Plagiarism Detection

When a member submits an application, Docket converts their responses into a word-frequency vector and computes the cosine similarity against every other application in your server's history. This is entirely local — no external API is involved and no data leaves your dashboard.

A similarity score above 68% triggers a plagiarism flag. The exact percentage match is shown to staff on the application card.

• Direct copy-paste of another member's answers
• Slightly rephrased versions of past accepted applications
• The same person applying again with a new account using the same backstory
• Answers that appear to be copied from publicly shared application templates

The system looks at the total combined text across all answers, not each answer individually. This means someone who rearranges question order or splits content differently will still be caught.

When an application is flagged, the staff notification card shows:

• A red Plagiarism Flagged badge on the application
• The percentage similarity match (e.g. 84% match with past application)
• The AI score is still calculated and shown — a flagged application is not automatically denied

Staff decide whether to deny based on the flag. A high similarity with a past denied application is strong grounds for denial. A high match with an approved application may indicate the same person using an alt account.

The default threshold of 68% catches most copied applications while avoiding false positives from common RP phrases (e.g. "I have 3 years of roleplay experience"). If you are seeing too many or too few flags, you can adjust this in Dashboard > Applications > Settings > Plagiarism Threshold.

Setting it lower (e.g. 55%) catches more borderline cases but may flag innocent applications. Setting it higher (e.g. 80%) reduces false positives but may miss looser rewrites.