Please note that the convention for the messages sent to CCDB populator was changed since #6150 in order to avoid clashes between the inputs from different calibrators. The documentation was not fully updated, just pushed #6234.
Now it should be e.g.

output.snapshot(Output{clbUtils::gDataOriginCDBPayload, "MFT_NoiseMap", 0}, *image.get());
output.snapshot(Output{clbUtils::gDataOriginCDBWrapper, "MFT_NoiseMap", 0}, info);

Hi Ruben, thanks for the comments. This is now corrected.

Respectively, in the getNoiseCalibratorSpec the outputs should be declared as

outputs.emplace_back(ConcreteDataTypeMatcher{clbUtils::gDataOriginCDBPayload, "MFT_NoiseMap"});
outputs.emplace_back(ConcreteDataTypeMatcher{clbUtils::gDataOriginCDBWrapper, "MFT_NoiseMap"});

What is the purpose of this option? It will miss the noisy channels which happen to be neighbors (e.g. noisy column).

This was a remanent of what was in ITS code, now removed.

The calibration may start from any orbit, so the nTF in general will not reflect the TFID. In other time-slot-based calibrations we are using auto tfcounter = o2::header::get<o2::framework::DataProcessingHeader*>(pc.inputs().get("input").header)->startTime;. This is also not really correct, but at least, once we come to agreement on the common TF counter definition, we can change all centrally. I would suggest to use the same: extract the TF counter in the CalibrationSpec and pass it to the Calibrator, e.g. like in https://github.com/AliceO2Group/AliceO2/blob/c2f765d42f324beb1e1ecda9cf5ba4acf29704cd/Detectors/TOF/calibration/testWorkflow/LHCClockCalibratorSpec.h#L51-L54
Then, you also will not need the mHBFperTF.
Actually, for you the specific slot should be irrelevant, since you have just 1 with 0 - infinity span.

Thanks for pointing this out. Indeed I had some troubles with this. I think the same line is in the ITS code.

You should set the length of the calibration slot, i.e. in your case infinite (e.g. std::numeric_limits<TFType>::max) since you have just 1 slot.

Just a comment : by default this PR doesn't use the NoiseSlotCalibrator class but the NoiseCalibrator which doesn't use TimeSlots, as Jouri mentioned during the discussion that in the end ITS doesn't use the former. So I included the appropriate line but commented it out.

Since you never need multiple slots, why don't you set it in the NoiseSlotCalibrator constructor?

Should be done now.

Why don't you do directly:

std::map<std::string, std::string> meta;
auto toKeyValPairs = [&meta](std::vector<std::string> const& tokens) {
  for (auto&amp; token : tokens) {
     auto keyval = Str::tokenize(token, '=', false);
     if (keyval.size() != 2) {
        LOG(ERROR) << "Illegal command-line key/value string: " << token;
        continue;
     }
     Str::trim(keyval[1]);
     meta[keyval[0]] = keyval[1];
  }
};
toKeyValPairs(Str::tokenize(mMeta, ';', true));

Thanks, indeed. Now fixed.