See how the code was written, not just whether it passed

Secure editor, or AI allowed on the record

Your choice per invitation. Blocked: copy and paste are refused and every attempt is logged, clipboard size included. Allowed (Growth and above): the candidate gets an AI assistant inside the editor, every prompt and reply is recorded, and the review adds a note on how well they used it.

Server-scored session telemetry (primary in-editor signal)

Paste volume, focus switches, typing rhythm, and whether the finished code traces back to keystrokes we actually observed, all re-scored on our servers. The reviewer also sees each time the candidate ran the visible tests and how many passed, so you can tell who tested as they went. Missing or stripped telemetry never scores clean.

Public git process evidence (take-home path)

Candidates can submit a public GitHub, GitLab, Bitbucket or Codeberg repo instead of a zip. We read its commit history and flag what doesn't look like a real working session: one bulk commit, a brand-new repo, work outside the assessment window, a different author. History can be rewritten, so we treat it as corroboration — never a sole reason to reject.

Cross-candidate similarity

Submissions are compared against other candidates at your company in the same language — same task and cross-task, with separate thresholds. With nothing to compare against, the check completes as no match instead of hanging. There is no shared public fingerprint corpus yet, so the first hire on a brand-new task starts without a baseline.

Public repository search

Two distinctive lines of a submission are looked up together in public GitHub repositories. When a file contains both, the repository link ships with the flag. This is verbatim search — a reworded copy is not caught, and a solution too short to search is reported as unsearched.