Diagnose. Find the flaw in someone else's proof.
A tutor that explains everything is a tutor that prevents you from thinking. This one does the opposite: it presents you with proofs that are almost right — proofs that look correct, sound plausible, and end at the intended conclusion — but that quietly cheat at one specific step. Your job is to find the step, name the rule it violates, and explain how a real proof would have made the move legally.
Not a chatbot
You are not asking the tutor to explain. You are diagnosing its work. The asymmetry matters: you are the active agent.
Errors are designed
Every flaw maps to a real, named failure mode — petitio principii, anachronistic dependency, unjustified existence, eyeball-equality, wrong common notion.
Wrong looks right
The figures and proofs are deliberately convincing. If they did not feel correct, the exercise would be useless — and so would the muscle it builds.
Five near-cases.
Canonical diagnosis
The rule that was violated
Tutor settings
No key configured.
No key yet? Mint one at console.anthropic.com → API keys. An Anthropic account with a small prepaid balance is required (see pricing). Each tutor call is short — typical cost per diagnosis is a fraction of a cent at current Sonnet pricing, and the doctrine prompt is cached so subsequent calls are cheaper still.
Your key is sent directly from your browser to api.anthropic.com. It is not transmitted to or stored by Euclid Lab. Keys held in sessionStorage clear when the tab closes; localStorage persists across sessions and should be treated as XSS-reachable — use it only on devices you control.
The full text of every prompt the AI receives is published at tutor-prompts. No hidden system messages.