Work out the questions worth asking, then answer them

You give it a subject rather than a question. It works out the handful of questions whose answers would actually change what you do about that subject, then researches each one and appends a row per question: the question, the answer, the page the answer came from, and what happened.

Every answer is written from the text of one page that was fetched during the run, and from nothing else. A question that page does not answer is left with an empty Answer cell and a row saying the page did not answer it - it is never filled in from what a model happens to know, because an unsourced answer sitting in a research sheet is indistinguishable from a sourced one a week later. A question no source could be found for is recorded the same way.

Set up: paste your spreadsheet address into 'sheetUrl' and name the subject in 'subject'. Rows are appended to a tab named Questions and answers in this column order: Asked on, Subject, Question, Answer, Source, Status. The Question column is also how it knows what it has already asked, so a question already on the tab is skipped rather than asked twice.

It runs weekly and asks at most 6 questions per run; change 'maxQuestions' to move that. Set it against a subject you want to keep learning about and the tab fills up a few questions at a time.