Please explain "confidence" parameter in errors

Hi guys,
here is an example: (I am using “picky” level)
"Access to clean and safe drinking water is a fundamental human right. "

but the system reports with confidence 0.8 that there should be a comma before “and”

   "rule": {
        "id": "COMMA_COMPOUND_SENTENCE_2",
        "subId": "4",
        "sourceFile": "grammar.xml",
        "description": "comma between independent clauses",
        "issueType": "uncategorized",
        "urls": [
            "value": ""
        "category": {
          "id": "PUNCTUATION",
          "name": "Punctuation"
        "isPremium": false,
        "confidence": 0.81
      "ignoreForIncompleteSentence": true,
      "contextForSureMatch": -1

I would like to avoid these kind of false positive. Can I rely on confidence parameter and filter errors by it? And what threshold would you recommend? 0.9?

Hi, the confidence is experimental and shouldn’t be used for this. But thanks for the report, I’ve forwarded it to our developers.

Thank you @dnaber for your reply. Just wanted to ask in general how to avoid false positives - if there any other parameter we can use? or maybe there is a list of rules which are “problematic” and have issues for being false-positive?

Also the strict picky level - does it have any influence on these false-positive errors?

Sorry, I don’t think there’s an easy way to turn off just the false positives. Picky level generally gives you more complaints, often style-related ones, which some users might consider false positives in a way.