---
name: research-gap-identification
description: >
  Determines what is not known, which gaps matter to the decision, and which are
  worth filling. Use for "what do we still not know", "what should we research
  next", "where are the gaps", "turn this review into a research plan", "is
  there enough evidence to decide", "do we need more research or can we decide
  now", "what would we have to find out", "prioritise the unknowns".
category: 10 Desk Research and Evidence Synthesis
ref: "10.04"
tier: 1
inherits: [K2, K3, K4, K5]
---

# Research Gap Identification

## 1. One-line description

A method for establishing what is not known, separating gaps in the literature from gaps that matter to the decision, testing whether each gap is real rather than an artefact of where you looked, and producing a prioritised register that says which gaps to fill, which to accept, and which mean the decision must be made under uncertainty now.

## 2. What this skill is used for

**The research problem it solves.** A review that reports what is known is a document. A review that also reports what is not known, and what to do about it, is a research plan. The step between them is where most work stops, and where it does not stop it usually goes wrong in one of two directions. The first is completeness thinking: every unanswered question in the literature is listed as a gap, producing a research agenda nobody can afford and nobody needed, because most literature gaps exist for the excellent reason that nobody had a decision resting on them. The second is the opposite failure, treating any absence as a reason to commission work, so that gaps get filled because they are visible rather than because filling them changes anything. Both miss the distinction that actually matters: whether the decision in front of the organisation turns on the missing information. This skill installs that distinction, classifies gaps by type because different types need different remedies, forces honesty about whether an apparent gap is really a gap in your searching, and treats "the gap cannot be filled in time, so decide now with mitigation" as a legitimate and often correct recommendation rather than an admission of failure.

**Where it sits in the research lifecycle.** Immediately after evidence assessment and immediately before design. It is the hinge: it takes a sufficiency judgement and converts it into either a research specification or a defensible decision to proceed without one. It also runs at the end of a project, where the gaps that remain become the next brief.

**Typical use cases.**
- Turning a completed literature or evidence review into a costed research plan.
- Deciding whether existing evidence is sufficient to make a decision now.
- Prioritising a research agenda when the list of unknowns exceeds the budget.
- Writing the "what we could not establish" section of a substantial report so that it is useful rather than defensive.
- Justifying to a stakeholder why a requested study is not worth running.
- Framing a research programme for a new area where almost nothing is known and everything looks like a priority.

**Who uses it.** Research directors and heads of insight setting an agenda; consultants converting a review into a proposal; evaluation and policy researchers specifying the next phase; product and UX research leads deciding what to test before a build; postgraduate researchers justifying the contribution of a proposed study.

## 3. When to use it

- A review or synthesis is complete and the question "so what do we research now" is live.
- A decision is imminent and someone needs to say whether the evidence is sufficient.
- The list of things the team would like to know is longer than the budget or the calendar.
- A stakeholder is requesting a study and you suspect the answer already exists, or that the answer would not change the decision.
- A report needs a credible account of its own limits.
- A new area is being entered and the evidence base has to be characterised before anything is commissioned.
- A previous study answered its questions and the organisation is asking what comes next.
- Two sources conflict irreconcilably and the question is whether resolving the conflict is worth funding.

## 4. When NOT to use it

- **The evidence has not been assembled or assessed yet.** You cannot identify a gap in an evidence base you have not built. Establish what exists through **10.01 Literature Review and Desk Research**, and what it collectively supports through **10.02 Evidence Synthesis** or **10.03 Multi-Source Research Synthesis**, before asking what is missing. A gap list produced without an evidence base is a list of things you personally do not know.
- **The decision is not identified.** Gap prioritisation is entirely relative to a decision. Without one, every gap is equally interesting and equally unrankable, and the output degenerates into a wish list. If no decision has been named, go to **01.02 Business Problem to Research Question** first and return with one.
- **The task is designing the study that fills a gap.** This skill specifies what needs to be known and to what precision. Choosing the method, the sample and the instrument is **01.04 Research Method Selection**, then the design skills. A gap register that has already fixed the methodology has usually skipped the question of whether the gap is fillable any other way.
- **The gap is not empirical.** Some things a decision waits on are not questions research can answer: what the organisation's risk appetite is, what it values, what it is willing to be seen doing, whether a trade-off is acceptable. These look like gaps and are not. Naming them as research gaps delays a decision that only leadership can make, and frames a values judgement as a data problem. Say plainly that the question is not empirical and return it. `RESEARCHER DECISION REQUIRED` (K5 §2.3).
- **The stakeholder wants a research plan to postpone a decision.** Commissioning research is a recognised way to defer an uncomfortable choice, and a gap register can be recruited into it. Where the evidence is already sufficient for the decision at hand, the honest output says so and names the specific further study as unnecessary, with the reasoning (K4 §4.2).
- **The purpose is academic contribution rather than decision support.** Justifying a study's novelty against a body of published literature is a different exercise with different criteria, where a gap's value comes from its theoretical significance rather than its decision relevance. See **15.06 Systematic Literature Review** and **15.01 Academic Research Design**. Applying decision-relevance filters to a doctoral contribution argument will discard exactly the gaps that matter there.
- **The search coverage is so poor that absence cannot be distinguished from failure to look.** Where the review consulted two source types in one language over an afternoon, everything looks like a gap. Improve the coverage or report the register as provisional with the coverage limits stated at the top, not in a footnote.
- **The gap is being used to avoid reporting an inconvenient finding.** "More research is needed" attached to a finding somebody dislikes is a rhetorical move, not a gap. A genuine gap is specific about what is missing; this one never is.

## 5. Required inputs

**Required.**
- **The decision the research would inform, with its timeline and its consequence.** Every prioritisation step depends on it. Without a decision there is no materiality test, no feasibility test and no basis for saying a gap does not matter.
- **An assessed evidence base,** ideally as a sufficiency table classifying each question as answered, partially answered, contested or absent. This is what the gap register is built from.
- **An honest account of search coverage:** what was searched, in what languages, over what period, and what was not reachable. The reality test in Step 4 is not possible without it, and a gap register that overstates absence sends people to look for things that are already known.

**Optional, and what each one adds.**
- **The budget and calendar available for further research.** Converts feasibility from an abstract judgement into a real constraint, and usually collapses a long gap list to three or four candidates immediately.
- **The decision's value at risk, even roughly.** Lets you compare the cost of research against what the decision is worth, which is the only defensible basis for saying a gap is not worth filling.
- **The organisation's existing internal evidence,** including unpublished and abandoned work. A surprising proportion of apparent gaps have been filled internally by someone who left.
- **Knowledge of what has been tried before and failed.** A gap that three previous studies failed to close is unlikely to be closed by a fourth of the same design, and that history changes the recommendation from "research it" to "research it differently, or accept it".
- **The decision-maker's current working assumption for each gap.** Reveals which gaps are already being filled with a guess, which is where research has the most value and the most political sensitivity.
- **Regulatory, ethical or access constraints.** Determine fillability directly. Some gaps cannot be filled because the study cannot be run.

## 6. Questions to ask before starting

1. **What decision does this feed, when is it made, and what is at stake?** Sets the materiality threshold and the feasibility window. Default if unanswered: assume the decision is imminent and moderate in consequence, and flag every gap whose priority would change under a different assumption.
2. **What is the decision-maker currently assuming about each unknown?** A gap where a confident wrong assumption is in place is far more valuable to fill than one where uncertainty is already acknowledged. Default: ask directly; the answer is usually offered readily and changes the ranking.
3. **Would a different answer change what gets done?** The central test. If the action is the same either way, the gap is interesting and not worth funding. Default: apply the test to every gap and record the answer, including where it is uncomfortable.
4. **How thoroughly was the evidence base actually searched?** Determines how much of the register is real absence and how much is unexplored territory. Default: state coverage limits at the head of the register and mark affected gaps provisional.
5. **What is the realistic budget and calendar?** Distinguishes gaps that can be filled before the decision from those that cannot, which is a different recommendation entirely. Default: assume the decision will not wait, and separate the register accordingly.
6. **Has anyone tried to answer this before?** Prevents recommissioning a study that already failed, and reveals whether the difficulty is practical or fundamental. Default: ask about abandoned and unpublished work specifically, because it does not appear in any search.
7. **Are any of these gaps unfillable in principle?** Changes the output from a research recommendation to a risk-management one. Default: run the fillability test in Step 7 on every gap rather than assuming research is always the answer.

## 7. Step-by-step methodology

**Step 1. Map what the decision depends on, before looking at the evidence.**
Write the decision as a choice between named options, then list every proposition that has to be true, or every quantity that has to fall in a particular range, for each option to be the right one. This is the dependency map, and building it from the decision rather than from the evidence base is what stops the register becoming a list of things the literature happens not to have covered. Include the propositions nobody is questioning, because the unexamined assumption is where the expensive surprises live: the dependency that everyone treats as settled, has no evidence behind it, and would reverse the decision if wrong. Mark each dependency with how sensitive the decision is to it, on a simple three-level scale of decisive, contributory or peripheral. *Correct result: a dependency list written from the decision, each item marked for sensitivity, including at least one dependency nobody had previously articulated.*

**Step 2. Set the evidence status against each dependency.**
Bring in the sufficiency assessment from the review or synthesis and attach it to the dependency map rather than the other way round. Each dependency gets an evidence status: established, partially established, contested, or unaddressed. Where a dependency is supported by evidence the synthesis rated at low confidence, that is not "established with a caveat", it is partially established, and the distinction is the whole point of the exercise. Two things become visible at this step and neither is visible from the evidence base alone: dependencies that are decisive and unaddressed, which are the real gaps, and evidence that is abundant and attached to nothing the decision turns on, which is where the previous research budget went. *Correct result: the dependency map with an evidence status per row, and an immediate read of which decisive dependencies are unsupported.*

**Step 3. Classify every gap by type, because the remedy differs.**
There are seven, and they are not interchangeable. **No evidence**: nothing addresses the question at all. **Weak evidence**: something addresses it, but the bases are too small, the method too thinly reported, or the sources too interested to carry the claim. **Conflicting evidence**: adequate sources disagree and the disagreement survived adjudication. **Different population or market**: good evidence exists about a population that is not yours, and the transfer is an assumption rather than a finding. **Out of date**: good evidence exists about a period that has passed, and something has changed since. **Different construct**: evidence exists about an adjacent thing, and the substitution is doing unexamined work. **Never applied**: the evidence exists, is adequate, and nobody has ever brought it to bear on this question. The last type is the most valuable and the most missed, because it looks like a gap and is actually a synthesis task that costs a week rather than a study that costs a quarter. The remedies differ sharply: a weak-evidence gap may need a better-powered version of an existing design, a construct gap needs measurement development before anything else, a transferability gap may be closed by a small calibration study rather than a full replication, and a never-applied gap needs no fieldwork at all. *Correct result: every gap carries one of the seven types, and the type has been used to name the remedy rather than defaulting to "run a study".*

**Step 4. Test whether the gap is real or an artefact of where you looked.**
Absence of retrieved evidence is a fact about the search, not about the world (K4 §2.4 in spirit, and 10.01 Step 10). For every gap classified as "no evidence", ask four questions and record the answers. Which source types would carry this evidence if it existed, and were they searched? What vocabulary would the community that studies this use, and was it searched? Would this evidence be published at all, or would it sit in commercial, internal or unpublished form where no search reaches it? Has the organisation itself already answered this, in work that was never written up? Then assign a coverage verdict: **confirmed absent** (the sources that would hold it were searched competently and it is not there), **probably absent**, or **not established** (the search cannot support an absence claim here). A register that presents all three as equivalent will send an expensive study after something that already exists, which is a specific and recoverable failure worth spending an hour to avoid. *Correct result: a coverage verdict per gap, and any "not established" gaps routed back to searching before they are costed.*

**Step 5. Apply the materiality test, and expect to discard most of the list.**
For each gap ask: if this were answered, at each plausible answer, would the decision change? Work through it concretely rather than in the abstract. Take the gap, imagine the two or three answers it could return, and trace each one to what the organisation would then do. Where every answer leads to the same action, the gap does not matter for this decision, and it belongs in a "noted, not prioritised" section rather than in the plan. This is the step that distinguishes a gap in the literature from a gap that matters, and it typically removes half to two thirds of the list. It also occasionally does the opposite and promotes something that looked minor: a peripheral-looking dependency whose plausible answers include one that would reverse the decision is a decisive gap regardless of how small it appeared. *Correct result: each gap marked decision-changing or not, with the reasoning recorded in one line, and a visibly shorter live list.*

**Step 6. Ask what a filled gap would actually be worth.**
Materiality says the answer could change the decision. This step asks whether it is worth paying to find out, and it has three components. **How much uncertainty is there really?** A gap where the plausible range is narrow and every point in it leads to the same action is not worth resolving even if it is technically unaddressed. **What does the decision have at risk?** Research costing a substantial fraction of the value at stake is rarely rational, and saying so is part of the job. **How much would the research actually reduce the uncertainty?** A study that would return a moderate-confidence answer to a question that needs a definitive one has not closed the gap, it has moved it. Do this qualitatively and transparently rather than with a spurious calculation; the reasoning is what the reader needs, and a number here would be false precision (K3 §4.4). *Correct result: a value judgement per decision-changing gap, stated in words, with the gaps that fail it moved out of the plan and the reason recorded.*

**Step 7. Test fillability, and separate what primary research can close from what it cannot.**
Not every gap is a research gap. Sort each into one of four. **Fillable by primary research**, with the design type named and the reason it would work. **Fillable, but not by research**: the answer sits in internal data nobody has queried, in a system nobody has interrogated, in a partner's records, or in a synthesis of evidence that already exists (the never-applied type from Step 3). **Fillable only over time**: the question is about what happens after a change that has not happened yet, or about a duration that has not elapsed, and no design compresses it. **Not fillable**: the study cannot be run for ethical, legal, access or practical reasons, or the question is about a counterfactual nobody can observe, or the phenomenon is not stable enough to measure. Getting this wrong in the optimistic direction is expensive: a commissioned study aimed at a gap that time alone can close produces a plausible-looking answer to a question that was not yet askable. Where a gap is not fillable, the output changes from a research recommendation to a risk-management one, and that is a legitimate deliverable. *Correct result: every live gap assigned to one of the four categories, with the design or the reason named, and the unfillable ones explicitly routed to mitigation rather than to a brief.*

**Step 8. Prioritise on stated criteria, with the rule written down.**
Rank the live gaps on three criteria, and state the rule you used so a reader can disagree with the ranking rather than with the register. **Decision relevance**: decisive, contributory or peripheral, from Step 1. **Changeability**: how likely the answer is to change the action, from Step 5. **Feasibility**: whether it can be closed within the budget and before the decision, from Step 7. The rule that works in practice is lexicographic rather than additive: decisive and feasible gaps come first in order of changeability, then decisive but infeasible gaps go to the mitigation list, then contributory and cheap gaps, and peripheral gaps do not enter the plan at all regardless of how easy they are. Do not build a weighted composite score. Weighted scores in prioritisation are almost always retrofitted until the ranking matches what someone already wanted, and the weights are unarguable because they are invisible. *Correct result: a ranked list with the ranking rule stated in the output, and every gap's position explicable from the three criteria.*

**Step 9. Identify the gaps where the honest recommendation is to decide under uncertainty.**
This is the step that most gap analyses omit, and it is frequently the most valuable output. A gap belongs here when it is decisive but not fillable before the decision, when it is not fillable at all, when the research would cost more than the decision is worth, or when the study would return an answer too weak to settle it. The recommendation is then explicit: make the decision now, on the current evidence, with the uncertainty named. That recommendation is not complete without three things. State what is being assumed in place of the missing evidence, in words the decision-maker would recognise as their own assumption. State what would happen if the assumption is wrong, and how badly. And specify the mitigation: a reversible first step, a staged commitment, a monitoring indicator that would reveal the assumption failing early, or a tripwire that triggers a review. A gap register that recommends delay for every unknown is not exercising judgement; delay has a cost, and it is usually invisible in the register while the cost of research is right there in the proposal. *Correct result: an explicit decide-now list, each item with its assumption, its exposure and its mitigation named.* `RESEARCHER DECISION REQUIRED` on each (K5 §2.5, §2.7).

**Step 10. Write the register and the specification it hands forward.**
The register carries every gap, including the ones that were discarded and why, because a reader who cannot see what was excluded cannot tell whether the list is a judgement or a selection. For each gap that enters the plan, write the specification: the question in answerable form naming population, construct and period; the precision required, which is the part most often left out and the part that determines cost; what would count as an adequate answer; and what the answer would change. That specification is the brief, and it hands directly to question formulation and method selection. Finally, state the register's own limits: the search coverage behind it, the decision it was built against, and its shelf life, because a gap register is valid for one decision at one moment and will be reused for another if you do not say so. *Correct result: a prioritised register with discards visible, a written specification per prioritised gap, and a stated shelf life.*

## 8. Analytical framework

One chain, applied per gap, and one disposition set at the end of it.

**The gap chain:**

    Decision → Dependency → Evidence status → Gap type → Coverage verdict →
    Materiality → Value of resolution → Fillability → Priority → Disposition

Reading it from the left is what makes the exercise useful; reading it from the right, which is what happens when you start from the evidence base, produces a list of literature gaps ranked by how visible they are. Two arrows carry most of the value. *Evidence status → Gap type* determines the remedy, and collapsing seven types into one ("we need research") is what generates unaffordable agendas. *Materiality → Value of resolution* is the pair of filters that turns a long list into a plan, and skipping either one is why research agendas grow faster than budgets.

**The four dispositions.** Every gap ends in exactly one, and the register is organised by them:

    FILL      Research it. The specification is written and handed to design.
    RESOLVE   The answer exists somewhere unsearched, unqueried or unsynthesised.
              Costs days, not quarters.
    ACCEPT    Decide now under stated uncertainty, with the assumption, the
              exposure and the mitigation named.
    MONITOR   Do not research and do not decide yet. Set an indicator and a
              tripwire, and revisit when it trips.

Applying it: the disposition follows from fillability and timing, not from how interesting the gap is. The most common error is assigning FILL by default, and the second most common is assigning ACCEPT to a gap that is genuinely RESOLVE, which means paying for uncertainty that a week of work would have removed.

## 9. Output format

**1. The decision and its dependencies.** The decision as a choice between options, the dependency map, and the sensitivity marking per dependency.

**2. Evidence status summary.** What the evidence base establishes against each dependency, carried through from the review or synthesis with its confidence levels intact.

**3. Search coverage statement.** What was searched, in what languages, over what period, and what was not reachable. Positioned before the register, because it qualifies every absence claim in it.

**4. The gap register.**

| ID | Gap | Dependency it sits under | Type (one of seven) | Coverage verdict | Decision-changing | Fillable by | Priority | Disposition |
|---|---|---|---|---|---|---|---|---|

**5. Prioritised research specification.** For each FILL gap: the question in answerable form, the precision required, what would count as an adequate answer, what it would change, and an indicative feasibility note.

**6. Quick resolutions.** The RESOLVE gaps, with where the answer is likely to be and what it would take to get it.

**7. Decide-now list.** The ACCEPT gaps, each with the assumption being made, the exposure if it is wrong, and the proposed mitigation.

**8. Monitoring list.** The MONITOR gaps with their indicators and tripwires.

**9. Noted and not prioritised.** The discarded gaps with the reason for discard. This section is required, not optional.

**10. Register limits and shelf life.** The decision it was built for, the coverage behind it, and when it expires.

**When the evidence is thin, the format must not force fabrication (K4 §1).** Where the evidence base is very small, most dependencies will be unaddressed and the register will be short and stark, and that is the correct output rather than a reason to elaborate. Do not invent plausible-sounding gaps to populate a register, and do not convert a gap you cannot characterise into a specification with an invented precision requirement. Where you cannot say what precision an answer needs, say that establishing the precision requirement is itself the first task. A register with four gaps and a clear recommendation is worth more than one with twenty and no ranking rule.

## 10. Quality checks

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

1. Was the dependency map written from the decision rather than derived from the evidence base?
2. Does the map include at least one dependency nobody had previously articulated, and has each been marked for sensitivity?
3. Does every gap carry one of the seven types, and has the type driven the remedy rather than a default recommendation to run a study?
4. Does every "no evidence" gap carry a coverage verdict, and have the "not established" ones been sent back to searching before being costed?
5. Has the materiality test been applied to every gap, with the reasoning recorded, and did it actually remove anything?
6. Is any gap prioritised where every plausible answer leads to the same action?
7. Has fillability been assessed, and is any gap assigned to primary research that only time, internal data access or a synthesis could close?
8. Is the prioritisation rule stated in the output, and is every gap's position explicable from it?
9. Is there a weighted composite score anywhere, and if so, why?
10. Does the decide-now list exist, and does every item carry its assumption, its exposure and its mitigation?
11. Are the discarded gaps visible, with reasons, so the register reads as a judgement rather than a selection?
12. Does every FILL specification state the precision required, not just the question?
13. Does the register state the decision it was built for and its shelf life?
14. Is any gap in the register really a values or strategy judgement being returned as a research question?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **The completeness agenda** | Every unanswered question in the literature listed as a gap | Build from the decision's dependencies, not the evidence base (Step 1) |
| **Literature gap mistaken for decision gap** | A well-argued gap that would not change any action | Materiality test with the answers traced to actions (Step 5) |
| **Absence of search read as absence of evidence** | "No research exists on this" after a narrow search | Coverage verdict per gap, stated search limits (Step 4) |
| **The internal answer nobody looked for** | A commissioned study duplicating unpublished internal work | Ask about abandoned and unwritten work explicitly |
| **Every gap becomes a study** | A register where every disposition is FILL | Four dispositions, assigned from fillability and timing (§8) |
| **The unfillable brief** | A study commissioned to answer a question only time can settle | Fillability test with four categories (Step 7) |
| **Construct gap treated as evidence gap** | More studies commissioned on a construct nobody has defined | Classify by type; construct gaps need measurement work first |
| **Precision left unstated** | A specification that says what to measure but not how precisely | Precision is a required field in every specification (Step 10) |
| **Weighted score prioritisation** | A composite ranking whose weights nobody can defend | Lexicographic rule, stated in the output (Step 8) |
| **Research as delay** | "More research is needed" attached to an unwelcome finding | A genuine gap is specific; a rhetorical one never is |
| **The invisible cost of waiting** | Delay recommended everywhere, its cost mentioned nowhere | The decide-now list is a required section (Step 9) |
| **Values question returned as a gap** | A research brief for a question about risk appetite | Name it as non-empirical and return it (§4) |
| **Register reused for a different decision** | A prioritisation applied to a decision it was not built for | State the decision and the shelf life on the register |
| **Discards hidden** | A short clean list with no record of what was excluded | Discards with reasons as a required section |

## 12. AI guardrails

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

1. **Never state that no research exists on a topic.** The honest form names the searches performed and the source types covered, and says that no adequate source was identified by them. Your inability to retrieve something is a fact about the retrieval, not about the world, and a confident absence claim sends people to fill a gap that may already be filled.

2. **Never generate gaps from general knowledge of what a literature "usually" lacks.** A gap is an absence in a specific assembled evidence base, established against a specific dependency. A plausible-sounding gap produced from pattern rather than from the register is a fabrication wearing the format of an analysis (K4 §1).

3. **Never assign a gap to primary research without naming what design would close it and why.** "Further research is needed" without a design is not a recommendation, and it is the phrase behind most unaffordable research agendas. If you cannot name the design, the honest answer is that the fillability of this gap has not been established.

4. **Never quantify the value of resolving a gap.** Expected-value arithmetic on invented probabilities produces a number that will be quoted and cannot be defended. State the reasoning in words (K3 §4.4).

5. **Never present a decision-changing test you have not actually run.** Trace each gap's plausible answers to actions explicitly, and where you cannot establish what the organisation would do differently, say that the test could not be completed rather than assuming the gap matters.

6. **Never let the register's structure generate its content.** If a category has no members, it is empty. A monitoring list with nothing in it says something true; a monitoring list padded to look balanced does not.

7. **Never recommend delay without stating its cost,** and never recommend deciding now without stating the exposure. Both recommendations are incomplete in the same way, and the asymmetry in how they are usually presented is what makes research agendas grow.

8. **Never convert a values, strategy or risk-appetite question into a research gap.** Where the missing input is a judgement only the organisation can make, name it as such and return it. Framing it as a data problem is a specific and consequential failure, because it produces a study that cannot succeed.

9. **Never carry a gap register across decisions.** Priority is relative to one decision at one moment. Reusing a register for a different decision silently reuses its materiality judgements, which were made against something else.

10. **Where the evidence base was assembled by a search you cannot characterise, mark the whole register provisional** and say so at the top, not in a limitations paragraph at the end. Every absence claim in it inherits that uncertainty.

## 13. Best-practice principles

- **Build from the decision backwards.** A dependency map written from the decision produces a different and much shorter list than one derived from the evidence base, and the difference is precisely the gaps worth funding.
- **The unexamined assumption is where the money is.** The dependency everyone treats as settled, that has no evidence behind it and would reverse the decision if wrong, is the highest-value gap in most projects and it never appears on anybody's list, because nobody is asking about it.
- **Most literature gaps are not worth filling,** and saying so is the professional act. A gap exists in the literature because nobody had a decision resting on it; that is usually a good reason, not an oversight.
- **The never-applied gap is the cheapest win available.** Evidence that exists, is adequate, and has simply never been brought to this question costs a week to resolve and is routinely mistaken for a gap requiring a quarter of fieldwork.
- **Distinguish "we do not know" from "we did not look".** The two are indistinguishable in a register and completely different in what they should trigger, and only an honest coverage statement separates them.
- **Ask what the decision-maker is currently assuming.** A gap where a confident assumption is already in place has far more value than one where uncertainty is acknowledged, because acknowledged uncertainty is already being managed.
- **Precision is the expensive part of a specification.** "How many" and "how many, to within five points, with segment-level reads" are the same question at three times the price. A specification without a precision requirement cannot be costed and will be scoped by whoever quotes for it.
- **Not every gap is a research gap.** Time gaps, access gaps, counterfactual gaps and values gaps all look like research questions and none of them can be answered by a study. Sorting them out early prevents a brief that cannot succeed.
- **Deciding under uncertainty is a legitimate recommendation,** and often the correct one. The cost of delay is real and is almost always invisible in a document whose subject is what we do not know.
- **A mitigation is part of the recommendation, not a consolation.** A reversible first step, a staged commitment or a monitoring tripwire converts an unfilled gap from an exposure into a managed risk.
- **Show the discards.** A register without them looks like a plan and is actually a selection, and the reader has no way to tell.
- **Give the register an expiry date.** It was built for one decision at one moment, and it will otherwise be reused for a different one with its judgements silently intact.

## 14. Worked example

Generic fictional scenario, digital product and UX.

**INPUT**

A public-facing digital service is deciding whether to rebuild its identity verification step, which is where the largest share of applicants abandon. Two options: a substantial rebuild introducing a new verification route, or a smaller set of interface changes to the existing route. A desk review and an internal evidence synthesis have been completed. The team has produced a list of eleven things "we still need to research", and the delivery deadline is four months away.

**PROCESS**

*Step 1.* The dependency map is written from the decision rather than from the eleven items. The rebuild is the right choice only if: abandonment at this step is caused by the verification requirement itself rather than by what precedes it; the new route would be usable by the applicant groups who abandon most; the new route would not shift abandonment to a different step; and the organisation can operate the new route within its existing assurance rules. The fourth dependency had never been written down. It is marked decisive.

*Step 2.* Evidence status. Abandonment location is established from analytics at high confidence. Cause is unaddressed. Usability of the proposed route for the affected groups is unaddressed. Displacement of abandonment is unaddressed. Operability under assurance rules is unaddressed and, as it turns out, is not a research question at all.

*Step 3.* The eleven items are classified. Four are never-applied gaps: the answers sit in existing session data and a previous accessibility audit that nobody has read against this question. Three are different-population gaps, resting on published usability evidence about a general population when the abandoning group skews towards applicants without standard documentation. Two are weak-evidence gaps. One is a construct gap: the team has been treating "abandonment" and "failure" as the same event when the data distinguishes them. One is not a gap at all.

*Step 4.* Coverage. The desk review searched two source types in one language. The three transferability gaps are marked probably absent; two of the weak-evidence gaps are marked not established, and are sent back for a targeted search before being costed.

*Step 5.* Materiality. Six of the eleven fail the test: whichever way they resolve, the team does the same thing. They move to the noted section with a line each.

*The judgement call.* The most-requested item on the original list was a large quantitative study of applicant attitudes to identity verification. It survives materiality, because attitudes could in principle change the choice of route. It fails Step 6. The plausible range of answers is narrow, the study would take three months of the four available, and at every plausible answer the team would still need to know whether the new route is usable by the affected groups, which is a different and cheaper study. It is recommended against explicitly, with the reasoning written out, because a recommendation against a study that senior stakeholders have already discussed needs to be arguable rather than merely stated.

*Step 7.* Fillability. Cause of abandonment: fillable, by moderated sessions with recruited applicants from the affected groups, roughly six weeks. Usability of the new route: fillable, by prototype testing, and it depends on the prototype existing. Displacement: fillable only over time, because it cannot be observed until the route is live; disposition MONITOR, with an instrumented indicator specified before launch. Operability under assurance rules: not a research question, it is a governance decision, returned to the service owner.

*Step 9.* One ACCEPT item. Whether the new route holds up at full volume cannot be established before the decision. The assumption is stated, the exposure is stated, and the mitigation is a staged rollout with a defined rollback trigger.

**OUTPUT**

A register that reduces eleven requested studies to two commissioned pieces of work totalling around eight weeks, four questions answerable from data already held, one monitoring indicator to be instrumented before launch, one item returned as a governance decision, and one accepted uncertainty with a staged rollout attached. The large attitudinal study is recommended against in writing, with the reasoning available for the stakeholder who asked for it.

`RESEARCHER DECISION REQUIRED` on the accepted uncertainty and its rollback trigger (K5 §2.5, §2.7). `RESEARCHER SIGN-OFF REQUIRED` on the recommendation against the requested study, which is a judgement with political consequences and should carry a named researcher (K5 §2.1).

## 15. Advanced usage

**The standing gap register.** In organisations with a continuing research programme, maintain the register as a living asset rather than a per-project artefact. Gaps close, new ones open, and the history of which gaps were accepted and what happened afterwards is the most useful calibration data an insight function can hold: it shows whether the acceptances were right. Pair with **14.03 Research Repository and Knowledge Curation**.

**Assumption auditing as a standalone exercise.** Run Step 1 alone against a decision that has already been made, and check what the organisation was assuming. This is fast, it is uncomfortable, and it regularly finds a decisive dependency with nothing behind it. It works particularly well as a pre-mortem input, and it produces a better research agenda than any amount of asking stakeholders what they would like to know.

**Gap analysis for a research proposal.** When responding to a brief, running this method on the client's own stated questions produces a proposal that costs less and demonstrates more judgement than one that prices what was asked for. Naming the questions their existing evidence already answers, and the ones a study cannot close, is the strongest available signal of expertise. It hands over to **01.03 Research Proposal Development**.

**Where the standard approach does not fit.** In genuinely new areas, almost every dependency is unaddressed and the register cannot discriminate. The productive move is to invert the exercise: instead of prioritising gaps, identify the two or three that are prerequisite to being able to ask any of the others, which are usually construct definition and population characterisation, and sequence rather than rank. For academic work, replace the decision-relevance criterion with theoretical contribution and keep the coverage test and the type classification, which transfer unchanged.

## 16. Skill chain

**Recommended previous skills:**
- **10.01 Literature Review and Desk Research.** Hands over the sufficiency table and the search log, without which no absence claim in the register is defensible.
- **10.02 Evidence Synthesis** and **10.03 Multi-Source Research Synthesis.** Hand over the claims that could not be established and the questions no source covered.
- **01.01 Research Brief Interrogation.** Hands over the decision, its timeline and its stakes.

**Recommended next skills:**
- **01.02 Business Problem to Research Question.** Takes each FILL specification and turns it into a formal research question set.
- **01.04 Research Method Selection.** Takes the specification with its precision requirement and designs the study that meets it.
- **01.03 Research Proposal Development,** where the prioritised register becomes a costed proposal.

**Runs well alongside:**
- **13.04 Bias Detection,** which checks whether the prioritisation accommodated a stakeholder preference.
- **12.03 Research Report Compilation,** which carries the register into the "what we could not establish" section of a deliverable.
- **14.03 Research Repository and Knowledge Curation,** where the register becomes a maintained asset.

---
A Yazi Supplied Skill and resource.