconds from controller start; no installation
461-byte CFC differential package	Version/build-date change from v1.18.0.2 to v1.18.0.3; installed a 2,142,104-byte application

The 213 KB trial stopped after completion because ota dev verify cannot verify ESP32 flash-backed staging through its memory-buffer diagnostic. This was not a remote permission failure. The recorded trial remains failed; a completed download is not proof of installation.

The small menu update installed body 63D3F15F24AE2BA0, confirmed over USB. Its first post-reboot menu check failed because the native ACL return PATH was volatile. The transport now re-establishes that PATH after a reset without a password or routine repeated bootstrap traffic. Menu recovery subsequently finished with install_verified=true and restored all owned settings.

Independent relay evidence

For the 104-block differential, observers captured 1,339 distinct identical DATA payloads on 909.5 MHz with path count zero and on 908.5 MHz with path count one. Every output path matched the relay's independently checked prefix. All 104 block indices were represented. Disabling relay crossing stalled the same transfer; re-enabling it resumed progress. The target's 907.5 → 908.5 MHz change was acknowledged over ESP-NOW and confirmed by transfer progress.

The full-package trial had 777 matching DATA payloads. Neither capture had parser or recorded path-metadata mismatches.

The capture contains legacy DATA for 103 differential blocks and a compressed v2 tail for block zero. Full-sized v2 frames are absent from this capture, so these counts do not measure all transmitted v2 packets. Package size savings must not be described as an entirely compressed OTA wire transfer.

Fresh laboratory menu run

PASS: the fresh run from the app0 baseline completed download, installation, reboot and exact runtime verification. The saved record ended at finished with install_verified=true, no recovery errors, both endpoint leases off and the original endpoint frequencies restored. USB independently confirmed the expected version and body hash.

LoRa administration and fresh menu installation

PASS: Heltec V4.3 → Heltec V4.2 → Heltec V3 used 909.5 MHz, 500 kHz, SF5, CR5 and preamble 128. ESP-NOW, radio2 and crossing were off on every participant. The USB source used an explicit three-byte relay prefix for authenticated CLI requests and the reverse return route. The menu opened and restored all three primary OTA windows, RXPS and OTA policies, and the relay's repeating setting.

The first run installed the earlier package and restored settings, but its capture contained no DATA. A second run used a newly compiled v1.18.0.4 image and new package ID 434E4672, excluding reuse of the earlier package. The 377-byte differential contained 163 payload bytes and reconstructed the full 2,142,104-byte application exactly on the host. The menu verified the new body 356D1F43B1B5E15A after reboot and finished without recovery errors. USB independently confirmed that body and version. All leases were off, the source native path was off and its host folder was detached afterward.

The fresh capture contains the new DATA block on the source's zero-hop path. Its v2 stream inflated from 156 to 163 bytes. Relay copies were captured for discovery, manifest and proof messages, but no matching DATA relay copy was captured. These co-located nodes can hear the source directly: this run proves fresh LoRa delivery and routed administration, not that its DATA required the relay. The separate two-frequency trial above supplies the isolated DATA relay evidence. Neither result is a multi-relay hardware qualification.

The measured USB source used 5b01060d8 plus the native path change committed as 925f37a44; its running body was 463F0AD2C11A7271. The target's small test updates changed version/build metadata in the same application code. They exercise transfer, full-image reconstruction, application-slot installation and reboot rather than functional changes in the receiver application.

Scope and handoff

The isolated fixture required a preconfigured dual-profile LoRa relay, ESP-NOW control link, ACLs and OTA exclusion rules. ESP-NOW was a test aid for changing frequencies and reading status; it is not a production requirement. The production menu instead uses existing ACL grants and an already working LoRa administration route. It keeps the saved channel and manages relay OTA windows. Only one physical relay was tested; selectable larger hop limits do not establish hardware coverage for longer routes.

After testing, all three infrastructure boards were flashed with normal v1.18.0.2 firmware from 925f37a44. Their running versions and application body hashes matched the built images. They were left with RF, repeating, radio2, crossing and ESP-NOW off, with no temporary leases. Both passive observers returned to their original radio settings and transmitted no packets. The evidence summary records the final firmware hashes and qualified outcomes without device identities.

Offline checks passed on Windows and Linux: 287 wizard tests and 21 relay controller tests per OS. Native routing tests also passed on Windows and under ASan/UBSan on Linux. Seven hosted-document tests and the Windows JavaScript link harness passed. These checks supplement the hardware evidence above.