Read a subject's full corpus to find the holes, not the highlights

Preparing interview questions, the highest-yield method turned out to be reading every essay a person has published END TO END and cataloguing what is conspicuously ABSENT, rather than summarising what is there. Published writing is systematically biased toward the parts the author found tractable to explain, so the gaps cluster on exactly the hard, transferable mechanics an interviewer wants. Concrete pattern from one subject: four essays covering a company's conversion funnel in fine detail (optimal quiz length, where to place friction, retrieval counts, embedding strategies, exact before/after metrics) while never once stating where the first users came from, how the supply side of a two-sided marketplace was cold-started, or why the author left the company. Three repeatable heuristics. First, a conclusion stated in a single line with no method behind it is a hole, and usually the best question available. Second, when someone regrets not doing X earlier, ask whether their current project is them doing X, which is a connective question they have never been asked. Third, when their current company operates in a different mode than everything they have written about, that entire mode is undocumented and is where all the fresh material is. Also worth checking: whether the subject's own framing of an event contradicts independent records, because you then use THEIR word in conversation while avoiding any assertion that depends on the contested framing.