How These Pages
Are Made

Set out plainly, so no one has to guess in either direction.

• The questions are mine. What gets asked and answered here comes from my own study, or from family and friends who asked me something I could not answer in a sentence. Nothing is written because a topic seemed popular.
• The standards are mine, and they are written down first. DOCTRINE.md fixes the theology: where we stand, how we read, and the list of teachings a page must never make. CLAUDE.md fixes the design and the page templates. Both are in the public repository. Every page is written to them rather than to whatever seemed good that day.
• The research is the AI's. Gathering the passages, tracing how a text has been read down the centuries, summarizing what a teacher actually argued and where he wrote it: that work is the model's, and it is more of it than I could do alone.
• Most of the prose is the AI's. First draft and usually the tenth. I redirect, cut, press for what a page is dodging, and send work back. But if a sentence here reads well, assume the model wrote it.
• The building is the AI's. These pages are generated by Python scripts in the repository, not typed by hand, which is why 66 books can carry one consistent template.
• The review is mine, and it decides what ships. Every page is read against the checklist in DOCTRINE.md before it goes up. Pages go back. Some do not go up at all.
• What stands, I stand behind. Nothing is published that I am unwilling to put my name to. That is the one line that makes the rest of this arrangement defensible.

Knowing the tool's failure modes is more use to you than a general warning, so here they are, named.

• Fluent and wrong is the failure mode. A language model produces confident, well-formed prose whether or not the claim underneath is true. It does not sound less certain when it is guessing. So on these pages, confidence is not evidence. Polish is not proof.
• Quotations carry the highest risk. A model can generate a sentence that sounds exactly like Calvin and that Calvin never wrote. This is the single most dangerous thing it can do to a commentary, because a fabricated quotation borrows a dead man's authority. It is why the Berean rule in DOCTRINE.md forbids attaching a name to a plausible guess, and why our pages say which lines are quoted and which are our summary of an argument.
• References can be right in form and wrong in content. A chapter and verse that looks correct is easy to produce. Look them up.
• It reaches for symmetry. A neat structure is easier to write than an honest mess. Where a page says a question is unsettled, or that the evidence runs out, that is usually a place I pushed back, and it is the part of a page most worth reading.
• It has no faith and nothing at stake. A model can explain assurance without having any, and write about the fear of God without fearing Him. Nothing here is true because a machine put it well. It is true if Scripture says it.

Every page here is one family's reading, and everything quoted from anyone else is a man's opinion about God's Word, never a substitute for it. That was always true of commentaries written entirely by hand. The tools change; your job does not.

So read these pages the way the Bereans read Paul. Check the references. Weigh the quotations. Test the argument against the whole of Scripture, and where this site and the text disagree, the text wins and we will fix the page. If you find something wrong here, telling me is a kindness, not a complaint.