research_gap_evaluation v1.0.0

name: "research_gap_evaluation" description: "Decide whether a research gap is worth a thesis, dissertation, or funded project before committing months to it: state the gap in one sentence and classify its type, run a kill search that tries to prove the gap is already closed, test whether closing it is a real contribution rather than a permutation, test whether it can be closed in the time available with the data, skills, ethics route, and supervision at hand, and package the result as a decision dossier a supervisor or panel can challenge. Use when a candidate gap, thesis topic, or project idea needs a go, shrink, or no-go decision, when a supervisor asks why this gap matters, or when several candidate gaps need ranking. For a pure has-this-been-done check on an idea, use research-novelty-check instead." metadata: { "includeInPrompt": true }

Research Gap Evaluation

Purpose

Give one candidate gap a defensible decision: go, go smaller, park, or drop. A gap being empty is not enough. Empty ground can be empty because nobody bothered, because somebody already built on it under different words, or because it cannot be built on with the resources at hand.

This skill runs three gates in order. A gap must pass all three. Averaging is banned: a brilliant contribution that cannot be finished in time is a no, and an easy project that adds nothing is a no.

Neighbouring jobs are different. Checking whether one idea has been done before is research-novelty-check. Writing the literature review that presents the gap is a separate writing task. This skill sits between them: it uses a search like a novelty check, but the output is a commitment decision, not a verdict on novelty alone.

Workflow

1. State the gap in one sentence, or stop. Use this shape: "Prior work shows [what is known], but [specific missing piece] remains unknown, so we cannot yet [decision, explanation, or action the missing piece blocks]." If the sentence cannot name what the missing piece blocks, there is no gap yet, only a topic. Send it back for narrowing.

1. Classify the gap type. Name one primary type from the Miles (2017) taxonomy: evidence, knowledge, practical-knowledge conflict, methodological, empirical, theoretical, or population. Full definitions and the test for each type are in references/gap-types-and-tests.md. The type matters because it sets what closing evidence looks like. A population gap closes with data from that population. An evidence gap, where findings contradict each other, closes by explaining the contradiction, not by adding one more small study to one side.

1. Gate 1: Run the kill search. Do not search for support for the gap. Search for the work that closes it.

• Write the closing study in one sentence: the study that, if it existed, would fill the gap.
• Search for that study directly in at least two indexes with different coverage, using its concepts and synonyms, sorted by date so recent closers surface first.
• Forward-search from the papers the gap claim comes from. Anyone who closed the gap would likely cite them, a verification step from Muller-Bloch and Kranz's framework for identifying gaps in literature reviews.
• Check where unfinished work lives: trial and study registries and PROSPERO for reviews in health fields, OSF Registries in other fields, the preprint server the field uses, and grey literature (theses, conference proceedings, agency reports).
• Log the search the way PRISMA 2020 reporting asks: source, date, full string, limits, hits, kept. A gap claim with no log is an opinion.
• Verdict for the gate: pass if no closer and no credible in-progress closer turned up, conditional if a partial closer narrows the gap (restate the smaller gap and rerun step 1), fail if a closer exists. Record the closer, do not argue with it.

1. Gate 2: Test contribution against permutation. Ask what changes if the gap closes.

• Name the reader whose mind, model, or practice changes: a theory gets revised or bounded, a contradiction gets explained, a practice decision gets an answer, a method gets a check it lacked. "More literature on X" is not a change.
• Run the permutation traps. Each of these alone is a permutation, not a contribution, unless the dossier gives the reason it matters here: same study in a new country, same model with one added variable, same method on a new small sample, same question with a newer tool. Context work can be a real contribution where findings conflict or evidence from that setting is genuinely scarce and decisions depend on it. The burden is to say why, in writing.
• Run the problematization check from Sandberg and Alvesson (2011). Gap-spotting that leaves every assumption in the literature untouched tends to produce small, safe additions. Ask: does this project challenge or test at least one assumption the field is relying on, or does it only fill a cell in a table the field already drew? Only filling a cell can still pass for a bounded masters or professional project, but label it honestly as an incremental contribution and do not pitch it as more.
• Score the FINER interest, novelty, and relevance points here (Hulley et al.): interesting to someone beyond the author, novel against the Gate 1 log, relevant to a named decision or debate.

1. Gate 3: Test whether it can actually be closed. A gap passes only with evidence, not optimism.

• Data: name the dataset, sample, archive, or field site, and the state of access today (in hand, applied for, not yet asked). Not yet asked counts as a risk with a date to resolve it, not as access.
• Time: work backward from the hard deadline. Put data collection, analysis, and writing on a line. If closing needs a longitudinal window, an ethics wait, or a recruitment rate the calendar cannot hold, the gate fails at this scope.
• Ethics and permissions: name the approval route and its typical local lead time if the institution publishes one. If no route exists for the design as stated, shrink the design or fail the gate.
• Skills and supervision: name who covers the method. A gap that needs a method nobody on the team can run or supervise is a training project first and a thesis second.
• Money and tools: list the paid items and confirm a source for each. Unfunded essentials fail the gate.
• Build the shrink ladder before deciding: full scope, reduced scope that still closes a smaller gap, and minimum publishable or passable unit. Many Gate 3 fails are go decisions at the middle rung.

1. Score the gates without averaging. Use scripts/gap_score.py with one verdict per gate (pass, conditional, fail) plus the evidence line for each. The rule is mechanical: any fail means no-go at this scope, any conditional means go only at the reduced scope named in the dossier, three passes mean go. Run it and paste its output into the dossier so the decision rule is visible, not just the conclusion.

python3 scripts/gap_score.py --gap "one sentence gap" --gate1 pass --gate2 conditional --gate3 pass --evidence1 "kill search log, 2 indexes, no closer" --evidence2 "contribution is incremental, context reason stated" --evidence3 "data in hand, 14 months available"

1. Rank candidates if there is more than one. Never rank by total score. Rank by weakest gate first: a candidate with three passes beats one with two passes and a conditional, which beats any with a fail. Break ties by Gate 2, because a finishable project that adds little is the most common waste.

1. Write the decision dossier. Use templates/dossier-template.md. Fill every section, including the strongest argument against the project and what would change the decision. A dossier with no case against is advocacy, not evaluation. Use templates/gate-worksheet.md during the work and keep it with the dossier.

1. Hand it to a challenger before committing. The dossier is built for a supervisor, panel member, or peer to attack: the kill search log is reproducible, the contribution claim is one paragraph, the feasibility claims name dates and access states. Record the challenger's objections and the responses in the dossier. If nobody is available to challenge it, say so in the dossier and treat the decision as provisional.

Output Contract

• One decision, in the first line: go, go at reduced scope, park with a named reopening condition, or drop.
• The gap sentence and its Miles type.
• Gate table: one row per gate, verdict, and the single strongest evidence line for that verdict.
• The kill search log for Gate 1: sources, dates, strings, hits, kept, registries and preprint servers checked.
• The contribution paragraph for Gate 2: what changes, for whom, and the permutation or assumption check result.
• The feasibility evidence for Gate 3: data access state, backward timeline, ethics route, skills cover, funded essentials, and the shrink ladder.
• The dossier file, completed from the template, with the case against and the challenger record.
• For a no-go: the restated smaller gap that might pass, or a plain statement that none was found. Do not soften a no-go into a maybe.

Operating Rules

1. Every factual claim about the literature traces to the Gate 1 log or a named source the evaluator read. Never invent papers, findings, dates, or counts to make a gap look empty or full. Unverifiable means the gate that rests on it is conditional at best.
2. A fail on any gate stops the project at that scope. Do not let high marks on other gates buy out a fail.
3. Absence in one index is not absence in the field. Gate 1 needs at least two indexes with different coverage plus registries, preprints, or grey literature where the field uses them.
4. Replication and context studies are not automatic fails and not automatic passes. Judge them on the written reason the context or the repeat changes what can be concluded.
5. Feasibility claims state their evidence and date. "Data should be available" is not evidence. In hand, permission granted, or application submitted on a stated date are evidence.
6. Keep the level honest. A doctoral thesis usually must show a substantial original contribution under the institution's own regulations. A masters or course project may pass Gate 2 as a bounded incremental contribution. State which standard was applied.
7. Date the decision. Fields move. A go decision older than the project's own start delay deserves a fresh Gate 1 search before work begins.
8. Style: plain words, no em dashes, no promotional language about the candidate project. The dossier argues from records, not adjectives.