---
name: research-brief-interrogation
description: >
  Reads a client or stakeholder brief and extracts what is actually being asked.
  Use for "here is the brief, what do you make of it", "interrogate this brief",
  "what are we actually being asked", "the client wants a survey but I am not sure
  why", "what should I ask them before we design this", "is this brief asking us to
  prove something they have already decided". Separates statement from assumption,
  recovers the decision behind the request, finds the unstated success criteria and
  the scope boundary, and produces the blocking clarifying questions separately from
  the nice-to-know ones.
category: 01 Research Strategy and Design
ref: "01.01"
tier: 1
inherits: [K2, K3, K4, K5]
---

# Research Brief Interrogation

## 1. One-line description
Takes a brief as written and produces what it actually contains: the decision behind the request, the assumptions carried in its wording, the unstated success criteria, the scope boundary, the stakeholder map, and the questions that must be answered before design can start.

## 2. What this skill is used for

**The research problem it solves.** A brief is written by someone describing a solution, not a problem. By the time it reaches a researcher it has usually passed through two or three people, acquired a method, lost the decision it was meant to serve, and hardened a set of assumptions into apparent facts. The characteristic damage is invisible at the time: the project proceeds politely from the brief's own framing, and every downstream choice inherits an assumption nobody tested. The questions that would have exposed it were either never asked, asked too late to change anything, or buried in a list of twenty so that the three that mattered were answered last. This skill separates what the brief says from what it assumes, recovers the decision, and produces a short list of questions that must be answered before design begins and a longer list that can be carried on a stated default.

**Where it sits in the research lifecycle.** First. Before question architecture, before method, before any proposal. It is the cheapest intervention available in a research project and the one most often skipped because the brief looks clear.

**Typical use cases.**
- A brief arrives naming a method and no question.
- A brief arrives as a forwarded email chain with no author and no owner.
- A tender or procurement document must be responded to with limited access to the buyer.
- An internal stakeholder describes a study verbally and expects it to be built from the description.
- A brief reads as though it wants a particular finding, and you need to establish whether it does.
- An inherited project is going wrong and the original brief has to be re-read to find out where.

**Who uses it.** Agency and consultancy researchers receiving client briefs, client-side insight managers receiving internal requests, UX and product researchers taking intake from a product team, evaluation and policy analysts receiving commissioning documents, and research directors triaging incoming work. Useful to a beginner as a structure for the awkward conversation, and to a director as a fast, documented basis for pushing back.

## 3. When to use it

- A brief, request or commissioning document has arrived and design has not started.
- You can restate the method the brief wants but not the decision it serves.
- The brief contains numbers, claims or characterisations of customers that nobody has sourced.
- Several people have contributed to the brief and it is unclear whose question it is.
- The brief is short and confident, which is more often a warning than a convenience.
- The brief is long and comprehensive, and you suspect it lists everything anyone has ever wanted to know.
- A project is already running badly and you need to establish what was actually agreed.
- You are about to write a proposal, and the proposal will inherit every ambiguity you do not resolve now.

## 4. When NOT to use it

- **When there is no brief and no requester to interrogate.** Interrogating a document nobody wrote produces invented intent. If all that exists is a topic and a person who will not engage, say so: the output is a statement that no brief exists, plus the minimum set of questions that would create one. Do not construct a plausible brief and proceed from it.
- **When the brief has already been interrogated and what is now needed is the question set.** This skill stops at a clarified problem with its open questions listed. Turning that into a decision, a business question, a research question and sub-questions belongs to **01.02 Business Problem to Research Question**.
- **When the method is the open question.** If the problem is clear and the argument is qual versus quant, or whether a survey fits, go to **01.04 Research Method Selection**. This skill records that a method was pre-specified and tests whether a question sits behind it; it does not choose the replacement.
- **When the output required is a commercial proposal.** Interrogate first, then use **01.08 Research Proposal and Scope Development**. A proposal written before interrogation prices an ambiguity.
- **When the real question is whether the research is lawful or ethical to run.** Sensitive topics, vulnerable participants, secondary use of personal data and identifiability in small bases are not brief-clarification matters. Route to **13.05 Research Ethics and Consent Design** in parallel, and do not let a clean brief substitute for an ethics review.
- **When access to the buyer is closed and the interrogation would have to be conducted by assumption.** In a competitive tender with a formal questions window, the correct output is a list of ambiguities and the assumption you will state against each in the response. Resolving them privately and silently is how a supplier wins work it cannot deliver.
- **When the exercise is being used to reopen a decision that has been taken and communicated.** This skill can name a ratification brief for what it is, which is a legitimate and important output. It cannot reverse an organisational decision, and presenting it as though it could wastes the researcher's credibility on the one occasion it is most needed.
- **When the brief is genuinely clear, the decision is named, the constraints are stated and the questions you have are curiosity.** Interrogation has a cost in goodwill and elapsed time. Spending it where nothing is at stake teaches the client that clarification is a ritual, which makes it easier to ignore when it matters.

## 5. Required inputs

**Required.**
- **The brief exactly as written**, in the requester's own words. Not a summary, not a retyped version, not a colleague's account of it. The wording is the evidence: an assumption survives in a noun phrase and disappears in a paraphrase.
- **Provenance.** Who wrote it, who asked them to write it, when, and in response to what. A brief with unknown authorship can still be interrogated, but the interrogation must record that the author is unknown, because half the useful questions have no addressee.
- **Access to at least one person who can answer questions**, or an explicit statement that access is closed. This determines whether the output is a set of questions or a set of stated assumptions. If neither is available, stop and say so.

**Optional, and what each one adds.**
- **The email or meeting thread the brief came out of.** Frequently contains the decision, the deadline and the real sponsor, all of which were tidied out of the formal document.
- **Previous briefs from the same requester.** Reveal habitual framings and recurrent method preferences, and let you tell a fresh assumption from a house one.
- **Previous research on the topic, with dates.** Lets you test whether the brief is asking for something already answered, which is the fastest available scope reduction.
- **An inventory of the organisation's own operational data.** Converts "we need to understand how often customers do X" from a research need into an internal query, before anyone has designed anything.
- **The stakeholder list with roles.** Turns the stakeholder map from inference into fact, and identifies the blocker who is not on the brief.
- **The budget order of magnitude and the decision date.** Converts vague ambition into a testable scope constraint. Without them, the scope conversation happens later and more expensively.

## 6. Questions to ask before starting

1. **Who wrote this, and who asked them to?** Why it matters: the author is often not the sponsor, and the sponsor's question is the one that must be answered. Default if unanswered: treat the author as the requester, record the sponsor as unknown, and flag every priority judgement as provisional.
2. **What decision does this feed, and when is it made?** Why it matters: it is the anchor for scope, priority and deadline simultaneously. Default: mark the decision as not established, proceed as exploratory, and put this at the top of the blocking list.
3. **What happens if we do nothing?** Why it matters: the default action reveals both the real stakes and whether any finding would change them. Default: assume the organisation proceeds on its current plan, and record that assumption where it bites.
4. **Is any part of this already decided?** Why it matters: it distinguishes a design exercise from a ratification exercise, and the two need different responses. Default: assume open, but record any expected finding stated in the brief as information about the stakeholder rather than about the world (K4 §4.2).
5. **What is explicitly out of scope?** Why it matters: an unstated exclusion is a scope dispute waiting to happen, and the exclusions are what make the inclusions affordable. Default: draft the exclusion list yourself and put it in the restatement for confirmation.
6. **Who else has to agree with the answer for it to be acted on?** Why it matters: an audience nobody named sets a defensibility standard nobody designed for. Default: assume the finding will be challenged by someone who prefers the status quo, and say you have assumed it.
7. **What has already been tried, and what happened?** Why it matters: it prevents a recommendation that the organisation has already implemented and abandoned, which is the fastest way to lose a room. Default: ask at the debrief stage if not now, and record the gap.

## 7. Step-by-step methodology

**Step 1. Preserve the brief and record its provenance.**
Copy the brief verbatim into the working document before touching it, and record who wrote it, who commissioned it, the date, and what prompted it. Every later observation refers to a line in that preserved text. A correct result is a working document in which any claim about the brief can be checked against the brief's own words. Never interrogate a summary, including your own: the summary has already resolved the ambiguities you are looking for.

**Step 2. Split every line into three registers.**
Read the brief line by line and tag each substantive statement as **fact**, **assumption** or **preference**. A fact is checkable against a source that exists (a figure from the organisation's records, a dated finding, a published statistic). An assumption is a statement about the world presented as settled but not sourced ("customers find the process confusing", "the younger segment is price-driven"). A preference is a statement about how the requester wants the work done or what they expect it to show ("a survey of 1,000", "we expect regional differences"). This pass is the core mechanic of the skill and it is done exhaustively, not selectively. A correct result is a table in which most lines are assumptions, because that is what briefs are made of, and in which every assumption carries a further column: *if this is wrong, what changes?* Assumptions whose falsity would change the design are the ones to verify; the rest can be carried and stated.

**Step 3. Recover the decision behind the request.**
Take the request and ask "and then what?" repeatedly until an action by a named role appears, or until the chain runs out. "We want to understand why churn rose" becomes "so we can decide where retention spend goes next year" becomes "which the retention lead decides in October". Write it in one sentence: actor, action, date, and what is at stake. A correct result is either that sentence, endorsed by someone who owns it, or an explicit statement that no decision was found. Never fill the slot by inference. An invented decision is the highest-damage output this skill can produce, because everything downstream will be optimised for a fiction that reads plausibly.

**Step 4. Test for the method-before-question brief.**
This is the commonest form a brief takes, and it needs its own procedure rather than a complaint. Strike the method from the brief and re-read what remains. Three outcomes. If a question survives the deletion, the method was a preference and can be tested on its merits later. If nothing survives, no question was ever written, and the brief is a request for an artefact rather than an answer. If a different question survives than the one you expected, the method was doing the work of narrowing an unstated ambition. Then run the reverse test: what question would this method genuinely answer well, and is that the question the decision needs? A correct result names which of the three cases applies, in writing, plus the question the requested method would actually answer. Do not argue the method here. Record the finding and hand it to 01.04.

**Step 5. Extract the unstated success criteria.**
Ask what would make this a good piece of research in the requester's eyes, and then find the answer in the brief rather than waiting to be told. Three reliable sources. The output format they name (a board paper implies defensibility; a workshop implies material people can argue with; a dashboard implies repeatability). The audience they name, which sets the evidential standard. And the verbs of purpose: "show", "prove", "demonstrate", "convince", "build the case for" all describe an argument to be won rather than a question to be answered, and each one is worth quoting back. Then write the anti-criterion: what would make the requester regret commissioning this? The answer is usually more informative than the stated objectives, and it frequently names a subgroup, a stakeholder or a level of certainty that appears nowhere in the document.

**Step 6. Map the stakeholders and establish whose question this is.**
Six roles, and one person can hold several: the author of the brief, the sponsor who wanted it, the decision owner who will act, the budget holder, the audience who must be convinced, and the blocker who can stop the finding being used. Fill each slot with a name or with "unknown". Then apply the ownership test: whose behaviour changes depending on the answer? That person's question is the study's question, and where it differs from the author's, the difference is the single most important thing this skill will produce. Unknowns are not tidied away; an unfilled decision-owner slot is a blocking question.

**Step 7. Draw the scope boundary in three parts.**
Not two. Write what is explicitly in, what is explicitly out, and the third and most useful category: what is silently assumed in. Silent inclusions are the markets nobody mentioned excluding, the segment everyone pictures when they say "customers", the language the fieldwork will not be run in, the channel that is not covered, and the time period nobody bounded. Every silent inclusion becomes either an explicit inclusion with a cost or an explicit exclusion in the restatement. A correct result is an exclusion list long enough to be uncomfortable, because scope disputes are always about the things nobody wrote down.

**Step 8. Test for a brief written to confirm a decision already taken.**
Six signals, checked deliberately: the expected finding appears in the brief; the requested output is a persuasion artefact (a one-page proof point, a statistic for a launch) rather than an answer; the timing sits after the decision meeting rather than before it; the sample or method is specified in a way that can only produce a favourable reading; there is no stated plan for what happens if the result is negative; and the language of purpose is argumentative rather than investigative (Step 5). Then apply the null test, directly, to the requester: *if the study finds the opposite, what will you do?* Three answers and three responses. If the answer names a changed action, the brief is live and proceeds. If the answer is that the decision stands regardless, the correct output is a written statement naming what is being asked for, what evidence the decision already rests on, and the offer of the two honest alternatives: a genuine test, or no study. If the answer is that the finding would not be published, that is the same case with a reputational risk attached, and it is escalated rather than absorbed. This judgement is a K5 §2.4 matter and does not belong to the AI.

**Step 9. Convert every constraint into what it eliminates.**
Budget order, decision date, audience access, language coverage, internal analysis capability, approval requirements and any regulatory restriction. Write each one as an elimination rather than as a number: "the October decision date eliminates any design requiring recruitment of a low-incidence audience plus a translation cycle", "no access to customer contact details eliminates any sample drawn from the organisation's own base". A correct result is a short list of designs that are already impossible, produced before anybody has proposed one. This is the most efficient minute in the whole exercise.

**Step 10. Split the questions into blocking and non-blocking.**
Every open question goes into one of two lists, and the test is precise: **would two different answers produce two different designs?** If yes, it is blocking, and design does not start until it is answered. If no, it is non-blocking, and it travels with a stated default. Blocking lists are short, typically three to six items, and they are ordered by what they unlock. Non-blocking questions each carry the assumption you will proceed on and the point at which that assumption bites. A correct result is a list a busy stakeholder will actually answer, which is what separates this from the twenty-question document that gets skimmed. Where the blocking questions cannot be answered before a deadline, say which design decisions are therefore being made on assumption, and mark them for confirmation later.

**Step 11. Write the restatement and send it back.**
One page, in your own words: what we understand you to be deciding, what we understand you to be asking, what is in and out, what we are assuming, what we still need to know. The restatement is the deliverable. Agreement on it is the foundation the project is built on, and disagreement with it is worth more than agreement, because it surfaces now what would otherwise surface at the debrief. A correct result is a document the requester either endorses or corrects, and either outcome closes the interrogation.

## 8. Analytical framework

    Brief as written
      → Register split          (fact / assumption / preference, line by line)
        → Decision recovered    (actor, action, date, stake) or "not established"
          → Success criteria    (stated, inferred, and the anti-criterion)
            → Stakeholder map   (author, sponsor, decision owner, budget, audience, blocker)
              → Boundary        (in / out / silently assumed in)
                → Constraints as eliminations
                  → Questions   (blocking / non-blocking with defaults)
                    → Restatement

Two tests run across the framework and do most of its work.

**The substitution test** (Step 4). Remove the method from the brief and see what remains. It separates a brief that contains a question from a brief that contains only a request for an artefact, and it does so without arguing about the method.

**The null test** (Step 8). Ask what the requester would do if the finding were the opposite. It distinguishes a live question from a ratification exercise more reliably than any amount of reading between the lines, and it can be asked in one sentence in a meeting.

Apply the framework in order. The register split comes before the decision recovery, because an assumption promoted to fact will otherwise be used as evidence that the decision is already settled.

## 9. Output format

A **Brief Interrogation Note**, in this order.

**1. Provenance.** Author, sponsor, date, what prompted the brief, and whether access to a respondent is open or closed.

**2. Restated request.** One paragraph in your own words, written to be corrected.

**3. Decision statement.** Actor, action, date, stake. Or the words *decision not established*, followed by what would establish it. Never inferred.

**4. Assumption register.**

| Line quoted from the brief | Register (fact / assumption / preference) | If this is wrong, what changes | Verified? | Verify by |

**5. Method-before-question finding.** Which of the three Step 4 cases applies, the question the requested method would actually answer, and whether it matches the decision.

**6. Success criteria.** Stated, inferred (with the evidence in the brief that supports the inference), and the anti-criterion.

**7. Stakeholder map.**

| Role | Name | What they want to know | What changes for them depending on the answer |

Roles: author, sponsor, decision owner, budget holder, audience, blocker. Unfilled slots stay visible.

**8. Scope boundary.**

| In scope | Out of scope | Silently assumed in (needs a decision) |

**9. Constraints and what each rules out.**

**10. Confirmation-seeking assessment.** Which of the six signals are present, the answer to the null test, and the recommendation. Where the brief is a ratification exercise, this section carries a `RESEARCHER DECISION REQUIRED` marker per K5 §2.4.

**11. Blocking questions.** Three to six, each with what it unlocks and what would change depending on the answer.

**12. Non-blocking questions**, each with the default that will be assumed if unanswered and the point in the design at which the assumption bites.

**When the evidence is thin.** Sections stay empty and say so. If no decision was found, the note says *not established* and the project proceeds explicitly as exploratory. If the author is unknown, the provenance section says so and every question that has no addressee is marked unaddressable. If the brief contains figures with no source, they appear in the assumption register as unverified rather than being repeated as context. Format is not evidence (K4 §1).

## 10. Quality checks

Run before the note is sent. These sit on top of K4 §8.

1. Is every observation about the brief traceable to a quoted line of the brief?
2. Has every substantive line been assigned a register, or only the interesting ones?
3. Does every assumption carry a consequence statement, so a reader can see which ones matter?
4. Is the decision statement either sourced from a person or explicitly marked as not established?
5. Has the substitution test been run, and is its result recorded even where the brief did contain a question?
6. Are the success criteria distinguished into stated and inferred, with the inference shown?
7. Does the stakeholder map contain at least one name for decision owner, or an explicit unknown flagged as blocking?
8. Does the scope table contain a silently-assumed-in column with entries in it? An empty one means the step was skipped.
9. Is every constraint written as an elimination rather than as a number?
10. Were all six confirmation signals checked, with the result of each recorded?
11. Would two different answers to each blocking question genuinely produce two different designs?
12. Does every non-blocking question carry the default that will be used instead?
13. Are any figures from the brief repeated anywhere in the note as though they were established?
14. Is the restatement written so that a requester could disagree with it specifically, rather than nod at it?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **Polite acceptance** | The note restates the brief in better prose and adds nothing | The register split at Step 2 is mandatory and exhaustive; a note with no assumptions listed has not been done |
| **The twenty-question document** | Every ambiguity converted into a question, none prioritised | Step 10's test: only questions that change the design are blocking |
| **Interrogating the summary** | Observations that cannot be traced to a quoted line | Preserve the brief verbatim at Step 1 and quote from it |
| **The invented decision** | A tidy decision statement nobody said, phrased with suspicious precision | Decisions come from people. Otherwise *not established* |
| **Method argument too early** | The note becomes a case against the survey | Record the method finding, hand the argument to 01.04 |
| **Missing sponsor** | The author's priorities drive the design and the sponsor's question is never answered | The ownership test at Step 6, applied by name |
| **Scope by omission** | Exclusions never written down, disputed at the debrief | The three-column boundary table, confirmed in the restatement |
| **Ratification absorbed** | A brief that wants a conclusion is quietly designed around | The null test asked out loud, and the answer recorded |
| **Constraint as aspiration** | "Tight timeline" recorded as a comment rather than as a rule-out | Step 9 converts every constraint into an elimination |
| **AI: coherence resolution** | Ambiguities in the brief silently resolved toward the reading that makes the brief make sense | Ambiguity is the finding. It goes in the register, not into a smoothed restatement |
| **AI: answering its own questions** | The clarifying questions list contains items already answered by an assumption the AI made two paragraphs earlier | Every default in the non-blocking list is labelled as an assumption, not as knowledge |
| **AI: brief assertions treated as evidence** | The brief's characterisation of customers reappears in the note as background | Everything unsourced is an assumption, including the parts that are probably true |
| **AI: inferred stakeholder names** | Plausible role holders appear who were never mentioned | Unknown is a permitted value and the correct one |

## 12. AI guardrails

Skill-specific. K4 applies in full and is not repeated here.

1. **Never infer a decision, a decision owner or a deadline.** If a person did not state it, the field reads *not established*. Plausibility is not sourcing, and a fluent decision statement is the most dangerous single sentence this skill can produce.
2. **Never promote a brief's assertion to a fact.** Statements about customers, markets, causes and competitors that arrive in a brief are assumptions until sourced, however reasonable they sound and however senior the author.
3. **Never repeat a figure from the brief without marking it unverified**, and never carry one into a downstream document as though the interrogation had confirmed it (K2 §6).
4. **Never resolve an ambiguity to make the brief coherent.** Where two readings are possible, both are recorded and the choice between them becomes a question. Coherence-seeking is the characteristic AI failure in this task.
5. **Never soften a confirmation-seeking finding.** If the six signals are present, they are listed. Whether to challenge the client is a human judgement (K5 §2.4), but the AI does not make that judgement easier by leaving the evidence out.
6. **Never generate clarifying questions to fill a section.** A question that would not change the design is noise, and noise is what causes the blocking questions to go unanswered.
7. **Never invent stakeholder names, roles or motives.** An empty slot in the map is information; a filled one that nobody confirmed is a fabrication with a plausible face.
8. **Never treat the absence of an exclusion as an inclusion, or the reverse.** Silence is recorded as silence, in the third column, and converted into a decision by a person.

**Human review points** (see K5 §2 for the classes):
- **Researcher decision required** where the confirmation-seeking assessment finds a ratification brief. Class 2.4, ethical appropriateness: whether and how to challenge a client is a professional judgement with relationship and liability consequences the AI does not carry.
- **Researcher decision required** on which questions are blocking, where answering them will cost elapsed time against a fixed decision date. Class 2.7.
- **Researcher review recommended** on the stakeholder map before it is shared, since it names people and their motives and may be politically readable in ways the AI cannot anticipate.

## 13. Best-practice principles

1. **The brief is evidence, not instruction.** Read it the way you would read a transcript: for what it reveals as much as for what it says. The word choices, the order of the objectives and the things left out are all data about the organisation.
2. **The first noun phrase is usually the assumption.** "Why customers are abandoning the portal" contains an unproven abandonment. "The barriers to adoption" contains unproven barriers. Underline the nouns before anything else.
3. **A method in a brief is a symptom, not a preference.** Somebody chose it for a reason, and the reason (a previous study, a board member's expectation, a budget line already coded) is more useful to know than the method is to argue with.
4. **Ask the null question early and once.** "If it comes back the other way, what happens?" takes ten seconds and settles more about a project than a page of objectives.
5. **Exclusions are a service, not a retreat.** Clients experience a written out-of-scope list as competence, because it demonstrates that the in-scope list was chosen rather than accepted.
6. **The person who wrote the brief is rarely the person who needs the answer.** Find the second one before design starts, even if it is awkward, because they will appear at the debrief regardless.
7. **Blocking questions must be few enough to be answered in one reply.** Three good questions get answered. Fifteen get delegated to someone junior and answered badly.
8. **Every unanswered question becomes a stated assumption, never a silent one.** Silence at this stage reappears as a dispute at delivery, and by then it has a cost attached.
9. **Interrogation is not interrogation of the person.** The tone that works is curiosity about the problem, and the output that works is a restatement they can improve rather than a critique they must defend.
10. **The strongest signal in any brief is what it does not ask.** A study of leavers with no interest in stayers, a service brief with no mention of staff, a pricing brief with no mention of competitors: absences are where the framing lives.
11. **Restate before you propose.** A proposal built on an unconfirmed understanding is a proposal that will be judged against a different understanding.
12. **Recording that a brief is ratification-shaped is a success, not a failure.** The document has done its job even if the project does not proceed, and it is the only artefact that will protect the researcher's position later.

## 14. Worked example

Fictional scenario, international NGO. All figures are illustrative and belong to the scenario.

    INPUT

    A one-page brief from the programmes director of a fictional water and
    sanitation NGO, forwarded by a programme manager:

    "We need a survey of 800 beneficiaries across our three districts to
    demonstrate the impact of the household water point programme, for the
    annual donor report in November. Please include satisfaction with the
    water points and the health benefits people have experienced. We know
    the programme is working, we just need the evidence."

**Process.**

*Step 2, register split.* Eleven substantive lines. Two facts (three districts, November report date). One preference stated as a requirement (a survey of 800). Eight assumptions, of which three would change the design if wrong: that the intended respondents are reachable and identifiable as programme beneficiaries; that health benefits are attributable to the programme and can be reported by respondents; and that "the programme is working" is established rather than assumed.

*Step 3, decision recovery.* Asking "and then what" produced no action. The donor report is a reporting obligation, not a decision. Pushing further reached a second, unstated decision: the programme's next funding cycle is agreed in February, and a decision exists about whether to extend the model to two further districts. That decision has an owner and a date. The stated purpose and the real decision are different, and they need different evidence.

*Step 4, substitution test.* Striking "a survey of 800 beneficiaries" left "demonstrate the impact of the programme", which is a purpose, not a question. Case two: no question was ever written. The reverse test asked what a beneficiary survey would genuinely answer well, and the answer is satisfaction, use patterns and self-reported change, which is useful and is not impact.

*Step 5 and Step 8, the judgement call.* The verbs are argumentative throughout: "demonstrate", "we just need the evidence". Four of the six confirmation signals are present. The null test was asked of the programmes director: if the survey found no self-reported health improvement, what would happen? The answer was that the report would use the satisfaction findings instead. That is a ratification brief on the reporting purpose. It is not a ratification brief on the February extension decision, which is genuinely open. The interrogation therefore split into two: the donor-report request, which was named as a communications need with a documented limitation on what a beneficiary survey can attribute; and the extension question, which is a live decision and can carry a real study.

*Step 7, boundary.* Silently assumed in: households that stopped using the water points, which the beneficiary list may no longer contain; the two districts under consideration for extension, which the brief does not cover at all; and non-beneficiary households in the same communities, without whom no comparison exists. The third of these is the difference between a satisfaction study and anything that could support an attribution claim.

*Step 10, blocking questions.* Four. Who owns the February extension decision and what are the options? Does the beneficiary list include households that have lapsed, and can they be contacted? Is there any pre-programme measurement or a comparison community, and if not, is the attribution claim to be dropped from the donor report? What ethical approval and consent arrangement covers surveying these households, given that some are receiving ongoing assistance (routed to 13.05)?

    OUTPUT

    A Brief Interrogation Note with: the two purposes separated and named;
    a decision statement for the February extension and an explicit "no
    decision" for the donor report; an eleven-line assumption register with
    three verification actions; a finding that the brief specifies a method
    and no question, with the question the method would actually answer;
    a confirmation-seeking assessment carrying a researcher decision marker
    on the attribution language; a three-column scope table with three
    silent inclusions; four blocking questions; six non-blocking questions
    with stated defaults; and a one-page restatement sent to the programmes
    director for correction.

## 15. Advanced usage

**Closed-access tenders.** Where a formal questions window exists, use it for the blocking list only, phrased so that the answers are usable by any bidder (procurement rules usually circulate answers to all). Everything non-blocking becomes a stated assumption in the response, in a visible assumptions register rather than in a footnote. A bidder who states assumptions clearly is easier to compare and harder to blame.

**Multi-author briefs.** Where a brief has been assembled from several contributors, the register split will show internal contradictions, usually between an objective and a constraint. Do not reconcile them. Present the contradiction to the requester as a choice, because reconciling it privately means choosing between stakeholders without authority.

**Re-interrogating mid-project.** When a project is going wrong, run this skill on the original brief and compare the result with what is being built. The gap is almost always one of three things: a silent inclusion that became work, a blocking question that was answered by assumption, or a sponsor who was never in the stakeholder map. Naming which one, with the original document as evidence, converts a difficult conversation into a factual one.

**The verbal brief.** Where the brief arrives as a meeting, write it up and send it back the same day, with the register split already applied and the assumptions marked as such. The write-up becomes the brief, and the act of correcting it is the interrogation.

**Procurement-written briefs.** A brief written by a procurement function to a template will have a specification where the question should be, and the specification will be precise about deliverables and vague about purpose. Interrogate for the internal client behind procurement, and expect the decision to be recoverable only from them.

**Running it as a triage.** Directors reviewing incoming work can run Steps 3, 4 and 8 alone, in a few minutes each, to sort briefs into three piles: live decisions worth designing for, artefact requests that need a scope conversation, and ratification exercises that need a different conversation entirely.

## 16. Skill chain

**Recommended previous skills**
- None. This is the library's entry point for incoming work. Where an inherited project is being reviewed rather than started, **13.01 Research Quality Review** can run first and hand over the specific defect to be traced back to the brief.

**Recommended next skills**
- **01.02 Business Problem to Research Question.** Receives the clarified problem, the decision, the assumption register, the stakeholder map and the scope boundary, and turns them into a defensible question set.
- **01.08 Research Proposal and Scope Development.** Receives the boundary and the assumptions where the output is a proposal to an external buyer.

**Runs well alongside**
- **01.04 Research Method Selection**, which receives the finding that a method was pre-specified and tests it on the merits.
- **13.05 Research Ethics and Consent Design**, wherever the interrogation surfaces vulnerable participants, sensitive topics or secondary use of personal data.
- **10.01 Literature Review and Desk Research**, for testing whether the brief's assumptions are already answered.

---
A Yazi Supplied Skill and resource.