
1. Give It a Role and a Reason
We reviewed the same handbook paragraph twice. Without a role: 134 words of writing advice. With one line naming the job: 235 words about the ambiguities that cause disputes.
Tag
10 pieces

We reviewed the same handbook paragraph twice. Without a role: 134 words of writing advice. With one line naming the job: 235 words about the ambiguities that cause disputes.

We put a prompt injection inside customer tickets and ran it with and without tags. Both ignored it. Here is why we still tag everything, and it is not the reason you think.

We asked for one set of numbers in three formats. The table came back with four rows when we only supplied three facts, and the fourth was a metric it derived itself.

We ran two trap problems while actively discouraging working out loud. It got both right. So what is step by step actually buying you? Not accuracy.

We built one prompt from an empty box to a saved template. The improvement did not come from a technique. It came from 90 seconds answering one question.

Same notes, same model. The one shot email never mentioned the competitor. Splitting it into analyze-then-write pulled the deadline forward and addressed the risk directly.

We hid one fact at line 9, line 131 and line 251 of a 260 line document. It found all three, with correct line numbers. The real limits are somewhere else entirely.

Four ideas, no mathematics, each one explaining a failure we hit while writing this course. Including why we will not call it autocomplete.

We ran one prompt five times without changing a character. Output ranged from 80 to 100 words. If you compare one run to one run, you measured the dice.