research-gap-finder

Find genuine research gaps in a topic and output a ranked, citation-backed gap report with a candidate research question per gap. Uses a stdlib-only CLI over key-free scholarly APIs (OpenAlex, Semantic Scholar, Crossref, Europe PMC, PubMed, arXiv) plus a guided human-in-the-loop layer (Litmaps, ResearchRabbit, Connected Papers, Elicit, SciSpace, Consensus, Scite, Inciteful, Dimensions). Grounded in gap frameworks (PICO/PICOS, six-type taxonomy, five-dimension importance rubric, AHRQ, Robinson, Arksey-O'Malley) with bibliometric mining and a zero-invented-citation honesty protocol.

Turn a topic into a ranked, citation-backed list of genuine research gaps, each with a candidate research question.

Two layers:

• Automated (CLI) — build an evidence matrix and classify/score/validate gaps via key-free APIs. Stdlib-only, Python 3.8+, no keys.
• Guided (human-in-the-loop) — browser tools (Litmaps, ResearchRabbit, Connected Papers, Elicit, SciSpace, Consensus, Scite, Inciteful, Dimensions) plus grants/patents/trials; use within their ToS and don't upload proprietary data.

Honesty guarantee: a gap is never credited to a source without a resolvable identifier (DOI/accession); unverifiable items are flagged Exploratory/Hypothetical.

When to use

Broad-topic or understudied-angle scanning for proposals, literature-review chapters, or "state of the field" briefs.

1 · Automated pipeline

bash

python3 scripts/research_gap_cli.py init     --dir proj --topic "T" [--pico '{"population":"...","outcome":"..."}']
python3 scripts/research_gap_cli.py search   --dir proj [--query "T"] [--engines openalex,semantic,crossref,europepmc,arxiv] [--limit 100] [--years 2018 2026]
python3 scripts/research_gap_cli.py extract  --dir proj [--cuefile cues.json]
python3 scripts/research_gap_cli.py classify --dir proj [--scores scores.json]
python3 scripts/research_gap_cli.py validate --dir proj [--check-web]
python3 scripts/research_gap_cli.py rank     --dir proj [--min-total 9] [--min-confidence medium] [--top N]
python3 scripts/research_gap_cli.py report   --dir proj [--top 8] [--out report.md]
python3 scripts/research_gap_cli.py status   --dir proj
python3 scripts/research_gap_cli.py selftest

Command	Result
init	config.json (topic, PICO, engines, years, limit, cache flags)
search	Queries chosen engines; dedupes by DOI/title; appends to evidence.json + evidence_matrix.csv; caches raw responses in proj/.cache/. A live API failure is skipped, never aborts. Fully offline → gaps labeled Exploratory/Hypothetical.
extract	Pulls gap-cue sentences (limitations/future-work/absence) into candidates; never fabricates one if none is found.
classify	Tags six-type taxonomy + five-dimension rubric scores (each 0–3, total /15). Baseline scores are marked estimated; override per gap via scores.json/--scores.
validate	Resolves sources via Crossref (or marks identifiers) and sets confidence: High = ≥2 independent sources & resolved; Medium = 1 source; Low = no resolvable source → Exploratory/Hypothetical.
rank/report	Sort by importance × confidence; emit the Markdown gap report.
status/selftest	Project state; offline self-test.

2 · Guided methodology (adds depth)

1. PICO/PICOS — define Population, Intervention, Comparison, Outcome, Setting; store in config.json.
2. Recent systematic reviews — pull 3–5 from PubMed/Cochrane/Scopus; combine with automated results.
3. Explicit gap cues — read each review's Limitations/Future Directions; log Source, Gap type, Evidence (CLI extract automates cue detection).
4. Citation networks — Litmaps (reference-gap method), ResearchRabbit/Connected Papers (citation "islands").
5. Semantic "what is unstudied?" — Elicit/Consensus questions over the topic + population.
6. Novelty cross-check — grants (NIH RePORTER, NSF, EU CORDIS), preprints, trials, patents; downgrade already-funded/published gaps.

3 · Classify & rank

Six-type gap taxonomy (CLI tags via cue lexicon):

Type	Signal
Evidence	insufficient/no/weak evidence, little is known, understudied
Methodological	weak design, small sample, no control, outdated instrument, reproducibility
Population	demographic/geographic/severity absent, rural, elderly, underrepresented
Contextual	setting/real-world/cultural omitted, low-resource, community
Theoretical	no explicit theory, conflicting explanations, mechanism, no consensus
Translational	not moved to practice/policy, implementation, uptake

Five-dimension importance rubric (0–3 each, max 15):

Dimension	Question
Theoretical	Advances core concepts / resolves contradictions?
Practical	Solves a real problem / informs policy-practice?
Feasibility	Data, methods, resources realistically obtainable?
Novelty	Genuinely unstudied (check preprints, trials, grants)?
Coherence	Logical next step of the field's trajectory?

13–15 proposal-ready; 9–12 needs feasibility work; ≤8 re-evaluate. Require Feasibility ≥2.

Identification ≠ importance. Filling the "Gap" cell ends identification; importance is judged afterwards. Many gaps are genuine but unimportant.

4 · Output report

Per gap: gap statement · type (+ secondary) · source evidence (DOI/resolvable ID) · five-dimension importance + total · confidence · candidate research question.

Honesty & anti-hallucination (non-negotiable)

1. Zero-invented citations — every gap traces to a resolvable source; otherwise it is Exploratory/Hypothetical (a hypothesis, never a citation-backed claim). The CLI enforces this by only emitting sources with a DOI/accession.
2. Separate identification from importance — run the rubric; never present a trivial absence as valuable.
3. Label confidence — AI-assisted findings unverified by a review cap at Medium.
4. No invented trends — if literature is sparse, say "insufficient literature to determine trends".
5. Absence ≠ algorithm failure — a missing paper may be a search artifact; cross-check with the AHRQ framework first.
6. Verify every citation you output — resolve DOIs/titles via Crossref (validate) before writing; never emit an unresolvable citation.

5 · Catalog

The full 100-resource catalog (frameworks, tools, databases, strategies, key papers) is in resources.md.

---

Authored and hardened with the repo's AI multi-model team; published under the Skill Publishing Standard.