RAK3401 recommended-preamble repeat, 23 October 2026

Status: RF measurement complete, delivery FAIL. Debugger capture integrity also FAILS its original sampling-gap limit. The firmware's settled recommendation of radio1 preamble32 / radio2 preamble200 did not eliminate the scanner's loss. Both fixed witnesses received all applicable packets. The sanitized receipt records the RF evidence and retains the debugger limitation.

Comparison with preamble384

Radio2 preamble	RAK primary / 100	RAK secondary / 100	Fixed primary / 100	Fixed secondary / 100
384, previous completed run	97	100	100	100
200, settled recommendation	96	98	100	100

Primary remained 917.000 MHz, SF7, BW62.5 kHz, CR5, preamble32; secondary remained 918.000 MHz, SF5, BW500 kHz, CR5, RX mode. Only the explicit secondary preamble changed, including the matching sender and secondary witness settings. RAK3401 still ran Full unified Repeater source adb11e7b; Tag sender and T1000-E primary witness ran Companion 08650f64, and the T096 secondary witness ran Companion 4773ac42. No firmware was built or flashed.

Before any measured transmission, the controller waited for the two-minute recommendation observation and checked the live response:

chirps=4.60,328.05; need=32,200; switch=1176us; loop=0us; test=4/4,4/4 ready

Live effective preambles were 32,200, with visits 9421,20995 us. The guarded switch allowance and visits matched the previous completed run. The controller would abort without transmitting if the settled recommendation differed from the selected values.

The Tag sent 200 fresh, unique, 92-byte RAW_CUSTOM payloads in the same seeded twenty-batch plan: 100 per profile, at 2 dBm. Physical packets were 94 bytes. The nominal post-TX_DONE gaps remained 0.5-0.9 seconds, extended as needed for confirmed sender retuning. The RF window lasted 203.294 seconds. There were exactly 200 physical direct transmissions, zero floods and no retries or replacement packets. This was successive traffic from one sender, not simultaneous transmissions on both profiles.

Packet and health evidence

RAK missed primary P032, P078, P079 and P087 (global zero-based ordinals 61, 147, 148 and 176), and secondary S014 and S029 (ordinals 33 and 48). Every missing packet reached its fixed witness. Both witnesses independently matched all 100 custom-payload pushes and all 100 corresponding raw-frame pushes. Each pair of observations counted as one delivery.

The RAK's physical receive counter advanced by 194, agreeing with its exact raw packet ledger. There were no log drops, collector errors, framing losses, duplicates or opposite-profile witness deliveries. All three receivers transmitted zero packets during the RF window, and all four radios added zero receive/CRC errors. Historical core masks remained 8/8/0/8 for RAK, Tag, T1000-E and T096 respectively. RAK soft/hard watchdog counters stayed 0/0; queues were empty and uptimes remained continuous.

These finite counts show that preamble200 was not an improvement in this run. One batch per preamble, collected at different times with uncontrolled ambient RF, does not establish a universal performance ranking or identify the cause of the six misses.

Debugger limitation and restoration

The same nonhalting read-only SWD plan completed 300.004 seconds and 27,086 samples, with exact before/after hardware identity matching. Samples showed no debug enable, halt, stepping, interrupt masking, reset or lockup. OpenOCD returned normally. However, one 27.5 ms midpoint gap between samples 11103 and 11104 exceeded the frozen 25 ms acceptance limit. Timestamps remained monotonic and all sample reads took at most 3 ms. The integrity failure is retained, not waived. The controller therefore did not award an overall pass.

The sampling gap occurred around 121.7 seconds into RF collection; the six misses occurred around 35.33, 46.84, 61.38, 150.90, 151.96 and 179.22 seconds. It did not overlap a missed packet. RF receipt validity comes from independent USB collectors, witness raw frames and exact physical counters; it does not rely on uninterrupted debugger sampling. This capture does not qualify CPU sleep: the active USB data/logging session keeps the repeater awake.

All four nodes passed settings preservation. Temporary leases were removed, RAK live radio2 returned to off, original power/logging settings were retained, and the unchanged gateway service reacquired its radio. Independent offline analysis confirmed the packet losses and the separate debugger failure.