Retrieval is the easy part of research, and it is where most tools stop. You ask, they fetch, you copy the answer into a deck. But a good decision is not a lookup — it is a loop. The work that actually moves an outcome runs in four stages, and the answer is only the second of them.
1. Generate the right questions
Most bad analysis is a confident answer to the wrong question. Before retrieving anything, the system should decompose the decision into the handful of questions whose answers would actually change it — and, just as important, name the ones no document can settle. Framing is leverage: get this stage right and the rest gets dramatically easier.
2. Find and verify the answers
Now retrieval earns its keep — but with provenance attached. Each answer is grounded in a primary source where one exists, labelled as inference where it does not, and routed to a human expert where the truth lives only in someone’s experience. The output is not a paragraph; it is a set of claims you can audit, each tagged verified, inferred, or unknown.
3. Simulate the execution
Here is the stage almost everyone skips. An answer is a fact; a decision is a bet on what that fact implies. Before you act, the system should play the answer forward — model the scenarios, stress-test the assumptions, and show where the conclusion breaks. “Pricing holds at current levels” is a claim. “If pricing holds, here is the return; here is the threshold where it stops working” is a decision you can actually make.
4. Iterate as reality returns signal
The loop does not end when you decide. Reality answers back — the deal closes, the rollout stalls, the number comes in light — and that signal should feed straight into the next pass. Questions get sharper, weak inferences get replaced by what actually happened, and the expert calls you paid for once become context the system never has to buy again.
A single answer is a snapshot. A decision loop is a system that gets less wrong every time it runs.
Strung together, the four stages turn research from a one-shot lookup into a compounding asset: the right question, a sourced answer, a simulated consequence, and an honest update when the world disagrees. That is the difference between a tool that fetches and a system that thinks alongside you — and it is the loop expertOS is built to run end to end.