How Pygar handles your questions.

Before Pygar runs the engine, every question passes a brief safeguarding screen. Almost everything goes straight through — the screen exists only to catch a few narrow situations where reasoning is not the right response. When it does step in, your question is neither processed nor stored.

There are three things the screen looks for:

• Someone in crisis. If a question suggests you, or a person you describe, may be at risk, Pygar steps aside and points you to people who can actually help — Samaritans, NHS 111, the National Domestic Abuse Helpline, or the emergency services.
• Clearly illegal acts. Pygar will not help carry out something unlawful. If there is a lawful decision underneath the question, you are welcome to rephrase around that.
• Judging people by protected characteristics. Pygar does not process questions that ask it to profile or judge identifiable people on the basis of special-category data — health, race or ethnicity, religion, political opinion, sexual orientation, and the like. You can usually reach the same decision by focusing on behaviour, performance, or the choice in front of you.

When the screen intervenes, the content of your question is never read by a person, never saved, and never sent to the engine. We keep only a content-free record — which of the three categories was triggered, and when — so we can understand how often the screen fires. Your words are not part of that record.

Everything else is reasoned through as you would expect. Pygar does not moderate the thinking it returns: the panel’s reasoning and the synthesis are never softened or withheld because a conclusion is uncomfortable. The screen guards the input; it never edits the answer.