The substitution that changes the result
Lesson 2 of 6 · Methods at the bench
1 · Learn the move · Cognitive verifier
The dangerous substitution is the reasonable one: a different detergent from the same shelf, the same enzyme from a different vendor, the same reagent from a new lot. Asked directly, a model will often bless the swap, because the two look interchangeable in text. The verifier decomposes instead: what job does the original do in this step; which of its properties does that job depend on; where does the substitute differ on exactly those properties; which downstream step would show the difference; what side-by-side control would detect it. The model never concludes a substitution is safe or equivalent - it lists what would have to be true, and equivalence is an experimental result from the paired run. Anything with a hazard dimension routes to your EHS office before the bench.
I want to substitute [substitute] for [original] in [step] of [protocol]. Do not tell me whether it is fine. Instead decompose: (1) what the original does in this step, (2) which properties that function depends on, (3) where the substitute differs on those properties, (4) which downstream steps would show a difference, (5) the side-by-side control that would detect it. Mark anything you are unsure of UNKNOWN. Any hazard aspect: list it as a question for EHS, not an answer.
2 · Your turn. You write the prompt
Your usual lysis detergent is backordered six weeks. A labmate says the nonionic detergent on the next shelf 'is basically the same thing' and the model you asked last night agreed in one breezy sentence. Two months of prep work rides on the next three purifications, and the swap decision is yours.
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.