TL;DR — The mystery byte in a Wolf read telegram is a CRC8 with polynomial 0x5c over the
low byte of the TelegramNr, XORed with the high byte. That was the last thing standing between
Wolf's own published parameter database and a complete ebusd config. You no longer need to
copy magic bytes out of somebody else's 08.csv. There is a script below that generates a
config for any of 48 Wolf device templates. Writes work too, and the write format is
different from the read format in a way that explains why so many people's w rows do nothing.

Tested on a Wolf FGB-K-28 (ebusd 25.1, eBUS shield, ioBroker). 22 of 22 generated IDs are
byte-identical to a hand-made config that has worked for years, plus 30 new registers that were
never readable before. Please try it on your CGB / CGB-2 / CGG-2 / COB / CHA / BM-2 and report
back — the derivation is device-independent but I can only prove it on the hardware I own.

1. Why this has been hard

Wolf publishes no eBUS documentation. Every Wolf config in the wild is hand-made, partial, and
propagated by copy-paste. --scanconfig can never help: a Wolf boiler answers the scan with
MF=65; ID=" &", ebusd does not know manufacturer 0x65, and the ID looks like line noise.

Reads are ZZ=08, PBSB=5022, three ID bytes:

r,heating,dhw_temp,,,08,5022,cc0e00,...
                            ^^^^^^
                            ?? 0e 00      0x000e = 14 = TelegramNr

The last two bytes are obviously the TelegramNr little-endian. The first byte is the problem.
It is different for every register, the forums call it "the CRC", and the accepted workaround has
been: if nobody has published the byte for the register you want, you cannot read that register.

2. The checksum

It is linear over GF(2), which is what makes it recoverable. Take known-good (byte, TelegramNr)
pairs out of any working Wolf 08.csv, treat each output bit as a linear equation over the 16
index bits plus a constant, and solve. On 22 verified pairs that is heavily over-determined and
gives a unique answer, whose bit-effects are exactly:

function wolfCrc(telegramNr) {
    let c = telegramNr & 0xff;
    for (let i = 0; i < 8; i++)
        c = (c & 0x80) ? (((c << 1) & 0xff) ^ 0x5c) : ((c << 1) & 0xff);
    return (c ^ (telegramNr >> 8)) & 0xff;
}

const hx = n => n.toString(16).padStart(2, '0');
const readId = tn => hx(wolfCrc(tn)) + hx(tn & 0xff) + hx((tn >> 8) & 0xff);

A CRC8 chain with polynomial 0x5c, init 0, over the LOW byte only — then the HIGH byte is
XORed in and never shifted. That last part is why it is not any textbook CRC8 and why
table-driven guesses fail.

Checks that matter:

• It reproduces 22 of 22 IDs in a config that has worked for years.
• The linear solve leaves index bits 10–15 unconstrained (no training sample used them, since
  all known registers were < 1024). The poly-0x5c form pins them as identity on the high byte.
  That extrapolation is confirmed on real hardware: TelegramNr 10010 (high byte 0x27, far
  outside the training set) generates 771a27 and reads 1.80 bar of system pressure on my
  boiler. Same for 10013, 10067, 10068, 65019.
• TelegramNr 554 and 644 both produce de — a genuine collision, because 0x2a and 0x84 shift
  to the same value. The high byte disambiguates them. If you have been confused by two registers
  sharing a first byte, that is why.

Bonus: running this over an existing config is a good way to find typos. It flagged one in
mine — hg74 had d5 where the algorithm says 91. It had never worked and nobody knew why.

3. Writes — and why yours probably do nothing

Writes use a different address and a different ID format:

w,heating,dhw_target_set,Warmwassersolltemperatur eingestellt,,03,5023,001300,dhw_target_set,,SIN,10,°C,,suffix,,HEX:4,=9d010000,,

Two things people get wrong:

1. The destination is the master address 03, not 08. I had this recorded as "the write
   rows are mis-addressed" — that was wrong, and it had simply never been tested. Sending the same
   write to ZZ=08 gives ERR: read timeout.
2. ebusctl write printing empty is SUCCESS, not an error. A Wolf write is a master-master
   telegram with no data response. empty is what a successful one looks like.

Proof, on a deliberately harmless register (hg07, heating pump run-on, valid range 0–30 min):

$ ebusctl read -m 0 -c heating hg07
2
$ ebusctl write -c heating hg07 3
empty
$ ebusctl read -m 0 -c heating hg07
3

And the one that matters — the DHW setpoint, TelegramNr 19, write ID 001300:

before: 45.0
empty
after:  50.0
live:   50.0      <- TelegramNr 3, the derived setpoint, followed

Test writes on something harmless first. hg07 or hg09 are ideal: in-range values change
nothing you will notice, and they read straight back.

4. Getting the register list for your Wolf device

You do not have to guess which TelegramNrs exist. Wolf's own SmartSet/ISM7 metadata is public
inside zivillian/ism7mqtt — parameter.xml,
converter.xml and device.xml under src/ism7mqtt/Resources/. Between them they give, for
every Wolf device: TelegramNr, data type, unit, min/max, and the value labels for enums.

wolf-registers.js (attached) downloads them and does the join:

$ node wolf-registers.js --list
  DTID    WRSDeviceIds  params  Name
  30000   0x18          80      BM
  220000  0x20          139     BM-2
  80000   0x6           45      CGB
  180000  0x21          97      CGB-2
  100000  0x1D          53      CGG-2
  110000  0x1C          53      CGU-2
  270000  0x29          178     CHA
  90000   0x1A          49      COB
  250000  0x26          64      FGB
  ...48 templates

Which one are you? Look at your ebusd scan string. Mine says MF=65; ID=" &". Those ID bytes
are 0x20 0x26 — and & is 0x26, which is exactly the WRSDeviceIds of the FGB
template. The "garbage" ID is not garbage; it contains your device id. Match it against the list.

$ node wolf-registers.js --device FGB
TelegramNr  readID   writeID  Type    Unit   Range        Name
13          280d00   000d00   SS10    °C                  Kesseltemperatur
14          cc0e00   000e00   SS10    °C                  Warmwassertemperatur
19          541300   001300   SS10    °C     30..80       Warmwassersolltemperatur eingestellt
270         cd0e01   000e01   SS100          0..30        Heizkurve
10010       771a27   001a27   SS100   bar                 Anlagendruck
...

$ node wolf-registers.js --device FGB --csv --write > 08-generated.csv

Type mapping: SS10→SIN divider 10, SS100→SIN divider 100, US10→UIN divider 10,
US→UIN, SS→SIN.

5. Things that cost me hours

ebusctl read -h only works for messages ebusd ALREADY KNOWS. It is not a raw frame sender.
An unknown ID returns ERR: element not found without touching the bus. So you cannot use it to
probe for new registers, and "a wrong checksum is rejected" proves nothing about your boiler — it
only proves ebusd's message table did not match.

ebusctl hex and read -def need --enablehex / --enabledefine. Off by default.

Validate a config offline, as a normal user, before installing it:

ebusd --checkconfig --configpath=/home/you/ebustest/ --device=ens:127.0.0.1:59999

Copy your config tree somewhere writable, edit, check. It prints found messages: N (... M poll)
and exits 0 without touching the running daemon or the bus.

Do not force-read many registers back to back. I ran 32 sequential read -m 0 calls and
roughly every other one came back empty, with the symbol rate spiking to 102. It looks exactly
like "this register is unsupported" and it is not. Read the poll cache instead —
ebusctl find -a -c heating — giving it a couple of minutes after ebusctl reload to refill.

Adding a write row whose name matches an existing read row does NOT increase messages:.
ebusd merges them into one message with two directions. Check with ebusctl find -w -c heating,
not the count. I briefly misread this as a rejected row.

Polling more costs nothing. I went from 28 to 60 polled messages and the bus symbol rate
stayed at 12, exactly as before. Poll priority (r1…r9) controls how often each is polled,
not how much traffic there is in total.

Values reading 32768 or 3276.8 are the eBUS null sentinel — "this register does not exist
on this device". Six of mine had been dead for years. Their TelegramNrs (39, 321, 362, 365, 368)
simply are not in the FGB parameter list; they had been copied from a different Wolf model. No
config change will ever fix those, and now you can tell the difference between "wrong byte" and
"not present" by checking the generated list.

6. ioBroker integration

MQTT. ebusd's own MQTT integration publishes reads fine:

--mqttint=/etc/ebusd/mqtt-integration.cfg --mqtthost=127.0.0.1 --mqttport 1883 ...

Note it ships filter-direction = r|u, i.e. read-only — it will not accept writes. To write
from a script, shell out:

exec('ebusctl write -c heating dhw_target_set 50', () => {
    exec('ebusctl read -m 0 -c heating dhw_target_set', (e, out) => log('now ' + out));
});

Always read back and verify. A write that silently does nothing is the failure mode here.

Poll priority gotcha. If you use the Home Assistant MQTT profile, filter-seen = 5 silently
promotes matched messages to poll priority 5. Remove it and everything goes to no data stored
because nothing is polled any more. Declare priority in the CSV instead (r5,heating,...).

Two scripts are attached:

• boiler_status_en.js — turns the raw enum numbers into readable states under
  0_userdata.0.heating.boiler.*: status, mode, burner_status, firing, fault, alive,
  plus a one-line summary like Firing at 28% — Heating mode. fault is the one worth
  alarming on: true when burner_status is 12 (Störung) or the safety limiter has tripped.
  alive watches staleness across several polled registers, because the boiler powers the eBUS —
  so a dead bus means a dead boiler.
• boiler_aliases.js — a stable alias.0.boiler.* layer so renaming an ebusd message does not
  break every widget and chart. It refuses to create an alias whose source does not exist and
  tells you which ones to fix.

Useful enum decodings (from Wolf's own KeyValueList, FGB — check yours with the tool):

burner_status  0 Aus · 1 Ruhekontakt · 2 Vorspülen · 3 Zünden · 4 Stabilisierung · 5 Ein ·
               6 Ventilprüfung · 7 Nachspülen · 8 Softstart · 9 Taktsperre ·
               10 Betrieb ohne Brenner · 11 Abgasklappe · 12 Störung · 13 Gradienten Überwachung ·
               14 Gasdruck · 15 Spreizung hoch · 16 Spreizung KF
operating_mode 0 Test · 1 Start · 2 Frost Heizkreis · 3 Frost Warmwasser · 4 Schornsteinfeger ·
               5 Kombibetrieb · 6 Parallelbetrieb · 7 Warmwasser · 8 Warmwassernachlauf ·
               9 Mindest-Kombizeit · 10 Heizbetrieb · 11 Nachlauf Heizkreispumpe ·
               12 Frostschutz · 13 Standby · 14 Kaskadenbetrieb · 15 GLT-Betrieb ·
               16-19 Kalibration · 20 WW Schnellstart · 21 Externe Deaktivierung
program (10100) 1 Aus · 3 Winterbetrieb · 5 Sommerbetrieb
TelegramNr 370 bit 3          = Brenner
TelegramNr 371 bits 0,1,2,4,5 = Zündung, Ventil 1, Ventil 2, Kesselkreispumpe, 3-Wege-Ventil

7. Safety

This is a gas appliance. A few things I would not do:

• Do not write to a register you have not looked up. The generated table gives the valid
  range; stay inside it. Writing outside a documented range is not something I have tested and
  not something I would.
• Prove writes on a harmless parameter first (hg07, hg09), read back, restore.
• Do not try to make ebusd the heating controller. Emulating the controller broadcast means
  repeating it forever, and if your automation dies in winter so does your frost protection. For
  simple on/off demand a relay across the room-thermostat contact fails safe; this does not.
• Back up 08.csv boiler_aliases.js boiler_status_en.js wolf-registers.js before every change and keep ebusd --checkconfig in the loop.
• Writes change settings in the boiler, persistently. They survive a reboot of everything
  else. Write down what you changed.

8. What it got me

A Wolf FGB-K-28 that exposed 28 registers, six of them permanently dead, now exposes 60 — at
identical bus load. New and working:

Anlagendruck (system pressure, bar) · Abgastemperatur (flue gas) · Modulationsgrad ·
Brennerstatus (17 states) · Betriebsart (20 modes) · Brenner, Zündung,
Ventil 1/2, Kesselkreispumpe, 3-Wege-Ventil (six real digital signals) ·
Drehzahl Kesselkreispumpe / ZHP WW · Heizkurve · Firmware · Anlagendruck ·
Auslauftemperatur Warmwasser · Sicherheitstemperaturbegrenzer · Anzahl Netz-Starts ·
and roughly fifteen more.

Plus working writes, which turned into an automated weekly DHW disinfection cycle and a
vacation setback.

Three errors found in a config that had been running for years: a wrong checksum on hg74,
hg75 typed as minutes when TelegramNr 364 is litres per minute (it is the DHW flow rate),
and six registers that do not exist on this device at all.

Files

Credit

The parameter database is zivillian/ism7mqtt's extraction
of Wolf's SmartSet metadata — without it none of this is possible. ebusd is
john30/ebusd. The forum threads that got me oriented are
ebusd discussion #779 and #721.

Corrections welcome, especially from anyone who can confirm or refute the checksum on a non-FGB
Wolf device.