Michlaude Discipleship Project

Appendix E. Where these tests came from, in short and honest

Who did what

AI did much of the work. I made every decision.

A blind reader is an AI in a fresh chat. It gets the rules and some cases. It is not told what answer I expect.

An assessor is another AI, in a separate chat. It sorted what the blind readers found. Which findings were real breaks, where a rule gave a wrong answer, or two answers? Which were limits to state? Which were out of scope? Then I answered the open questions.

A skeptic is an AI asked to raise the strongest objections one view would raise.

A last reader, one more AI, took the pooled objections, with the names removed and the order mixed. It marked each part of each objection in one of four ways. The rules agree with the claim. The rules give the opposite. The rules leave it open. Or no rule speaks to it. An agreement does not mean the objector's own reading has been shown. A mark is never a verdict on a view or a person.

I also wrote a long paper, which I call the long argument. In it I apply these rules to Jesus' talk on the Mount of Olives. This book does not argue it. An AI editor builds each copy of it and checks it. AI scripts check every quote. Then the editor and I read. I decided every status. I did not hand that to a machine.

Two kinds of round

A round is one full pass of testing. There were two kinds.

A rule round tests the wording of a rule. Blind readers read it, and then an assessor sorts what they found. Seven of the rules in this book went through three rule rounds, with two blind readers each time and fixes between rounds.

An argument round tests one copy of the long argument, in three steps. First blind readers, then three skeptics, then the last reader. It tests my long argument. It does not test the Bible. A round is clean when no new confirmed error turns up in it. A confirmed error is a problem I checked and found true against the NET, the Greek and Hebrew word list, or the source the argument cites. A misquote is one. A wrong reference is one. So is a claim that goes against a verse it cites. A note that only asks me to mark something as my own reading is not an error.