Should you trust this #1 claim?

• The claim names a specific leaderboard and version
• A third party ran the test, not the model's maker
• The score shows a confidence interval or vote count
• The lead is bigger than the stated margin of error
• The benchmark matches the work you actually do
• The model is publicly available to test today
• Settings (prompt, effort level, tools) are published
• A second, different leaderboard agrees