Troubleshooting a failed prep, systematically
Lesson 3 of 6 · Methods at the bench
1 · Learn the move · Flipped interaction
After a failed prep, the temptation is to ask what went wrong and take the first fluent answer - and a model will always have one, assembled from every troubleshooting page it has seen, connected to your bench by nothing. Flip the interaction: the model asks, you answer, one question at a time. What changed since the last working run - reagent lots, water source, timings, temperatures, operator, instrument. Only after the interview does it rank hypotheses, ordered by the cost of the check that would discriminate them, cheapest first. The model proposes checks; it does not diagnose. A diagnosis is what you call the hypothesis after your discriminating check confirms it, and the interview is what keeps the check list about your bench instead of everyone's.
My [prep/reaction] failed: [one-line symptom]. Interview me before proposing anything - one question at a time, starting with what changed since the last working run: lots, dates, timings, temperatures, operator, water, instrument. After the interview, rank hypotheses with the cheapest discriminating check first. Propose checks only; do not diagnose. Stop when the checks are cheaper than continuing to ask.
2 · Your turn. You write the prompt
Plasmid preps that ran fine for a year dropped tenfold in yield this week - for everyone in the lab at once, on different constructs. Three people have three theories and someone has already ordered a replacement kit on the theory it is the columns. You want the cheapest path to the actual cause.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.