A bad legal research memo is almost always a bad question

How does an outsourced research team get trained on your case files?

Training happens through finished memos, not through orientation calls. Pull the last ten research memos a partner accepted without rewriting. Attach the original assignment to each one. House style, depth, citation convention, and the length a reader will tolerate are all legible from that pairing. A call can describe depth, and it cannot show it.

The assignment is the half firms never keep. Without it, a new researcher sees ten good answers and cannot see what was asked, so they imitate the format and guess at the standard. Write the refusals into the same packet: where to stop, what never goes in a memo, and what gets flagged rather than resolved. The supervision a lawyer owes over nonlawyer assistants does not change because the draft was written in another time zone, so the stopping points matter more than the permissions. A researcher who knows when to stop is cheaper to supervise than one who knows every tool in the database.

Send the accepted ones, with their assignments attached.

What happens to a half-finished memo when the researcher leaves?

A half-finished memo leaves with the researcher, because most of it was never a document. It is a folder of annotated PDFs, a search history, a list of dead ends already ruled out, and a running note of what the partner clarified on a call three weeks ago. Ask where that material physically lives, and ask it before you ask what the engagement costs. Case knowledge that lived only in one researcher's head leaves the building with that person. So does the confidentiality posture around it, if nothing was ever centralized.

Reassignment mid-matter is the same problem with a friendlier label. The second researcher inherits the memo and not the question, which means they inherit the guesses without knowing which ones were guesses. They will defend those guesses, because on the page the guesses read like settled ground. A repository fixes the documents. Only a written question fixes the reasoning. The written ask is the part of a matter that survives its author, and it takes a paragraph to write.

What does a usable legal research question have to name?

Write the ask as if the reader cannot reach you, because on a 12-hour offset they effectively cannot. "See if there is anything on preemption" is not a research question. That is a mood. The cure is not a longer assignment, because a page of background still leaves the researcher choosing between two readings alone. What removes the guesswork is naming the things a junior would have asked in the doorway, in writing, before the work starts. Four of them turn a mood into an assignment somebody can be graded against.

• The jurisdiction and the forum, named. Not the state alone: the court, and whether federal authority is in or out.
• The decision the answer feeds. A memo written for a motion reads differently from one written for a client call, and the researcher cannot infer which.
• The stop rule. How deep to go, what to do when the authority is unsettled, and the point at which they come back rather than keep reading.
• The answer you expect. Saying what you think the law says lets the researcher tell you that you are wrong, which is the entire value of the exercise.

Naming those four takes a paragraph. Skipping them costs a week of somebody's reading.