Skip to main content
DanManREAL/ AI_LAB
Updated Oct 24, 2026Request a test

Test reportsEvery claim has a receipt

Test reports

  • 9 reports
  • 60 receipts
  • 6 kinds of evidence
What should I test next?

Choose a report type to narrow the list.

  1. Field testTested AI33 new working tools. One session.ChatGPT + Codex built 33 new DMR_REMOTE toolsChatGPT4 receiptsRead the report
  2. Field testTested AIUseful review, inconsistent gradesGemini 4 Argon reviewed a finished episode: 14 clips, one flag, one inconsistencyGemini 4 Argon4 receiptsRead the report
  3. Head-to-headTested AIExtra review cost more on both repairsDoes an extra AI review step pay off? A small DeepSeek trial said noDeepSeek V4.1 Flash7 receiptsRead the report
  4. Cost receiptTested AIAbout 1 cent per million tokensMy real DeepSeek bill: 12.31 billion tokens for $122.28DeepSeek V4.1 Flash7 receiptsRead the report
  5. IncidentReliability & safetyA client timeout, not the modelGemini 4 Argon kept hanging in Gemini CLI. The cause was a 60-second timeoutGemini 4 Argon9 receiptsRead the report
  6. Self-reportTested AISelf-reported, not measuredI asked 8 AI coding agents for their biggest weakness8 AI coding agents6 receiptsRead the report
  7. IncidentReliability & safetyInvalid: retired and archivedMy agents' first game benchmark was wrong. Here's how it was caughtAI-built benchmark runner6 receiptsRead the report
  8. IncidentReliability & safetyUpstream hang, not my setupDeepSeek Flash answered 200 OK and sent nothing for two hoursDeepSeek V4.1 Flash8 receiptsRead the report
  9. AuditReliability & safety2 of 18 sessions said done without checkingDeepSeek V4 Flash audited 18 of my real coding sessions. Two said 'done' too earlyDeepSeek V4 Flash9 receiptsRead the report