...
A misprint looks a lot like any other candle with excessive real volume, which is why they are so hard to narrow down. In the ETH/EUR case of June 19, there was a misprint for a few minutes. A moment's pause in our internal system caused this. Normally, the system is cleared out and the visual interface should reflect this but rarely, a misprint can happen.
When the system pause happens, there is no actual volume traded. At the end of this pause, however, the cancelled volume falsely appears within the minute of system restart. This means that a 1 minute candle collects the data for more than a minute. As a result, a few minutes volume becomes falsely visible within a 1 minute candle.
...
Do I understand you correct that one could distinguish between real and false peaks by taking a look at the volume? Thus a peak/dip with very low volume is a false peak/dip, and if it is real it is paired with a large trade volume.
Taking a look at the ETH/EUR peak on June 19 using a 1 h interval, the volume is rather low for this type of peak, especially when comparing to the dip on June 22th which is paired with a rather high volume. However, we need to put ourselves in the situation on June 19th. So, we only look at trade data before the peak and neglect the trade data after the peak. Then the volume is not particularly low, actually it is higher than in most other 1h intervals within the last 48h.
After taking a look at the Dash/BTC market, we could further test this way to distinguish between real and false peaks. 1h intervals are used as it is not possible in Cryptowatch to go back to all peaks using a 15min interval. Times are given in CET.
1) June 29 04:00-06:00 a double peak of 2.5% with very low volume: Real or not real
2) July 3rd around 08:00 a peak of 1.5% with a bit higher volume: Real or not real
3) July 4th around 08:00 big dip and peak, both roughly 4%, relative high volume: Real or not real
According to your way to distinguish, the first is a false peak, as the volume is rather low, and peak 2 and 3 are real because they have relative high trade volumes.
Am I correct, or is number 1 also a real peak?
As you might see, the way to distinguish does not guarantee a correct result. And even if this technique would yield a guaranteed correct result, it seems a bit odd that traders need to check your data for consistency in order to know whether the data is real or false.
Last, but not least. You are right. A misprint/mistake can happen. However, if one is responsible and is aware of a fault then one should try to correct it. Especially, when your customers trust you to display correct OHLC data. Let's not forget that customers pay fees for your service. Furthermore, this mistake is happening on a regular basis since more than a year. Therefore, I doubt that Kraken has taken this issue seriously.