Thank you for the message. If there is a mistake in any sentence and you leave a comment, but the author does not reply in 14 days, let a corpus maintainer know, and they can make the appropriate changes for you. (It is one of our rules)

Note that on our Downloads page, we provide two files for the sentences. One of the files contains more information about the sentence, including the username of the owner of the sentence.

When you extract Kabyle sentences from the CSV file, you can define a whitelist and extract Kabyle sentences only from the whitelisted users, or you can define a blacklist and exclude all sentences from the blacklisted users.

I know it can be frustrating to see bad quality sentences in the corpus, but whitelisting/blacklisting is one good way to filter the corpus to better meet your quality standards.

You can also take into account information from the "User skill per language" to extract only sentences from native speakers, or you can use the "Users' sentence rating" to exclude sentences on an individual basis.

For the sentence rating, this requires you and your peers to enable the feature from your Settings, under "Experimental options" > "Activate the feature to rate sentences...". You can then start marking bad sentences as "not OK".
In your script that extracts the Kabyle sentences, you can then define a list of trusted users, and exclude any sentence that has been rated "not OK" by your trusted users.

This is just to say, Tatoeba provides some basic mechanism for you to filter out bad quality sentences. You'll need of course to spend a little bit more effort on your script, but then won't have to worry about every single sentence that is added into the Kabyle corpus.