Most AI research fails for one boring reason. It blends three things that should never share a sentence.
Fact: what the source literally says. Inference: what you concluded by connecting facts. Opinion: what you believe it means or what it's worth.
When they blend, the memo sounds confident and reads clean. That is the trap. A confident sentence built on a hidden assumption is how a thesis goes wrong quietly.
The fix is a labeling pass. Make the AI tag every line before you read it:
-> [FACT] only if it is stated in the filing, the transcript, or the comp set, with a page or timestamp. -> [INFERENCE] when it connects facts into a conclusion, and the assumption gets written down next to it. -> [OPINION] when it is a judgment about quality, direction, or value, carrying no source.
Then read in that order. Facts you verify. Inferences you stress-test against the named assumption. Opinions you own, because that is the part AI is not allowed to hand you as truth.
AI builds the labeled draft. You judge what survives. A line with no label does not get used.
Save this and run it before you trust your next memo.
Educational content only. Not investment advice, and not a recommendation to buy, sell, or hold any security. Wall Street Prompt. Always verify against the primary source.