Crafty — Blackbox Review of an 18-Arm Session

Eighteen arms on one continuous power cycle; thirteen left the ground and five were bench arms. The tune is healthy. Two hardware findings are not.

Every log carries DeviceUID 002700443235511330303938, the same flight controller the backups record as Crafty. The board was logging as AIR75 F at the time, so those names refer to one drone. It is filed under Crafty here.

Session

Parameter	Value
Logging	1 kHz effective, 1/4 decimation, bidirectional DShot
Arms / flights	18 / 13
Hover throttle	~35 % stick, `throttle_limit_percent = 75`
Mean hover RPM	~20 000

The only configuration change is throttle_limit_percent 100 → 75 between btfl_004 and btfl_005. PIDs, rates and filters are identical across all 18 logs, so every trend is physical.

The centre of gravity sits forward

The two front motors hold 12.8 % more RPM than the rear pair in every level flight, while left against right stays within 1.4 %. The pitch I-term parks at −27 rather than averaging to zero. Two independent measurements, one from the mixer and one from the controller, agree.

Log	Front RPM	Rear RPM	Split	I-term pitch
`btfl_006`	21 453	18 837	+13.0 %	−25.5
`btfl_009`	21 439	18 708	+13.6 %	−27.6
`btfl_012`	21 595	18 397	+16.0 %	−34.3
`btfl_015`	21 205	18 915	+11.4 %	−23.9
`btfl_018`	20 671	18 353	+11.9 %	−24.2

Thrust goes with RPM squared, so 12.8 % is about 27 % more thrust on the front pair, or a centre of gravity roughly 3 mm ahead of centre. Moving the pack back a few millimetres and refitting a hover should settle the pitch I-term nearer zero.

The voltage readings were the decoder, not the sensor

blackbox_decode treats the vbatLatest field as a raw sensor count and divides by vbat_scale, but Betaflight 2026.6 already logs it in hundredths of a volt. Every printed voltage is low by about 1.13.

Source	Value
Raw logged field (`--unit-vbat raw`)	`419`
Header `vbatref`	`421`
True voltage	4.19 V
OSD	4.2 V
CSV as written by the decoder	3.714 V

blackbox_decode --unit-vbat raw --stdout btfl_009.bbl | head -3

vbat_scale = 113 is correct. An earlier reading of this data concluded it under-read by 15 %. That was wrong.

Corrected, one pack runs from 4.11 V to 3.63 V over ten minutes, with in-flight minima to 3.09 V under punches. Ordinary 1S behaviour.

The current sensor pins to full scale

btfl_003, 005, 013 and 016 each contain a reading of 1320.2 A, which also corrupts the mAh counter (one log claims 1360 mAh). Amperage is unusable in those four logs. The rest show a sane 4–5 A hover and about 15 mΩ of pack-plus-wiring resistance.

Healthy: filters and tracking

Noise floor: the only significant peak is the motor fundamental at 315–325 Hz, exactly 20 000 RPM across 12 poles. The RPM filter and dynamic notch remove 21× of it.
D-term: RMS near 5 in calm flight. High figures are confined to the last second of a log.
Tracking: setpoint-to-gyro error of 3–5 °/s on calm flights.

Impacts

Eight of thirteen flights peak above 4 g, six ending within 1.5 s of that peak. Noise floor and motor balance are unchanged after every impact, so no prop cracked and no motor bell shifted.

Loose ends

motor_kv = 1960 cannot produce 20 000 RPM on one cell. Harmless unless an RPM-limit feature is used.
throttle_limit_percent was 75 here against the 80 % preference.