---
name: literature-review-and-desk-research
description: >
  Finds, appraises and synthesises existing evidence before or alongside primary
  research. Use for "what is already known about", "do a literature review",
  "desk research", "secondary research", "review the existing evidence",
  "is there published data on", "do we need primary research or can we answer
  this from what exists", "find sources on", "these two reports disagree".
category: 10 Desk Research and Evidence Synthesis
ref: "10.01"
tier: 0
inherits: [K2, K3, K4, K5]
---

# Literature Review and Desk Research

## 1. One-line description

A disciplined method for finding, verifying, quality-appraising and synthesising existing evidence, so that a research question is answered from what already exists where possible, and primary research is commissioned only where it is genuinely needed.

## 2. What this skill is used for

**The research problem it solves.** Most research questions have been partly answered already, by someone, somewhere. Without a method, desk research fails predictably. It becomes a collection of whatever surfaced first, weighted by findability rather than quality. It flattens sources of wildly different standing into an undifferentiated "the research says". And with AI assistance it acquires a further failure worse than the others combined: confident, well-formatted citations to sources that do not exist. This skill installs a search strategy, a source-quality framework, a verification gate no unverified source passes, and a synthesis method that handles disagreement rather than averaging it away.

**Where it sits in the research lifecycle.** Usually first, before design, where it establishes what is known, what is contested, and what the primary study actually needs to add. It also runs alongside analysis, to contextualise findings, and after it, to check whether a result is consistent with the wider evidence base.

**Typical use cases.**
- Sizing and characterising a market or a population before deciding whether to field a study.
- Establishing what is already known so a primary study is not commissioned to re-discover it.
- Building the context and background section of a report or proposal.
- Adjudicating between two published figures that disagree.
- Assembling a defensible evidence base for a business case where no primary budget exists.
- Establishing prior art and existing constructs before designing an instrument.
- Producing the narrative literature review that sits behind a commercial or applied academic piece of work.

**Who uses it.** Research managers and directors scoping a project; strategists and analysts working without a fieldwork budget; consultants building a point of view; postgraduate and applied academic researchers writing a narrative review; insight teams asked a question on a timeline too short for primary research.

## 3. When to use it

- The question is "what is already known about X", in whole or in part.
- A stakeholder wants a number, a trend or a market context and there may be a published source for it.
- A primary study is being scoped and you need to know what it should not spend money re-establishing.
- Two or more sources circulating in the business disagree, and someone needs to decide which to believe.
- Budget or timeline rules out primary research and the question still has to be answered as well as it honestly can be.
- A report needs a context section that is more than assertion.
- You are being asked to validate or contextualise a primary finding against external evidence.
- You need to establish whether a construct, scale or segmentation already exists before building one.

## 4. When NOT to use it

- **The question is about your own customers, product or service specifically.** No amount of published evidence tells you what your customers think of your onboarding flow. Desk research can tell you what is generally true of onboarding; it cannot substitute for asking them. Go to **01.04 Research Method Selection**, then to primary design.
- **The question concerns something new, proprietary or not yet in the world.** A concept that has not launched, a price point never tested, a message never run. There is nothing to find. Presenting adjacent evidence as though it answers the question is worse than saying it cannot be answered.
- **The decision is high-stakes and the available evidence is thin.** Where a material investment rests on it, a synthesis of three weak secondary sources is not a stronger basis than an honest "we do not know". It is a weaker one, because it looks like knowledge. Report the gap and specify the primary research that would close it. `RESEARCHER DECISION REQUIRED` (K5 §2.5).
- **The evidence base is stale relative to the thing being studied.** In fast-moving categories, behaviour reported four years ago may describe a world that no longer exists. Where every available source predates a structural change (a regulatory shift, a technology change, a major market event), the correct output is a dated picture explicitly labelled as historical, not a current answer.
- **The output needs to establish causality.** Desk evidence can assemble a case and identify what has been tested, but a synthesis of observational sources does not license a causal claim (K4 §3.2). If the decision needs to know whether X causes Y, you need a design that can answer that. See **06.04 Experiment and A/B Test Analysis**.
- **Formal academic rigour is required.** A narrative review is not a systematic review. Where the work will be examined, peer reviewed, or must be reproducible by a third party, the search must be exhaustive, pre-specified, and reported to protocol standard. Hand over to **15.06 Systematic Literature Review**, or **15.14 Advanced Systematic Review and Meta-Analysis** where quantitative pooling is intended. This skill's search is purposive and documented; it is not exhaustive, and it must never be described as systematic.
- **The task is verifying citations in an existing document rather than building an evidence base.** That is **13.02 Source and Citation Verification**.
- **No source can be accessed.** If a body of evidence is known to exist but sits behind access you do not have, do not reconstruct it from abstracts, summaries or memory. State what exists, state that it could not be read, and stop. K4 §6.4.

## 5. Required inputs

**Required.**
- **The decision or research question the review has to serve.** Not a topic. "Beverage packaging" is a topic; "should we assume consumers will accept a refill format at parity price" is a question. If only a topic is supplied, ask. A review without a question has no stopping rule and no relevance test.
- **The scope boundary**: markets or geographies, time window, and population of interest. If absent, propose one explicitly and state it as an assumption in the output.
- **Access to at least one real source that can actually be read.** Not a search result snippet, not a summary of a summary. If nothing can be opened, stop and say so.

**Optional, and what each one adds.**
- **A date cutoff or recency requirement.** Lets you exclude on age at screening rather than discovering the whole evidence base is stale at synthesis.
- **Named sources the stakeholder already believes.** Reveals the prior the review will be read against, and lets you appraise those sources explicitly rather than appearing to ignore them.
- **The intended audience and format.** Determines citation density and whether the output carries an evidence table or inline references.
- **An existing internal evidence base (past studies, client data).** Lets the review distinguish what is already known internally from what must come from outside, and prevents commissioning work that has been done before.
- **A definition of the key construct.** Without it, sources measuring different things get pooled. With it, definitional mismatch becomes a screening criterion rather than a synthesis error.
- **Budget and timeline for possible primary research.** Turns the sufficiency judgement in Step 10 from abstract into decidable.

## 6. Questions to ask before starting

1. **What decision does this review feed, and who makes it?** Determines the relevance threshold and the depth required. Default if unanswered: assume an internal decision of moderate consequence, and flag every conclusion that would change with the stakes.
2. **Which markets and what time window count as in scope?** Determines what gets excluded on geography and recency. Default: the client's home market, sources from the last five years, both stated explicitly in the output as an assumption.
3. **Is a directional answer acceptable, or does this need a defensible number?** A directional answer can rest on convergent weak sources. A defensible number cannot. Default: assume directional, and mark clearly where a number is being asked to carry more weight than its source supports.
4. **Is primary research a live option?** Changes the purpose of the review from "answer the question" to "answer what it can and specify the rest". Default: assume it is, and produce the gap statement either way.
5. **Are there sources the organisation is already committed to or already citing?** These must be appraised, not quietly bypassed. Default: ask at the outset, because discovering them late forces a rewrite.
6. **What would make this review wrong in six months?** Identifies the volatile assumptions. Default: state the shelf life of the conclusion explicitly.
7. **What language, and does that limit the evidence base?** A review conducted in one language in a multilingual market has a coverage problem (K3 §6). Default: state the language limitation.

## 7. Step-by-step methodology

**Step 1. Convert the decision into answerable review questions, and set a stopping rule.**
Break the decision into three to six specific questions that evidence could settle. Each must name a population, a construct and, where relevant, a period. "Do consumers care about repairability" is not answerable. "Among domestic appliance buyers in this market since 2021, what proportion report repairability influencing choice, and does stated importance track observed purchasing?" is. Then write the stopping rule before searching: the review stops when each question has either converging evidence from at least two independently produced sources of adequate quality, or a documented finding that no adequate source exists. Without a stopping rule, desk research expands until the deadline. *Correct result: a numbered question list, each with the evidence that would answer it named in advance.*

**Step 2. Map the evidence landscape before you search.**
For each question, list the *types* of source that would carry the answer if it exists: peer-reviewed academic work, official statistics, regulator and standards-body output, industry and trade association reporting, published commercial research, company disclosures and filings, trade press, non-governmental research bodies, and grey literature such as conference and working papers. Mapping first prevents the commonest structural failure in desk research, a review built entirely from whatever a general web search surfaced, over-weighted toward trade press because trade press is optimised to be found. *Correct result: a short table of question against expected source types, with the accessible ones marked.*

**Step 3. Build the search strategy and open the search log.**
For each question, decompose into concepts, then list synonyms, near-terms, industry jargon and academic terminology for each. Terms differ by community: the same behaviour may be "churn" commercially and "attrition" academically, and searching one vocabulary silently excludes half the evidence. Combine concepts with AND, synonyms with OR, and use exclusion terms only where a term is genuinely ambiguous. Adapt syntax to each source type: controlled vocabulary and field-limited search in academic databases, table and dataset navigation in statistical sources, publication indexes for regulators and associations. Open a search log immediately and record for every search: the source searched, the exact string, the date, the hits, and how many were taken forward. The log is what makes the review checkable and what turns "we could not find anything" into a defensible statement rather than an admission. *Correct result: a search log with at least one row per source consulted, including the searches that returned nothing.*

**Step 4. Screen in two passes.**
First pass on title, abstract or executive summary against explicit inclusion criteria: population, geography, period, construct match, source type. Second pass on the full source, where most exclusions actually happen, because a title promising a finding routinely conceals a different population or a definitional mismatch. Record the reason for every exclusion in one of a small set of standard categories (wrong population, out of period, construct mismatch, restatement of a source already included, could not access, insufficient method reporting). Categorical reasons let you see the shape of the exclusions, and expose the case where you have excluded most of the inconvenient evidence on grounds that will not survive scrutiny. *Correct result: counts at each stage (found, screened, included, excluded by reason), and an included set you have read in full.*

**Step 5. Verify before any source is allowed to carry weight.**
This is the gate, and nothing passes it on plausibility. For every source: confirm you have opened it, take the locating details from the source itself rather than from a citing document, and confirm the claim you intend to draw appears in it and means what you think it means. Then trace provenance. A trade article reporting "a study found 62%" is not the study; find the study. Where the original cannot be found, the claim is attributed to the intermediary and labelled an unverified restatement, or dropped. Watch for citation chains where several sources trace to one original, creating a false impression of convergence: this is the most common way desk research overstates its evidence base. Count the original once, not the outlets that repeated it. *Correct result: every included source carries complete locating details, a note of what it actually claims, and a provenance marking of primary, secondary or unverified.*

**Step 6. Appraise quality on eight dimensions.**
Score every included source. Do not average the dimensions into one number: a fatal weakness on one is not offset by strength on another.

| Dimension | What to establish | Weak signal |
|---|---|---|
| Authority | Who produced it, their standing, whether the work was reviewed or edited independently | No named author or institution; no review process |
| Recency | Date of the underlying data, not the date of publication | Publication date recent, fieldwork undated or years older |
| Methodology | Design, instrument, mode, analysis; whether enough is reported to judge it | Method described as "a survey of consumers" and nothing more |
| Sample | Size, frame, selection method, response rate, base sizes for the specific claim | Headline claim on an undisclosed or tiny base |
| Geography | The markets actually covered, and whether the finding is claimed beyond them | A single-market study reported as a regional or global truth |
| Relevance | Whether the population, construct and period match your question | Adjacent construct treated as the same construct |
| Conflicts of interest | Who funded it, who benefits from the result, whether it doubles as marketing | Research published by a party selling into the category, concluding the category is growing |
| Evidence type | Primary data, secondary analysis, restatement, or opinion | Commentary presented in the register of finding |

Then assign a tier: **A** (primary evidence, method fully reported, independent, directly relevant), **B** (primary evidence with a material limitation, or independent secondary analysis of primary data), **C** (restatement, or primary evidence too thinly reported to appraise), **D** (opinion, promotional, or unverifiable). Tier D sources do not carry claims. They may be cited as evidence of what is being said in a market, which is a different claim and must be worded as one. *Correct result: a tier for every source, with reasons recorded.*

**Step 7. Extract into a structured table, from the source, in front of you.**
Never summarise a source from memory of reading it, and never from a search result snippet. Open it and extract: the claim in the source's own terms, the figure with its base and exact question wording where quantitative, the population and period, the method, and the page or section locator. Everything downstream inherits what you write here. Where a source gives a percentage, capture what it is a percentage of; "41% of consumers" and "41% of category buyers who replaced an appliance in the last two years" are different claims that look identical once shortened. *Correct result: one row per claim, not one row per source, each traceable to a locator.*

**Step 8. Synthesise by claim, not by source.**
Reorganise the extraction table around the review questions. Under each, group the claims that speak to it and state: what the sources agree on, what they disagree on, how many independent sources support each position, and at what tier. Never write a synthesis as a sequence of source summaries ("Source A found... Source B found..."). That is an annotated bibliography, and it pushes the analytical work onto the reader. *Correct result: for each review question, a claim-level statement with its supporting sources named and its tier profile visible.*

**Step 9. Adjudicate disagreement rather than averaging it.**
Diagnose a conflict before judging it. Work through six causes in order: **definitional** (different constructs under the same word), **methodological** (stated importance versus observed behaviour, different modes or question wording), **sampling** (different populations or frames), **temporal** (different periods, and the thing genuinely changed), **geographic** (different markets), and only then **genuine** (comparable studies, incompatible results). The first five are far more common than the sixth, and most apparent contradictions dissolve into definitional or methodological differences. Where the conflict is genuine, report both positions with their tiers and say which is better evidenced and why; do not split the difference, and never quietly pick the one that suits the narrative (K4 §4.1). *Correct result: every conflict named, diagnosed by cause, and resolved with reasons or explicitly left open.*

**Step 10. Judge sufficiency, and answer the primary research question.**
Classify each review question: **answered** (converging tier A or B sources, directly relevant, adequate bases), **partially answered** (evidence exists with a material limitation on relevance, recency, geography or quality), **contested** (genuine unresolved disagreement), or **absent** (no adequate source found, with the searches that establish this in the log). Then make the recommendation the review exists to support: what can be decided on desk evidence alone, what needs primary research, and what that research would have to measure. Absence of evidence found is not evidence of absence; word it as "no adequate source was identified by the searches recorded in the log", and name where you looked. *Correct result: a sufficiency table, and a primary research recommendation specific about the questions still open.* `RESEARCHER DECISION REQUIRED` where the recommendation is to proceed without primary research on a material decision (K5 §2.5, §2.7).

**Step 11. Write up with the chain intact.**
Every claim carries its source reference in the K2 §4.3 format. Every conclusion carries a K3 confidence level, capped by the weakest source beneath it. Keep the K2 finding-to-interpretation boundary visible: what the sources say is a finding, what you conclude from reading them together is an interpretation, and the reader must see the join. Disclose the review's own limitations next to the conclusions they affect: languages searched, sources not accessible, date cutoff, and the purposive rather than systematic search.

## 8. Analytical framework

The review is built on two structures that interlock.

**The evidence chain, per claim:**

    Review question → Sources found → Quality tier → Claim extracted →
    Convergence or conflict → Adjudication → Synthesised answer →
    Confidence → Remaining gap

Each arrow is a place where the review can fail, and each is inspectable. The critical arrow is *Convergence or conflict → Adjudication*: this is where independent corroboration must be distinguished from repetition of a single original, and where conflict must be diagnosed rather than smoothed.

**The evidence hierarchy, per source.** Sources are not equal and must not be pooled as if they were:

    Primary evidence with method reported
      → Independent secondary analysis of primary data
        → Secondary restatement with the original traceable
          → Secondary restatement with the original untraceable
            → Commentary and opinion

Applying it: a claim's confidence is set by the highest tier that independently supports it, capped by relevance. A tier A source about the wrong population supports nothing. Two tier C sources restating the same tier A original count as one source, not three. And a claim supported only below the third level is reportable as "what is being said" about a market, never as "what is true" of it.

## 9. Output format

**1. Scope and question.** The decision served, the review questions, the scope boundary (markets, period, population, language).

**2. Method and search log.** Source types consulted, search strings, dates, and screening counts. Stated plainly as a purposive, documented review, not a systematic one.

**3. Source table.**

| ID | Source (full reference) | Type | Date of data | Population and geography | Method and base | Evidence type | Tier | Interest to declare |
|---|---|---|---|---|---|---|---|---|

**4. Findings by review question.** For each: the synthesised answer, the supporting source IDs, the tier profile, any conflict and its adjudication, and a confidence level in K3 language.

**5. Where sources disagree.** Each conflict, its diagnosed cause, and the resolution or the reason it remains open.

**6. What the evidence does not establish.** The absent and contested questions, with the searches that establish the absence.

**7. Primary research recommendation.** What still needs to be measured, on what population, and why desk evidence cannot close it. Or an explicit statement that primary research is not required, with the reasoning.

**8. Limitations.** Access, language, recency, and the purposive search.

**When the evidence is thin, the format must not force fabrication (K4 §1).** A review question with no adequate source is reported as such, in Section 6, with the search log as the evidence that looking was done. The source table shrinks; it is never padded. A tier D source is never promoted to fill a row. If the whole review returns little, the correct output is a short review that says so and a strong primary research recommendation, and that is a successful outcome, not a failed one.

## 10. Quality checks

Run before anything is presented. These sit on top of K4 §8.

1. Has every source been opened and read, or is it marked as inaccessible and excluded from carrying claims?
2. Are the locating details for every source taken from the source itself, not from a document citing it?
3. Does every claim in the output appear in the source it is attributed to, in the sense it is used?
4. Has every multi-source claim been checked for citation chaining back to a single original?
5. Does every quantitative claim carry its base, its population and, where it matters, the question wording?
6. Is the date of the underlying data recorded, distinct from the publication date?
7. Has every source been tiered, and does any conclusion rest on tier C or D sources without being worded accordingly?
8. Are funding and interest declared for every source where they exist, and does any conclusion rest mainly on interested sources?
9. Is every conflict between sources reported and diagnosed, with none resolved by silent selection?
10. Does the search log contain the searches that returned nothing, so that absence claims are supported?
11. Is the geography of every finding stated, and has any single-market finding been generalised?
12. Is the review described as purposive rather than systematic wherever its status could be misread?
13. Does the output distinguish what the sources say from what you conclude from reading them together?
14. Would a reader with no access to your files be able to locate and check any cited source from the reference alone?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **Invented citations** | A reference that looks perfectly formed, plausible authors, a real-sounding journal, and cannot be found | Nothing enters the source table without being opened. See §12 |
| **Half-remembered sources** | "There is a well-known study showing..." with no locating detail | Treat the absence of a locator as the absence of a source (K4 §2.4) |
| **Snippet citation** | The claim matches a search result summary but not the source body | Extraction happens only with the source open (Step 7) |
| **Citation chaining** | Five sources, one original, presented as convergence | Trace every claim to its origin before counting sources (Step 5) |
| **Trade press as evidence** | The review is built from articles because articles are findable | Map source types before searching (Step 2) |
| **Publication date mistaken for data date** | A 2025 report reporting 2019 fieldwork, cited as current | Record data date separately (Step 6) |
| **Definitional pooling** | Two figures about "loyalty" combined when they measure different things | Adjudicate by cause before synthesising (Step 9) |
| **Averaging disagreement** | A single figure appears with no trace of the conflict behind it | Report both positions and adjudicate with reasons (K4 §4.1) |
| **Geographic drift** | A finding from one market stated as a general truth | Carry geography in the claim itself, not in an appendix |
| **Interested sources uncaught** | Every source concludes the category is growing; all are published by the category | Declare funding and interest per source (Step 6) |
| **Annotated bibliography** | The output is a sequence of source summaries | Synthesise by claim, not by source (Step 8) |
| **Absence read as absence** | "There is no research on this" after three searches | Phrase absence against the log, and say where you looked |
| **The confident gap-filler** | A number appears in a summary that is not in any source table row | Every number in a summary traces to a row (K2 §5) |

## 12. AI guardrails

Skill-specific. The universal prohibitions in K4 apply in full and are not repeated.

1. **A reference you cannot locate is not a reference.** K4 §2.4 prohibits fabricating sources. The operational rule here: if you cannot state, from the source itself, enough to let someone else find it, the source does not exist for this project. Do not offer it as a lead, a starting point, or "something along these lines". A plausible-sounding pointer costs the researcher more time to disprove than it would have cost to find a real one.

2. **Recognise the half-remembered reference as a distinct failure state, and stop when it occurs.** The signature is a strong sense of familiarity with a topic combined with vagueness about the specifics: roughly this author, roughly this year, roughly this journal, definitely this finding. That pattern is model knowledge reconstructing a plausible artefact, not recall of a document. When it occurs, the correct output is the *claim* stated as unsourced background, together with a description of the kind of source that would support it and where to look, explicitly labelled as a search lead rather than a citation. Never render it in reference format. Formatting is what makes it dangerous: an unformatted hunch reads as a hunch, and the same hunch in a reference list reads as evidence.

3. **Never generate a reference list, a bibliography or a "key sources" section from model knowledge.** Every entry is either something opened in this project or is absent. A bibliography assembled without access is fabrication in bulk, regardless of how many of its entries happen to be real.

4. **Never complete a partial reference.** Do not supply a missing year, volume, page range, DOI, publisher or journal name by inference or pattern. Leave the field marked incomplete and say what is missing. A reference with an invented DOI is more harmful than one with a blank DOI, because it will be checked and appear to fail for the wrong reason.

5. **Never describe searching a source you did not search.** Do not narrate a search process, report hit counts, or state that a database was consulted unless it was. If you cannot search a source type, record it in the log as not consulted, with the reason.

6. **Never upgrade the evidence type.** Reading a summary of a study is not reading the study. A source that mentions a finding is not a source for the finding. Record provenance honestly even when the upgrade would make the review look stronger.

7. **Never reconcile conflicting sources by inventing a middle value.** No figure enters the output that is not in a source. Where sources disagree and no adjudication is possible, both figures appear, with their sources.

8. **Never present the absence of retrieved evidence as the absence of evidence.** The honest statement names the searches performed. Your inability to find something is a fact about the search, not about the literature.

9. **Do not silently substitute an adjacent construct.** If the question is about repair and the available evidence is about durability, say that the evidence is about durability and let the researcher judge the transfer. The substitution is often reasonable; making it invisibly never is.

10. **Where a claim rests on a source you could not fully appraise, cap its confidence at low and say why** (K3 §3.4). Method reporting too thin to appraise is a quality finding, not a detail to pass over.

## 13. Best-practice principles

- **The stopping rule is written before the searching starts.** Desk research has no natural end, so an unstated stopping rule becomes "when the deadline arrives", which systematically over-weights whatever was found first.
- **Search the vocabulary of every community that studies the thing.** Academic, regulatory, commercial and trade communities use different words for the same construct, and a single-vocabulary search returns a biased slice while feeling comprehensive.
- **Count originals, not documents.** Corroboration means independently produced evidence. Thirty articles repeating one study is one piece of evidence with good distribution.
- **The date that matters is the date of the data.** Publication lag routinely exceeds two years, and a report can be new and its evidence old.
- **Read the method section first.** It decides whether the headline is worth reading. Experienced desk researchers spend more time on method and base than on findings, because the finding is only as good as what produced it.
- **Interest is not disqualifying, but it is always relevant.** Research funded by a party with a stake can be excellent. It requires the funding declared, the method scrutinised harder, and independent corroboration before it carries a conclusion alone.
- **A conflict between sources is usually a definitional or methodological difference wearing a disguise.** Diagnose before adjudicating. The disagreement, once diagnosed, is frequently the most useful finding in the review, because it reveals that the construct is measured in incompatible ways.
- **Distinguish stated importance from observed behaviour ruthlessly.** They diverge systematically and predictably, and pooling them is one of the most common ways commercial desk research reaches a confidently wrong conclusion.
- **Grey literature is worth the effort.** Working papers, conference material, regulator consultations and association reporting often hold the most current and most specific evidence, and are the material a search-engine-led review misses.
- **A short honest review beats a long padded one.** The value of desk research is often the finding that the question is not answered by existing evidence, which is exactly the finding that saves the cost of a study designed to re-establish what is known.
- **Write the gap statement even when the review succeeds.** What the evidence does not cover is the brief for whatever comes next.
- **Keep the log even under time pressure.** The log is what lets a second researcher continue the work, what lets you defend an absence claim, and what stops the third re-run of the same search two weeks later.

## 14. Worked example

Generic fictional scenario, commercial.

**INPUT**

A mid-sized manufacturer of domestic appliances is deciding whether to invest in a repairable product architecture. The category team believes repairability now influences purchase. Finance wants evidence before funding a primary study. The question arriving is: "is repairability a purchase driver, and do we need research to find out?"

**PROCESS**

*Step 1.* The question is reframed into three answerable ones. (Q1) Among domestic appliance buyers in the home market since 2021, what proportion report repairability influencing choice? (Q2) Does stated importance correspond to observed purchasing or price sensitivity? (Q3) Has regulation changed the information available to buyers at the point of sale? Stopping rule: two independent adequate sources per question, or a documented absence.

*Step 2.* Landscape map. Q1 is likely to sit in published consumer research and trade association reporting. Q2 is likely academic or in behavioural data, not in trade press. Q3 sits with the regulator and standards bodies, and will be authoritative and findable.

*Step 3.* Search strings built across three vocabularies: repairability, reparability, right to repair, product lifetime, durability, serviceability. The academic vocabulary turns out to favour "product lifetime extension", which the commercial search would have missed. Logged.

*Step 4 and 5.* Nineteen candidates screened, eight retained. Verification catches the key problem: four trade articles all report the same headline proportion of buyers saying repairability matters, and all four trace to a single survey commissioned by an industry body. That is one source, not four. Two further items are excluded as inaccessible and recorded as such.

*Step 6.* Tiering. The regulatory source is tier A for Q3 (primary, authoritative, directly relevant). The industry-body survey is tier B for Q1: primary data, adequate base, but the commissioning body has an interest in the answer and the question wording is a stated-importance rating, which is a known inflator. One academic study of actual purchase behaviour in an adjacent market is tier A for method but only partially relevant on geography.

*The judgement call.* Q1 now has two numbers that appear to conflict: the industry survey reports a large majority saying repairability influences their choice, and the academic behavioural study finds a far smaller effect on actual purchases. The temptation is to average, or to pick the one the category team will like. Instead the conflict is diagnosed against the six causes in Step 9. It is not genuine disagreement: it is methodological (stated importance versus revealed preference) compounded by geographic (different market). Both are reported. The synthesised answer is that a large majority of buyers say repairability matters, that the one available behavioural study in an adjacent market finds a much weaker effect on actual purchase, and that this gap is the finding, not a problem in the sources.

*Step 10.* Sufficiency. Q3 is **answered** (high confidence, tier A regulatory source). Q1 is **partially answered**: the stated-importance figure is real but interested and known to inflate, moderate confidence at best. Q2 is **contested and under-evidenced**: no source measures behaviour in this market, and the stated-behaviour gap is precisely what the investment decision turns on.

**OUTPUT**

A review that answers the regulatory question outright, reports the stated-importance evidence with its interest and wording flagged, names the stated-versus-observed gap as the central unresolved issue, and recommends a primary study designed to measure willingness to pay for repairability at realistic price differentials rather than to re-measure stated importance, which the desk evidence already establishes. It also saves the cost of the study the category team originally wanted, which would have produced the same inflated number.

`RESEARCHER SIGN-OFF REQUIRED` on the investment-facing recommendation (K5 §2.5).

## 15. Advanced usage

**Living reviews.** Where a market is moving, keep the search log and source table as a maintained asset with a scheduled re-run of the recorded searches. The value is the diff: what is new, what has been superseded, and which conclusions have expired. This is where a review stops being a document and becomes infrastructure. See **14.03 Research Repository and Knowledge Curation**.

**Reviews as instrument design input.** Extract the exact question wordings and scale formats used by the sources, not only their findings. Reusing a validated wording makes your primary result comparable to the existing evidence base, which is worth more than a marginally better question that compares to nothing. Hands over to **02.01 Survey Questionnaire Design** and **02.07 Scale and Measurement Selection**.

**Deliberate disconfirmation.** After synthesis, run a search designed to find evidence against the conclusion, using the vocabulary a sceptic would use. This is the most effective single guard against a review that assembled itself around a stakeholder's prior, and it takes under an hour. See **13.04 Bias Detection**.

**Escalating rigour.** Where the review turns out to matter more than expected, the upgrade path is explicit: pre-specify the protocol, make the search exhaustive, add a second independent screener, and report to systematic-review standard. That is a different method with a different cost, and it is **15.06 Systematic Literature Review**. Do not retrofit the label onto a purposive review; the search log will not support it.

**Where the standard approach does not fit.** For emerging topics with almost no formal literature, shift the source map toward regulator consultations, patent and filing disclosures, conference material and practitioner communities, and lower the tier expectation explicitly rather than silently. For heavily commercial topics where most published material is promotional, expect a tier C and D evidence base and word every conclusion as what is being claimed in the market rather than what is established.

## 16. Skill chain

**Recommended previous skills:**
- **01.02 Business Problem to Research Question.** Hands over the decision and the question the review must serve, without which there is no relevance test or stopping rule.
- **01.01 Research Brief Interrogation.** Hands over scope, stakeholders and the priors the review will be read against.

**Recommended next skills:**
- **10.04 Research Gap Identification.** Takes the sufficiency table and turns the absent and contested questions into a specification for what must be researched.
- **01.04 Research Method Selection.** Takes the primary research recommendation and designs the study that closes the gap.
- **10.02 Evidence Synthesis** and **10.03 Multi-Source Research Synthesis.** Take the appraised source set and integrate it with primary and internal evidence.
- **15.06 Systematic Literature Review.** Takes over entirely where academic or protocol-standard rigour is required.

**Runs well alongside:**
- **13.02 Source and Citation Verification**, which audits the reference set this skill produces.
- **10.05 Competitive and Market Research**, which uses the same appraisal framework on company and market sources.
- **12.04 Research Evidence Integration**, where desk evidence sits beside primary findings in a single report.

---
A Yazi Supplied Skill and resource.