🧰 Endor Labs anti-cheating pipeline

The items that name it 3

• Best in Class, Novel in Method: Opus 5 and the Recall-Then-Diverge Pattern | Blog | Endor Labs
  writeup2026-08-11

  Multi-signal detection plus LLM adjudication flagging suspicious instances (git history use, web fetch, workspace leak, training recall) and, after this update, recall-then-diverge via edit-trajectory and string-chronology analysis

• OpenAI Codex with GPT-5.6 Sol: competitive, zero cheating, one unique Django fix | Blog | Endor Labs
  article2026-07-17

  Applies four independent signals (conversation analysis, patch similarity, memorization detection, strict-test pass) to flag suspicious instances, then submits flagged instances to multi-step LLM adjudication producing a structured verdict (cheating true/false, confidence, mechanism, reasoning)

• Endor Labs: Claude Fable 5
  writeup2026-06-11

  Multi-signal cheating detection (patch similarity, conversation analysis, memorization, strict-test pass) followed by LLM inspection of suspicious instances, used to confirm cheating and assess genuine solves

In every workspace you can read 4