[{"data":1,"prerenderedAt":104},["ShallowReactive",2],{"content:\u002Fglossary\u002Ffalse-positive":3,"ArticleBody_uJWS2j03FGU0uh4CS0F2cRZPXNJDgsyzTjqLC0n28":99},{"kind":4,"path":5,"locale":6,"title":7,"htmlTitle":8,"description":9,"navTitle":7,"published":10,"updated":10,"alternates":11,"breadcrumbs":12,"indexable":20,"twin":21,"collection":22,"slug":23,"file":24,"content":25,"hub":18,"titleEm":27,"eyebrow":17,"lead":28,"cta":27,"faq":29,"faqJsonLd":45,"sources":51,"related":77,"topic":27,"jsonLdType":97,"blocks":98,"shortDefinition":28,"feed":20},"term","\u002Fglossary\u002Ffalse-positive","en","False positive","False Positives in Content Moderation · Detectivision","A false positive is harmless content a moderation system flags. Why they happen, how to measure them on your own data, and how thresholds and review help.","2026-10-16T18:00:00+03:00",[],[13,16,19],{"name":14,"path":15},"Home","\u002F",{"name":17,"path":18},"Glossary","\u002Fglossary",{"name":7,"path":5},true,"\u002Fglossary\u002Ffalse-positive.md","glossary","false-positive","content\u002Fen\u002Fglossary\u002Ffalse-positive.md",{"collection":26,"path":5},"glossary_en",null,"Content a moderation system flags although it breaks no rule, such as a harmless word that looks like a slur. Its opposite is a false negative.",[30,33,36,39,42],{"q":31,"a":32},"What is the difference between a false positive and a false negative?","A false positive is harmless content that gets flagged, such as a town name that contains a swear word. A false negative is harmful content that gets through, such as an insult with a deliberate misspelling. Moving your line trades one for the other, so decide per flag which mistake costs you more.",{"q":34,"a":35},"What is a good false positive rate for content moderation?","There is no universal number. Measure precision and recall per flag on a labelled sample of your own content, then set targets by what each flag triggers: you can accept more false positives when they only send an item to a person, and fewer when they remove a post or ban an account. Averages from other platforms describe other content, other rules and other models.",{"q":37,"a":38},"Does a higher confidence mean the content is worse?","No. Confidence says how strongly the check found a category; it is not a severity scale. An insult at 0.95 means the check is surer that it is an insult, not that the insult is worse than one at 0.6. Confidence is also not comparable across flags, so 0.8 for hate and 0.8 for insult do not mean the same.",{"q":40,"a":41},"How many items should I label to measure false positives?","There is no fixed number. Precision needs enough flagged items for each flag you act on; recall needs enough rule-breaking items, which are rare, so sample flagged and unflagged content separately and label more where mistakes cost most. Measure again after any change to your rules, your lines or the flags you check.",{"q":43,"a":44},"Should a false positive count against the user?","No. When a person overturns a flag, clear it from the user's record, so strikes, rate limits and trust scores reflect what the user did, not what the filter got wrong. Keep the overturned item: it is a labelled example for your next measurement.",[46,47,48,49,50],{"q":31,"a":32},{"q":34,"a":35},{"q":37,"a":38},{"q":40,"a":41},{"q":43,"a":44},[52,57,60,63,66,70,74],{"title":53,"url":54,"publisher":55,"accessed":56},"Flags","https:\u002F\u002Fdetectivision.ai\u002Fdocs\u002Fflags","Detectivision","2026-10-16",{"title":58,"url":59,"publisher":55,"accessed":56},"Responses","https:\u002F\u002Fdetectivision.ai\u002Fdocs\u002Fresponses",{"title":61,"url":62,"publisher":55,"accessed":56},"Text moderation","https:\u002F\u002Fdetectivision.ai\u002Fdocs\u002Ftext",{"title":64,"url":65,"publisher":55,"accessed":56},"Metadata and actions","https:\u002F\u002Fdetectivision.ai\u002Fdocs\u002Factions",{"title":67,"url":68,"publisher":69,"accessed":56},"Precision and recall","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FPrecision_and_recall","Wikipedia",{"title":71,"url":72,"publisher":73,"accessed":56},"Regulation (EU) 2022\u002F2065, the Digital Services Act (Articles 3 and 15)","https:\u002F\u002Feur-lex.europa.eu\u002Feli\u002Freg\u002F2022\u002F2065\u002Foj","EUR-Lex",{"title":75,"url":76,"publisher":55,"accessed":56},"Service status, GET \u002Fapi\u002Fv1\u002Fstatus","https:\u002F\u002Fapi.detectivision.ai\u002Fapi\u002Fv1\u002Fstatus",[78,81,85,89,93],{"path":79,"title":53,"description":80},"\u002Fdocs\u002Fflags","The categories a Detectivision moderation check looks for, such as hate, insults, sexual content and scams, and how to get your account's own list.",{"path":82,"title":83,"description":84},"\u002Fglossary\u002Fscunthorpe-problem","Scunthorpe problem","The Scunthorpe problem is a filter blocking harmless text because a banned word hides inside it. Why it happens in every language, and the fixes.",{"path":86,"title":87,"description":88},"\u002Fglossary\u002Fhuman-in-the-loop-moderation","Human-in-the-loop moderation","Human-in-the-loop moderation sends what software is unsure about to people. How a review queue works, what to record, and where the law requires a person.",{"path":90,"title":91,"description":92},"\u002Fguides\u002Fcontent-moderation-system-design","How to design a content moderation system: checks, rules and queues","A reference design for moderating user content: where to check, when to block or hold, how to route doubtful cases to people, and how decisions reach your app.",{"path":94,"title":95,"description":96},"\u002Fguides\u002Fai-hate-speech-moderation","Can AI moderate text for hate speech and harassment?","Yes, for most of it: a model reads every message and says which kind of abuse it finds and how sure it is. What it does well, where people decide.","Article",false,["Island",100],{"key":101,"result":102},"ArticleBody_uJWS2j03FGU0uh4CS0F2cRZPXNJDgsyzTjqLC0n28",{"head":103},{},1792166153887]