An AI score is a signal, not a verdict. Codequiry's detector scores each file from 0–100% based on statistical patterns characteristic of LLM output (consistency of indentation and naming, idiom choices common to ChatGPT / Claude / Copilot, low-entropy stylistic tells). It is a probability, not proof.

Known false-positive sources. Boilerplate or instructor-provided starter code, students who write very clean idiomatic code, code that follows a strict style guide, and short snippets without enough material to analyse can all score higher than they should. Enable Auto-Detect Boilerplate Code in Settings to mitigate the first case.

Before acting on a high score: open the submission in the Result Driller to read the actual flagged code, check whether peer or web similarity corroborates, and consider an oral defense or in-person follow-up. We never claim certainty — treat a high score as a reason to open a conversation, not to close one.