contracts/docs/impl/reports/Z27AG-DECODE-2.md

type, date

type	date
report	2026-10-15

Z27AG CAN: second pass, against OBD-II and GNSS

Sessions 000007-20261015T192938 and 000008-20261015T200202 (car Orin): doors, tailgate, bonnet and wipers (checklist step 10–11), then a short drive with GNSS (3D fix, helix antenna) and OBD-II through the ELM327 on the K-line, all on the Orin's clock. The first time CAN, OBD-II and GNSS were recorded together. About 30 minutes; 1190 OBD-II samples of each fast value.

Confirmed

Signal	Where	Evidence	Registered as
Engine speed	210 bytes 4–5, BE, 1 rpm (pass 1)	equal to OBD-II 0C: median ratio 1.0000, median difference 4 rpm, 1190 samples	engine_rpm (unchanged)
Coolant temperature	449 byte 2, −40 °C	equal to OBD-II 05 within ±1 °C (median 0.0) over 86–97 °C; no other byte tracks it	coolant_temp_c
Vehicle speed	231 bytes 6–7, little-endian, 0.0577 km/h per count	steady cruising (> 15 km/h, GNSS change < 1 km/h over 2 s, 117 fixes): spread 2.6 % against GNSS	vehicle_speed_kph
Four wheel speeds	235 four words, little-endian (pass 1 left the byte order open)	big-endian misses by 28 km/h rms; little-endian fits. Words 0, 1: 0.0578; words 2, 3: 0.0581 km/h per count against GNSS	not yet: which word is which wheel waits for the jack-stands run (fronts turn, rears stay)

Scales are against GNSS, so they include this car's tyres today. OBD-II speed reads 3 % low against GNSS (0.9706): the ECU's figure is no better a reference than the ABS frames. The two word pairs of 235 differ by 0.5 %, the usual size of a front/rear tyre difference.

Doors, tailgate, bonnet, wipers

Thing	Bit	Notes
Driver's door	423.0.4	its own bit (pass 1); also sets any-door
Passenger, rear left, rear right	only 423.2.1 (any door)	no bit of their own
Tailgate	only 423.2.1	with the driver's door open, 423.0.4 cleared on closing it while 423.2.1 stayed set until the tailgate shut
Bonnet	nothing	a mechanical latch with no switch, as expected
Front wiper running	423.3.3	set through "low" and "high"; nothing in a 4 s "interval" window. Medium confidence; also explains pass 1's unexplained 423.3.3 stretch

IMU calibration, checked against GNSS

Level from session 000006 (two still stretches, the ground's 1.9° slope cancelled), forward from 000008 (three stops, one launch; 3512 samples, mean spread 12.7°). The module sits upside down, about 24° off level, turned so its X axis points forward. Calibrated, the marked stretches read:

Stretch	IMU longitudinal (0.25 s mean, peak)	GNSS
Stop, 48 → 0 km/h	−0.35 g	−0.28 g
Stop, 45 → 0	−0.17 g	−0.17 g
Stop, 40 → 0	−0.21 g	−0.18 g
Launch	+0.32 g	+0.36 g

Lateral stays at about 0 in all four: forward is straight ahead. GNSS reports about once a second, which smooths short peaks.

Session 000009: lights at night, corners

z27ag-20261015-203117-lights-corners-drive: step 9 at night, then a 35-minute errand drive (no OBD-II: the obd-poller was off after the freeze). References: GNSS and the calibrated IMU.

Signal	Where	Evidence	Registered as
Main beam	423.2.7	set exactly through the marked stretch; also fired once with the other lights off: flash-to-pass	main_beam
Front fog lamps	423.2.4	set exactly through the marked stretch (lights on)	front_fogs
Steering wheel angle	0C0 bytes 0–1, little-endian, signed, 0.1° per count, positive left	correlation +0.995 with how sharply the car turns (IMU yaw rate ÷ speed), 1021 s of driving. Lock to lock −479…+471°; with a 2.5 m wheelbase that is a 16.3:1 steering ratio, typical. Straight driving reads −2.7° (camber or alignment)	steering_angle_deg (the layout format gained signed)
Dipped beam?	423.3.1	clean in daylight on the 13th; at night it flickers on for 0.1–0.2 s at each switch movement. The cluster has no dipped-beam lamp (only side lights, main beam and fogs, no rear fogs), and 423 matches the cluster's lamps, so this bit is probably something else that moves with the lights (illumination, dimming). A walk through Off → Auto → Dim → Main settles it	dipped_beam kept, description says so

IMU, lateral axis: lateral g against speed × yaw rate gives a slope of −0.998 (correlation −0.91; the sign is right: yaw counts a left turn positive, lateral g counts right positive). With the stops above, both horizontal axes of the calibration are confirmed.

149 bytes 1–2 is not a yaw-rate sensor (correlation 0.02 with the IMU's yaw rate).

Not found

• Intake air temperature: no byte of 449 tracks OBD-II 0F (29–51 °C here).
• 449 bytes 0, 1, 3 move like temperatures but match nothing on OBD-II.

Notes on the recording

• The Orin froze mid-drive (19:59; hardware watchdog reset at 20:01:47), splitting the drive over two sessions. orin-nano-env-setup has the investigation.
• OBD-II through the ELM gave each fast value about every 2 s (K-line, ~250 ms per request).