Research

Scope a multistate survey so that it ends

Turns "check every state" into a bounded project: the question rewritten so it can be answered identically everywhere, tiers that decide where you go deep, the sources per tier, the traps, and the rule for stopping.

About 30 minadvancedIn-house, Regulatory, Employment

Your prompt6,370 characters

Still to fill in: The question, Jurisdictions in scope, What the client actually does

RoleYou are an in-house lawyer who has commissioned a fifty-state survey and received back a binder nobody could use, because the question was never pinned and every state got answered in its own vocabulary. You scope before you research: you fix the question, you tier the jurisdictions by where the client actually has exposure, and you write the answer template before the first search. You would rather deliver twelve states with a documented gap than fifty states of uneven confidence.What I needScope a survey of The question across Jurisdictions in scope, sized to what the client actually does.InputsThe question as asked: The question Jurisdictions in scope: Jurisdictions in scope What the client actually does: What the client actually does Time and money: Time and money What I can search: What you can searchHow to work this1. Rewrite The question into a form that can be answered identically everywhere: a yes or no, or a value with a unit. Pin every definition the question depends on, because a word that means different things in different states destroys the comparison before the research starts. If the question as asked cannot be made answerable, say so and give the two or three narrower questions that can be. 2. Tier the jurisdictions in Jurisdictions in scope before researching any of them, using What the client actually does rather than headcount alone: tier one is where a bad answer changes what the client does, tier two is where the law is known to be an outlier or where a bad answer is survivable, tier three is everything else. Say what depth each tier gets, and produce an explicit not-researched list rather than leaving gaps unmarked. 3. Fix the answer template before the first search, and make it the same for every jurisdiction: the fields, in order, including a value for "no law on point," a value for "statute silent but agency guidance exists," and a confidence value. A matrix whose cells were written in different shapes cannot be read across. 4. Name the source of first resort for each tier and the order to work them: the statute or regulation itself, the enforcing agency's published guidance, a named survey or treatise chapter if one genuinely exists, and the practitioner source of last resort. Mark any publication you are not certain exists or is current as [UNVERIFIED - confirm it exists and check the edition]. 5. Name the traps that invalidate a survey after it is finished: home-rule cities and counties with their own rules, effective dates and grandfathering, whether federal law preempts the question in some states and not others, contractual choice-of-law clauses that make the survey irrelevant for some agreements, and the gap between what a statute says and what the agency actually enforces. 6. Give a stopping rule per tier, stated as a condition rather than a time box, and an escalation rule: what makes a tier-three jurisdiction jump to tier one in the middle of the project. 7. Produce the matrix blank. Do not fill in any jurisdiction's substantive law, and say plainly why: any rule you supply will be wrong somewhere, and a wrong cell in a survey is more dangerous than an empty one because nobody re-checks a filled cell.Ask me firstYou have the question, the jurisdiction list, the operations, and the constraints. Ask me these four, which none of that tells you, then stop and wait: 1. What decision does this feed, and would one bad jurisdiction actually change it? A survey that picks one state to operate in and a survey that papers a national policy are different projects. 2. Which jurisdictions would the business genuinely refuse to exit if the answer came back badly? That is the real tier one, and it is rarely the same as the headcount list. 3. Who has to be able to defend this work later: a regulator, a board, an auditor, or nobody? That decides how much of the negative research has to be written down as it happens. 4. Has anyone run this before, inside or out, and how stale is it? An old survey is a starting structure even when every answer in it has moved. Do not begin until I answer. If I tell you to proceed anyway, state each assumption at the top of your output and mark it [ASSUMPTION - verify].Output formatOpen with the rewritten question and the pinned definitions. Then the tiering: three named lists with the depth each gets and an explicit not-researched list. Then the answer template as a field list with its allowed values. Then the source order per tier. Then the traps, each with the check that catches it. Then the stopping rule and the escalation rule. Then the blank matrix with headers only, and one line saying why it is blank. Close with one line naming the two of my answers that most changed this scope, and how many jurisdictions would have been in tier one without them. If an answer changed nothing, say so, because it means the question was not worth a round-trip.Never do this- If the scope would fit any multistate question, it is too generic. The tiers have to come from what this client actually does and the traps from what this question actually turns on. - No hedging filler. "Varies by state" is the premise, not a finding. Do not tell me to consult an attorney; I am the attorney, and I am the one who will be asked to sign this. - Never state any jurisdiction's rule, threshold, number, or effective date, and never cite a case or statute. The matrix ships blank. Any publication you name is marked [UNVERIFIED - confirm it exists]. - Where you cannot tell whether the question is preempted or displaced in some jurisdictions, say you do not know and put it in the traps list. Do not smooth over the gap with a confident scope. - Do not pad. If the honest answer is that six jurisdictions matter and the rest are a form letter, say that. Length is not value.Before you answer- Can the rewritten question be answered with the same shape of answer in every jurisdiction? - Does the template include a value for no law on point and a confidence value? - Did I supply any substantive rule for any jurisdiction? - Is the not-researched list explicit, or did I leave gaps implied? - Would this scope be useless for a different question? It should be.

The run walks turn one, the pressure test, the follow-up, and a check on what came back. The Cockpit adds driver's-seat tunes. Your values come with either one.

Pressure-test it

Makes the AI switch hats and attack its own answer.

Your general counsel reads the finished matrix and asks the question it cannot answer: which of these cells did somebody actually research, and which did somebody assume? Work through the scope as the general counsel who has to stand behind it: name the tier where the depth is thin enough that a filled cell will imply more than was done, the trap most likely to invalidate a whole column after delivery, and the jurisdiction you would bet is wrong. Then tell me what to change in the scope so that answer is defensible instead of embarrassing.

Go deeper

Pushes the work further once the basics are right.

Matrices get built once and then go stale on a shared drive. Write the maintenance plan: which cells carry an expiry date and why, the two or three signals that would force an unscheduled refresh, who owns the watch, and the one-page cover note that goes on the front of the matrix stating what it covers, what it does not, and the date after which nobody should rely on it without checking.

4Check what came backPaste the answer here and work a checklist against this prompt's own rules.

Before you run it

What to gather first

The question as the business asked it, before you translated it
The list of jurisdictions actually in scope, and why each is on it
Where the client has employees, customers, revenue, or registered entities
The deadline and who is paying for it
Any earlier survey on this question, however stale

Watch for

The model has no reliable knowledge of any state's current rule and will produce thresholds, effective dates, and citations that are wrong. This prompt exists to scope the work, not to do it.
A filled cell never gets re-checked. An empty cell marked "not researched" is safer than a plausible guess, and this is the reason the matrix ships blank.
Home-rule cities and counties defeat state-level surveys routinely. A state column that is correct can still be wrong for the city where most of the workforce sits.
Effective dates and grandfathering are where multistate work goes wrong quietly. Confirm which version applies to agreements already signed before you rely on any answer.
Do not paste client-confidential operational data unless your company's AI policy permits it. Headcount by state is often more sensitive than it looks.

What comes back

The rewritten question with pinned definitions, three tiered jurisdiction lists with the depth each receives and an explicit not-researched list, the answer template as fields with allowed values, the source order per tier, the traps each with its catching check, the stopping and escalation rules, a headers-only blank matrix with the reason it is blank, and the answer-consumption line.

See an example of what you’ll get

*(After I answered: this feeds whether we adopt one national handbook or state supplements, we would not exit California or New York under any circumstance, the audit committee has to be able to defend it, and there is a 2022 survey from prior outside counsel.)* Rewritten question. For each jurisdiction: (a) is a pre-dispute arbitration agreement covering statutory wage claims enforceable against an employee, yes or no or unsettled; and (b) if yes, does enforceability require a signature rather than continued employment or a handbook acknowledgment, yes or no or unsettled. Pinned definitions. "Wage claim" means a claim under the jurisdiction's own wage and hour statute, not a contract claim for unpaid compensation and not a federal claim. "Acknowledgment" means an electronic or paper receipt of a handbook that does not itself contain the arbitration terms. "Employee" excludes anyone classified as an independent contractor, which is a different survey. Tiering - Tier one, full research (6): California, New York, Illinois, Washington, Texas, and the jurisdiction where the entity is organized. You said you would not exit California or New York regardless of the answer, which means those two are where a bad answer changes the document rather than the footprint. These get the statute, the agency's position, and the practitioner source, with the answer written up in full. - Tier two, confirmation only (roughly 16): the remaining states where you have employees. One source each, the primary one. If the answer is not clear from it in fifteen minutes, the jurisdiction escalates to tier one rather than getting a guess. - Tier three, not researched (the balance): no employees, no entity, no customers. These are on the not-researched list by name, not omitted. - Explicit not-researched list. Every tier-three jurisdiction listed by name in the deliverable, with the sentence: "Not researched. No operations. If headcount appears here, this question must be answered before the first hire." Answer template. Same eight fields, same order, every jurisdiction: Jurisdiction | Enforceable for statutory wage claims (yes / no / unsettled / no law on point) | Signature required (yes / no / unsettled / not reached) | Primary source consulted | Date of source | Agency position differs from statute (yes / no / not checked) | Confidence (high / medium / low) | Researcher and date. The "no law on point" value and the "not checked" value are the two that make this readable. Without them a researcher who found nothing writes a paragraph explaining that they found nothing, and the column stops being comparable. Source order - Tier one: the statute or code section itself, then the enforcing agency's own published guidance or opinion letters, then a national treatise chapter on arbitration of employment claims [UNVERIFIED - confirm the treatise exists, is current, and covers state statutory wage claims specifically rather than federal preemption], then a state bar practice guide if one exists. - Tier two: the statute alone. Nothing else. - Tier three: none. Traps | Trap | The check that catches it | |---|---| | Home-rule cities with their own wage ordinances | For every tier-one jurisdiction, check whether the largest city where you have people has its own wage ordinance and whether it addresses arbitration | | Effective dates and grandfathering | Ask, per jurisdiction, which version applies to acknowledgments already signed. Your existing workforce is on the old document | | Federal preemption operating unevenly | Preemption may resolve the question in some jurisdictions and be contested in others, which means a "yes" in two states can rest on completely different reasoning. Record the reason, not just the answer | | Choice-of-law clause | If the handbook or an agreement designates one state's law, part of this survey may be irrelevant for part of the workforce. Confirm before researching state twenty | | Statute versus enforcement | The agency's litigating position can differ from the text. The template has a field for this, and tier two never fills it, which the confidence value has to reflect | Stopping rule. Tier one stops when the statute has been read against the pinned definitions, the agency's published position has been located or its absence recorded, and one secondary source confirms nothing has moved in the last eighteen months. Tier two stops at fifteen minutes with the primary source, recording either an answer at medium confidence or an escalation. Tier three does not start. Escalation rule. A tier-three jurisdiction moves to tier one the moment headcount appears there. A tier-two jurisdiction moves to tier one if the primary source is silent, if the answer is "unsettled," or if the answer differs from the tier-one consensus, because a single outlier in a national policy is worth more attention than four states that agree. The matrix, blank. | Jurisdiction | Enforceable | Signature required | Primary source | Source date | Agency differs | Confidence | Researcher / date | |---|---|---|---|---|---|---|---| | | | | | | | | | It ships blank on purpose. Any rule I supplied here would be wrong in some jurisdiction, and a filled cell in a survey never gets re-checked by the person who reads it. Assumptions. That "employee" in your question excludes contractors [verify: if it includes them, this is two surveys]. That the 2022 prior survey asked the same question rather than a federal preemption question [verify: it changes whether it is a starting structure or a distraction]. That the audit committee wants a defensible method rather than a number [safe, given your answer]. What your answers changed. That the audit committee has to defend this is why tier two carries a hard fifteen-minute escalation rather than a best-efforts answer, and why the not-researched list is named jurisdiction by jurisdiction instead of described as "remaining states." Without that answer tier two absorbs the doubt and the matrix looks complete. That you would not exit California or New York under any circumstance is what put them in tier one ahead of Texas despite headcount, and it is why tier one has six jurisdictions rather than the nine a headcount cut would have produced. Your answer about the 2022 survey changed the source order only at the margin, so it barely earned its round-trip.

Why this prompt is built the way it is

## Framework 1. **Rewrite the question so every jurisdiction can answer it the same way.** A yes or no, or a value, with the definitions pinned. 2. **Tier the jurisdictions before researching any of them:** real exposure, known outliers, the rest. Different depth for each, and an explicit not-researched list. 3. **Fix the answer template first,** including a value for "no law on point" and a confidence value, so the answers are comparable. 4. **Name the source of first resort per tier,** in order, and mark any title you are not certain exists. 5. **Name the traps:** home rule, effective dates, federal preemption, choice-of-law clauses that moot the question, and the gap between the statute and how it is enforced. 6. **Give the stopping rule and the escalation rule.** 7. **Produce a blank matrix.** Supplying substantive law for fifty states guarantees it is wrong somewhere.