- Update scripts/validate.py with advertise block and control action validation
- Update templates.yaml with mytown_parking_beacon advertising TX template
- Add commented parking beacon TX/RX example to config.yaml
- Document advertise schema, counter_mode, counter_start, and control actions in README.md / README.ko.md
Both earlier explanations for the missing gear sensors were wrong, and the
response latencies in the logs rule them out.
The 21 A0 reply is 61A0FFFF0000 in all 115 logged responses, and it is a
finished reply rather than a first frame with the rest missing. Median
round-trip is 92ms, against 79ms for the single-frame 010C and 76ms for
21 01, which returns 41 bytes across multiple frames. A stalled multi-frame
would have to sit through the adapter's wait for the frames that never come
before it prints a prompt, so it would be far slower, not the same speed.
Multi-frame on 7E0 also completes normally in the very same session, so flow
control is not implicated either. The ATFCSH7E1 setup added in the previous
two commits was speculation and is removed.
The 1.0.9 idle-wait change is not responsible either: sendCommand blocks on
the prompt and the poll loop awaits it, so a later command cannot be written
into the middle of a reply. The idle wait only covers gaps when no command is
outstanding.
So the ECU really does return four data bytes here, and awesomeit's byte 13,
20 and 21 cannot be reached from it. Whether that is because the profile
describes a different specification, or because this TCU only fills the block
under conditions not present while parked, is still open — every log so far
was captured in park with the engine idling.
Replace the three sensors with probes exposing bytes 0 through 3 directly, so
shifting through P/R/N/D shows whether any of them move. If they do, that
byte carries the gear; if FF FF 00 00 never changes, this TCU does not serve
the block and the sensors should go for good.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Reverts the mode change from f86de06. Mode 21, PID A0 and header 7E1 were all
correct — as the awesomeit profile says and as this car was observed doing.
What was missing is flow control.
Every one of the 115 logged replies was 61A0FFFF0000, exactly six bytes. That
is not a short answer, it is the payload size of an ISO-TP First Frame: two
PCI bytes plus exactly six data bytes, with the rest carried in consecutive
frames that never arrived. The ECU was sending the whole block; only the first
frame got through.
Two things corroborate it. Multi-frame from 7E0 works fine on this setup —
21 03 returns 63 bytes and 21 01 returns 41 — so the reassembly path is sound
and only 7E1 truncates, which is what a stalled flow-control negotiation looks
like when an ELM327's automatic handling does not cover a non-default address.
And the Carnival YP log sets ATFCSH7E1 explicitly before querying 7E1; its
reply 6201A0FFFF84... shares the FF FF prefix with the Accent's 61A0FFFF0000,
so both are the same block, one complete and one cut off at the first frame.
Keep the flow control pre_commands added in f86de06 and put the mode and PID
back to 21 / A0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Removing these sensors was the wrong call. The header was right and the app
parsed the reply correctly — the service mode was wrong.
21 A0 at 7E1 answers on this car, but only ever with 61A0FFFF0000: four data
bytes, FF FF 00 00, identical across all 115 responses in the logs. The
awesomeit profile reads bytes 13, 20 and 21 from this block, all past the end
of that stub, so nothing could ever be published.
The Carnival YP diagnostic log shows what that stub is a truncation of. On the
same 7E1:
21 A0 -> 7F 21 11 (serviceNotSupported)
22 01A0 -> 6201A0 FF FF 84 00 00 00 00 13 ... (42-byte multi-frame)
Both replies open with FF FF, so the Accent's mode 21 answer looks like an
abbreviated form of the same block, with the real contents behind mode 22.
The awesomeit profile's 21 A0 is likely written for an older TCU generation.
Restore the three sensors with mode 22 and DID 01A0, keeping awesomeit's byte
positions unchanged so only the mode differs. Flow control is set the way the
Carnival log sets it before querying 7E1, since the full reply is multi-frame.
If this car answers short or negatively, decode returns null and the sensors
stay empty rather than reporting nonsense, and the log will say which.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
21 A0 at 7E1 answers on this car, but with nothing in it. Every one of 115
responses in the logs was identical:
61A0FFFF0000
Four data bytes after the echo, fixed at FF FF 00 00. The awesomeit profile
reads transmission fluid temperature from byte 13, gear position from byte 20
and gear from byte 21, all far past the end of that response, so all three
sensors could never produce a value. decode returning null out of range is
why they stayed blank instead of reporting nonsense.
The profile presumably describes a different trim or transmission than this
car has. Remove the three sensors and record the measured response so this is
not retried.
Sensor count: 11 -> 8.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
hassble-android's OBD loop (fixed in #65, shipped 1.0.9) no longer wakes every
10ms regardless of interval, but the interval itself still sets how much real
BLE traffic goes out — every poll is a write + wait for a genuine response
from the adapter, unlike the old idle-loop overhead which was pure software
busy-checking. Values that don't need sub-minute freshness were still being
polled as often as live telemetry.
Keep rpm and speed at 1s across every template — that immediacy is the point
of installing an OBD gateway. Loosen everything else to match how fast it
actually changes:
fuel_level 30s -> 60s (already had a 60s publish min_interval
in two templates; polling faster than
you publish was pure waste)
cluster_voltage 60s -> 120s (diagnostic, near-constant)
gm gear / prnd 1s -> 2s (still feels live for a dashboard)
gm oil_pressure 2s -> 5s
accent intake_temp 10s -> 20s (thermal mass, slow by nature)
accent barometric_pressure
30s -> 60s (near-constant while driving)
accent trans_oil_temp 10s -> 20s
accent gear / gear_position
1s -> 2s
accent odometer + at_last_dpf_regen + distance_since_dpf_regen
60s -> 300s (no reason for minute-level odometer
freshness, and this is the heaviest
exchange in the config: a 63-byte
multi-frame response, versus a few
bytes for rpm/speed)
accent odometer_flow_control
60s -> 300s (diagnostic probe, matches the above)
coolant_temp (10s) and oil_pressure's GM preset default were already
reasonable and are unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
meatpiHQ/wican-fw#478 has two real Hyundai/Kia vehicles reading the odometer
from the same header (7C6) and PID (22 B002) our car refuses:
Ioniq 28kWh (WiCAN): pid_init "ATSH7C6;ATFCSH7C6" — works
Ioniq 5 (canair): 22B002 queried directly — works
Neither needed a diagnostic session (10 03). That weakens the earlier theory
that our 7F 22 22 (NRC 0x22, conditionsNotCorrect) just needs a session to
clear — the working examples got a positive response without one, using only
flow control setup. Replace odometer_ext_session's session attempt with
odometer_flow_control, matching what those cases actually needed
(ATFCSH/ATFCSD/ATFCSM1), which our earlier attempt never set explicitly.
The linked #514 in that same issue is instructive for a different reason: it
is a WiCAN bug where ISO-TP PCI bytes leak into the payload it hands to user
expressions, shifting every byte offset. That is the same defect class as the
multi-frame length-header bug we fixed in this app, but it is not our bug —
our own odometer cross-check (odometer_at_last_dpf_regen + distance_since_dpf_regen
== odometer, using OBDb's PCI-stripped byte positions) already proved this
app reassembles ISO-TP correctly, so there is nothing to fix here.
That issue also shows the odometer's position inside a 22 B002 response is not
fixed across Hyundai/Kia generations: byte 6-8 (g:h:i) for the Ioniq 5 and
OBDb's eleven ICE models, but byte 9-11 (j:k:l) for the older 28kWh Ioniq
(recovered from WiCAN's buggy offset via the conversion table posted in #514).
Noted in a comment: if odometer_flow_control answers but the value looks
wrong, try j*65536+k*256+l before giving up on it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
offset_value -1436.5 fails a basic sanity check: applied to a hypothetical
new car reading 0 km on the engine counter, it would report a -1436.5 km
dash reading. An additive constant is only physically justified by a
one-time reset event, and the sign is wrong for that anyway — a reset
counter would read lower than the dash, not 0.73% higher.
A proportional error fits the actual mechanism. This DPF-service counter
converts wheel-speed pulses to distance with its own calibration constant,
separate from the cluster's regulated one, so it drifts from the true
distance in proportion to distance travelled, not by a fixed amount. That
model passes through the origin: a new car reads 0 km on both.
Replace scale 0.001 + offset_value -1436.5 with scale 0.000992723 (the ratio
195958/197394.5) on all three DPF-block sensors, not just the two the
constant version touched — a proportional correction must be applied
uniformly for odometer = odometer_at_last_dpf_regen + distance_since_dpf_regen
to still hold, unlike an additive one which happened to cancel out either way.
Add odometer_ext_session as a live retest of the cluster's own 22 B002.
Its earlier 7F 22 22 carried NRC 0x22 (conditionsNotCorrect), which per
ISO 14229-1 means the identifier exists but the ECU's current state refuses
it — unlike NRC 0x11/0x31, which would mean it does not exist at all. Send
a diagnostic session control (10 03) ahead of the read and watch the log: a
positive response there would give the dash value directly, replacing every
calibration assumption above with a value that needs none.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The dash-matching cluster/transmission/standard-PID sweep (34 diagnostic
sensors probing 22 B002 at 7C6, 21 A0 at 7E1, and the 01 A6 preset) found
nothing at 195958 km. The confirmed working sensor stays the 21 03 @ 7E0
engine block reading, already calibrated with offset_value -1436.5 to match
the dash.
Remove the scan sensors, the two 22 B001/B003 existence probes, and the
01 A6 preset attempt. Keep the gear/gear position/transmission oil temp
sensors from the 21 A0 @ 7E1 block: they are unrelated to the odometer
search — the awesomeit profile for this exact car reads them from bytes 13,
20 and 21, which the 0-15 scan window covered without ever matching an
odometer-shaped value, so the sweep's failure says nothing about whether
this block answers at all. If 21 A0 turns out not to respond, those three
should go too.
Sensor count: 45 -> 10.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The previous probes read byte 0 with formula "a" and never looked at an
odometer at all, so they could not answer the question they were added for.
Sweep the offset instead. 21 01 at 7C6 and 21 A0 at 7E1 each get sixteen
sensors reading three bytes as an unsigned big-endian value in kilometres,
at offsets 0 through 15. Whichever one reports 195958 has found the odometer,
and the rest can be deleted along with it.
The sweep is safe because decode returns null when offset plus length runs
past the response, and emit drops nulls, so offsets that do not exist stay
empty rather than publishing noise. Length 3 with type uint32 reads 24 bits,
since toLong only shifts across the slice it is given and uint32 skips sign
handling. Twenty-four bits scaled by one is the shape the cluster odometer
takes in 22 B002 on this platform; the engine's 32-bit thousandths form is
already covered by the working odometer sensor.
22 B001 and 22 B003 stay as single existence probes because nothing is known
about whether they answer at all. If they do, their response lands in the log
and the same sweep can be pointed at them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every candidate had formula "a", which was right for two of them and lazy for
the rest.
21 A0 at 7E1 is not a guess. The awesomeit profile for this exact model, year
and engine reads gear, gear position and transmission fluid temperature from
that block, so it becomes three real sensors with the published byte
positions: N at byte 13 minus 40 for the fluid temperature, U at byte 20 and
V at byte 21 through their lookup tables. Gear and position need decode
because bytes 20 and 21 are past the range formula variables can address. If
the block answers, the raw response also lands in the log for the odometer
search, so nothing is lost by making these real.
01 A6 has a standard formula, so it uses the built-in odometer preset instead
of reading one byte. If the car answers it, the value is the dash reading
directly with no calibration.
21 01, 22 B001 and 22 B003 at 7C6 keep formula "a" because no byte layout for
them exists anywhere — the awesomeit database and OBDb both carry only 22 B002
on 7C6. Writing a formula would be inventing one. These stay existence probes;
the raw hex in the log is the deliverable, and the real formula gets written
once that hex says where the value sits. The comment now states this outright
rather than leaving "a" unexplained.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The working odometer comes from the engine ECU's 21 03 block and sits 1436.5 km
above the dash, corrected with offset_value. Probing more 7E0 blocks cannot
improve on that — they read the same counter and carry the same bias. A value
that matches the dash has to come from a different module.
Add read-only probes on the two independent candidates:
21 01 @ 7C6 the cluster's mode 21 block. 7C6 refuses 22 B002 but answers
with 7F 22 22 rather than staying silent, so the module is
present and other identifiers are untested
22 B001 @ 7C6 identifiers adjacent to B002, unrecorded anywhere
22 B003 @ 7C6
21 A0 @ 7E1 the transmission. The awesomeit profile for this exact model
reads gear and fluid temperature from this block, so it
exists here and may hold its own distance counter
01 A6 the standard odometer PID. The mode 01 support list for a
2013 Accent stops at 0145 so this is a long shot, but it
would return the dash value directly and costs one request
Deliberately left out: 22 B002 at 7C6, which this car refuses outright, and
01 2F, which answers NO DATA. Both are recorded in the comment so they do not
get retried.
Note in the template that OBDb's Accent captures come from North American
petrol cars and do not describe this diesel — 7E0.2103 is marked unsupported
there yet works here — so candidates are settled on the car, not from the
capture.
Probes publish only the first byte; the point is the raw response in the log.
Remove the block once the question is answered.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Real readings from the car on 2026-07-30:
odometer 197394.5 km
odometer at last regen 197055.5 km
distance since regen 339.0 km
dash 195958 km
197055.5 + 339.0 = 197394.5 to the decimal, which confirms bytes 49, 53 and
57 are all decoded correctly. The ECU counter simply runs 1436.5 km — 0.733%
— above the cluster. It exists to time DPF service intervals, is fed by a
different speed source than the legal odometer, and drifts from it over a
vehicle's life.
Apply that as offset_value on the odometer, and on the odometer at last
regeneration since it reads the same counter and the cross-check only holds
if both move together. Distance since regeneration is a difference, carries
no bias, and stays as it is.
One reading cannot tell a constant error from a proportional one. A constant
is assumed for now and the comment records how to settle it: after 100 km or
more, a gap still at 1436.5 confirms the constant, while a gap near 1437.2
means it scales and offset_value should give way to scale 0.000992723.
Drop fuel_percent. Standard 01 2F answers NO DATA on this car — OBDb's Accent
capture that recorded it as supported is a North American petrol car — and
the cluster refuses 22 B002, so there is no route to fuel level here.
Drop probe_2103 now that the real sensors report, and give both DPF sensors a
state_class so Home Assistant stops warning that a unit is set without one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Replace the log-only probe with actual sensors now that OBD sensors can
extract values by byte offset. All three come from the same 21 03 response on
the engine ECU:
offset 49 odometer at the last DPF regeneration
offset 53 distance driven since that regeneration
offset 57 total odometer
Each is a 32-bit big-endian value scaled by 0.001 to give kilometres. The two
DPF fields are marked diagnostic and exist as a cross-check: the value at 49
plus the value at 53 should equal the value at 57, which tells us the offsets
are right rather than merely plausible.
Whether this car's 21 03 frame reaches byte 57 is still unverified. If it does
not, decodeStructured returns null and nothing is published, so the sensors
stay empty instead of reporting a wrong distance. The probe is kept for now so
the raw response still lands in the log either way.
Document `decode` on OBD sensors in both READMEs.
Needs hassble-android feat/obd-offset-decode.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The Accent does not expose the odometer through the cluster. OBDb's captured
command support for real cars records 7C6.7CE.22B002 as unsupported on the
2013, 2015 and 2019 Accent (2018 is the lone exception), which is exactly the
7F 22 22 conditionsNotCorrect this car returns. So obd_hyundai_kia, which is
correct for the 35 other HKMC models, cannot work here.
Add a template for the Accent RB diesel (1.6 TCI-U2) carrying what is known
to work on this car, plus a probe for the remaining unknown:
- fuel level as a percentage via standard 012F, recorded as supported on 7E0
for every captured Accent year
- intake temperature and barometric pressure from 21 01, taken from the
awesomeit profile for this exact model, year and engine
- probe_2103, which requests 21 03 purely so the response lands in the app
log as raw hex
21 03 on the engine ECU is where OBDb's make-wide HYUNDAI_ODO_V2 lives, at
bytes 57-60 divided by 1000. Two neighbouring fields sit in the same frame —
the odometer at the last DPF regeneration at bytes 49-52, and the distance
driven since that regeneration at bytes 53-56 — and they are distinct values,
not the odometer. The awesomeit profile for this car reads bytes 53-55 of the
same 21 03 response, so the frame is known to arrive at least that far; the
odometer is two bytes beyond.
The probe is needed because formula variables only reach byte 19, so byte 57
cannot be addressed yet. Once the raw response confirms the frame length and
the value matches the dash, OBD sensors need offset-based decode and the probe
can be replaced by a real sensor.
Note the exception in obd_hyundai_kia so it points at this template.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The comment claimed some Kia clusters shift the odometer to i:k:k:l. That
misread a search snippet: bytes j:k:l are a separate field holding the same
odometer in miles for North American clusters, not an alternate km layout.
Verified against OBDb (obdb.community), where 11 Hyundai/Kia models —
Elantra, Sonata, Tucson, Kona, Santa Fe, Veloster, Forte, K5, Seltos,
Sportage, Telluride — all publish the identical layout for 7C6 / 22 B002:
e fuel level (liters)
f cluster volts div 12.8
g:h:i odometer 24-bit km
j:h:l odometer 24-bit miles
OBDb's 12.8 divisor for byte f equals the 0.078125 multiplier already in the
template, which confirms OBDb indexes bytes the same way Torque letters do,
i.e. excluding the 62 B0 02 echo. So g:h:i is right as written.
Fuel level stays at e*0.5. OBDb is split — Santa Fe and Soul carry div 2
while ten others carry no scale and max 255, the uncalibrated 8-bit default.
The awesomeit plugin (35 models) and the Elantra CN7 list both scale by 0.5,
and e*1.0989 % maps to a 45.5 L full tank, so 0.5 is the calibrated value.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The obd_hyundai_kia template shipped four PIDs that belong to other makes, so
a Hyundai or Kia never answered them:
22 132A → GM/Chevrolet (Malibu, Cruze, Trax, Spark)
22 2203 + ATSH743 → Renault Samsung (SM5, SM6, QM3, QM6)
22 FD40 → SsangYong (Korando, Rexton W, Tivoli)
22 F22E → SsangYong Tivoli
The odometer formulas were wrong independently of the PID: FD40 reads the
odometer from bytes U:V, not a:b.
Replace them with the HKMC instrument cluster DID, which carries odometer,
fuel level and cluster voltage in one response:
ATSH7C6 + 22 B002 → 62 B0 02 [a][b][c][d][e][f][g][h][i] …
e → fuel level e*0.5 L
f → cluster volts f*0.078125 V
g:h:i → odometer 24-bit km
Cross-checked against the Elantra CN7 PID list, JejuSoul/OBD-PIDs-for-HKMC-EVs
and the awesomeit Torque plugin database (35 HKMC models, 2013 onwards).
Requires hassble-android to group sensors by (mode, pid) and to expose
response bytes past h — all three sensors share one PID and the odometer
needs byte i.
Also document the formula byte range and the ATSH restore rule in both READMEs.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
1. parking_beacon: manufacturer_hex_prefix "04" 추가
- SmartKey 인증 패킷(07xx, 22B)이 parking_floor/location을 오염시키던 문제 수정
- length: 14 (정확한 길이 일치)로 22B 패킷 디코드 단계에서도 차단
2. apt_key_door 신규 추가
- manufacturer_id 861 + prefix "02" → 출입 이벤트 6바이트 패킷만 매칭
- instance_mode: shared, MAC 지정 없음
- door_status: entering / exiting / idle 텍스트 센서
- publish.min_interval: 1s (기본 10s 오버라이드, 이벤트가 수 초 내 전환)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove redundant OBD initialization commands (ATZ, ATE0, ATL0, ATS0, ATH0) from configs. config.yaml now only uses ATSP6 for the device, and templates.yaml clears init_commands for one template and sets only ATSP6 for the other. This reduces unnecessary reset/echo/linefeed commands and relies on explicit protocol selection.
state_class: none is parsed as the string "none" by kaml/snakeyaml (not null).
ConfigValidator treated any unknown state_class as ERROR, so parking_floor was
silently excluded from entity declarations — it never reached Home Assistant.
- parking_floor: changed to state_class: measurement (correct for floor number)
- parking_location: removed state_class: none (meaningless for text_sensor)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
README.md rewritten as English-only with cross-link to README.ko.md.
README.ko.md added as Korean-only translation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Full documentation of every config.yaml field — root structure, defaults,
all three device sources (advertisement/gatt_notify/obd), match conditions,
decode types, controls, publish rules, OBD preset table, and 5 full examples.
Includes common pitfalls (hex literals, AND-logic, instance_mode, source_field).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
service_data_uuid + manufacturer_id 두 조건 AND 매칭이 실패 원인:
Android BLE 스캔 결과에서 F51C 서비스 UUID가 service data 또는 service UUID
목록에 포함되지 않으면 항상 NO MATCH 발생.
Fix: service_data_uuid 제거, manufacturer_hex_prefix "0215" (iBeacon 타입 마커)로
대체 → Apple 제조사 데이터(ID 76)로 시작하는 iBeacon 포맷 기기만 매칭.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Configure MAC-per-instance Jaalee decoding and manufacturer 0x035D parking time, battery, and location.
Co-authored-by: Cursor <cursoragent@cursor.com>