AI hallucinates.
Gaptrium™ counts.
Every AI tool in your research workflow introduces a risk that systematic review methodology cannot accept: the risk of a finding that does not exist in your sources.
The three problems
AI creates for researchers.
AI tools are useful for many things. Systematic literature review, gap analysis, and corpus analysis are not among them — not because the technology is bad, but because the methodology requires properties that probabilistic models cannot provide.
✗ Problem 1 — Hallucination
Large language models predict plausible text. In literature review, plausible means citations, findings, and author names that sound correct but do not exist. A single hallucinated citation in a systematic review invalidates the entire analysis for peer review.
✗ Problem 2 — Irreproducibility
Run the same query twice on the same AI tool and you will get different results. Systematic reviews require that any researcher with the same corpus and the same protocol reaches the same conclusion. Probabilistic models are structurally incompatible with this requirement.
✗ Problem 3 — Architectural limits
AI platforms process 5–20 documents per session, typically capped at 10–25 MB per file. These are not pricing tiers you can upgrade through — they are server-side architectural constraints. A literature review of 200 papers cannot be run in a single AI session. Period.
The comparison
that matters.
This is not a feature list comparison. It is a methodological requirements comparison — the criteria a systematic review protocol must satisfy, and whether each tool category satisfies them. Three categories are commonly used in literature analysis today; they are not interchangeable.
| Requirement | AI / LLM tools | Bibliometric tools | Gaptrium™ |
|---|---|---|---|
| Reproducibility — same corpus, same result | ✗ Probabilistic — varies between runs | High — dataset-dependent | ✔ Deterministic — byte-identical output |
| No hallucinated findings or citations | ✗ Structural risk — LLMs predict plausible text | Not applicable — citation data only | ✔ Only counts what is in your documents |
| Process 100+ documents in one analysis | ✗ 5–20 files per session (architectural limit) | Varies by database coverage | ✔ Scope: 100 · Signal: 500 · Core: unlimited |
| Documents never leave your device | ✗ Every file uploaded to external servers | Local / hybrid, depending on tool | ✔ Local processing — zero server exposure |
| Suitable for privileged / classified material | ✗ Data retention and jurisdiction risks | Depends on hosting model | ✔ Zero capture — compatible by architecture |
| Every finding traceable to source document | ✗ Black-box reasoning — not auditable | Partial — traces to citation, not content | ✔ Every count traceable to exact document |
| No file size limit | ✗ Typically 10–25 MB per file | Not applicable — no file processing | ✔ Bounded only by your device memory |
| Research gap detection | ✗ Not a core function — user must direct it | Partial / indirect — via citation gaps | ✔ Core functionality |
| Peer-reviewed methodology | ✗ Proprietary, undisclosed, or general-purpose | Varies by tool | ✔ ECKM 2021 · IEEE 2023 · Cited in Springer Nature 2026 |
AI tool and bibliometric tool figures based on publicly documented category characteristics as of 2026. Lacuna Labs does not name or endorse any specific competitor.
What deterministic
actually means.
The word is used loosely. In the context of literature analysis, it means something precise: given the same input, the algorithm always produces the same output.
Gaptrium™ builds a term frequency matrix across your corpus. Each cell in that matrix is a count — the exact number of times term X appears in document Y. There is no probability involved. There is no model making a prediction. There is only counting.
The consequence is that any researcher with the same corpus of documents, the same set of analytical terms, and the same Gaptrium™ configuration will produce exactly the same Heat Map, Gap Analysis, and Cluster output as any other researcher — on any device, in any country, at any point in time.
This is not how AI works. Language models are trained on data, and their outputs reflect statistical patterns in that training data — not the content of your specific documents. When an AI tool summarises your literature, it is blending what is in your documents with what it learned from billions of other documents. That blend is not auditable, not traceable, and not reproducible.
✔ Every finding is a count
A Gap Analysis result showing that "quantum entanglement" appears in 87% of papers with 0% patent coverage is a count you can verify by hand. No model. No inference. No hallucination possible.
✔ Every count traces to a document
Click any cell in the Heat Map and see which document it comes from, which passage contains the term, and how many times it appears. Every result has a full audit trail.
✔ No training data contamination
Gaptrium™ has no training data. It knows nothing about your topic before you load your documents. Its output reflects only the documents you provide — nothing else.
Peer-reviewed
provenance.
The methodology underlying Gaptrium™ — the MTTR framework developed at London Metropolitan University — has been peer-reviewed and published in two institutional contexts over five years, and is cited in a third.
IEEE Transactions on Services Computing · 2023
Applied as the methodology for trust assessment in cloud governance — a peer-reviewed application of the deterministic cross-reference matrix in a high-stakes institutional context.
Read the paper →London Metropolitan University · Doctoral Defence · 2021
The MTTR methodology was formalised and peer-reviewed as part of a doctoral defence at the Intelligent Systems Research Centre — the academic institution where the method was originally developed.
Read the thesis →Springer Nature · Systematic Reviews · 2026
The original 2021 MTTR paper is cited as one of hundreds of tools and methods referenced in a scoping review of AI tools and platforms for evidence synthesis. The citation reflects bibliographic coverage, not an evaluation of Gaptrium™ by the review's authors.
Read the paper →Gaptrium™'s deterministic methodology has a peer-reviewed academic record spanning two independent institutional contexts and five years — and continues to appear in the literature on evidence synthesis tooling.
A different category of research infrastructure
Unlike generative AI systems optimised for language generation and conversational assistance, Gaptrium™ is designed for:
- Structured scientific evidence processing
- Reproducible analytical pipelines
- Reduction of interpretative ambiguity in literature reviews
- Deterministic workflow execution
This positions Gaptrium™ within the category of deterministic research infrastructure for evidence-based science — adjacent to, but functionally distinct from, both generative AI assistants and traditional bibliometric tools.
References to academic literature relate to the scientific methodologies underlying the system and/or publications where related approaches are discussed in peer-reviewed research. This does not imply endorsement, certification, or validation by any academic publisher, journal, or institution.
When AI is fine.
When it isn't.
This is not an anti-AI argument. AI tools are genuinely useful for many research tasks. The question is which tasks.
AI is appropriate for: generating a first draft of a section, summarising a single paper you have already read, translating a document, suggesting search terms for your protocol, formatting citations.
AI is not appropriate for: systematic gap analysis across a corpus, any finding that will appear in a methods section, any analysis that must be reproducible by an independent researcher, any corpus containing legally privileged or confidential documents, any analysis where you cannot tolerate a hallucinated citation.
The distinction matters because researchers are currently using AI for the second category — not because it is appropriate, but because no good alternative existed. Gaptrium™ exists to be that alternative.
Frequently asked questions
Can I use ChatGPT for systematic literature reviews?
ChatGPT and similar LLMs can summarise individual documents, but they cannot reliably perform systematic gap analysis across a large corpus. They hallucinate citations, cannot process more than 5–20 documents per session, and produce different results each time you run the same query — making them unsuitable for peer-reviewed systematic reviews that require reproducibility.
What does AI hallucination mean for academic research?
AI hallucination in academic research means the tool invents citations, paper titles, author names, or findings that do not exist. In a systematic review, a hallucinated citation cannot be verified and invalidates the analysis. Deterministic tools like Gaptrium™ only report what is present in the documents you provide — they cannot invent findings because they do not predict text. They count it.
What is a deterministic analysis tool?
A deterministic analysis tool produces the same result every time it processes the same input. In literature analysis, this means the same corpus with the same terms always produces the same Heat Map, Gap Analysis, or Cluster output — on any device, by any researcher, at any point in time. This reproducibility is a methodological requirement for peer-reviewed systematic reviews, where independent verification must be possible.
Why can't AI tools process my entire literature corpus?
AI platforms have hard architectural limits — typically 5 to 20 files per session, 10 to 25 MB per file. These are not pricing limits; they are server-side processing constraints that cannot be removed regardless of which plan you pay for. Gaptrium™ processes your documents locally in your browser using your own device's memory, so the only limit is your hardware. Scope handles 100 documents. Signal handles 500. Core is unlimited.
Is Gaptrium™ validated in peer-reviewed research?
The methodology underlying Gaptrium™ — the MTTR framework developed at London Metropolitan University — was peer-reviewed and published at ECKM 2021, defended in a doctoral thesis at London Metropolitan University, and applied in IEEE Transactions on Services Computing (2023). It is also cited, among hundreds of other tools, in a 2026 scoping review of AI tools for evidence synthesis published in Systematic Reviews (Springer Nature).
What is the difference between AI document analysis and deterministic corpus analysis?
AI document analysis uses probabilistic language models that predict likely outputs — results vary between runs, citations may be invented, and the reasoning process is not traceable to specific passages in your documents. Deterministic corpus analysis counts exact term frequencies across your documents — every result is a number you can verify by counting manually, and every result is traceable to the specific document and the specific passage where the term appears.
Does Gaptrium™ use any AI internally?
The core analytical engine — Heat Map, Gap Analysis, Bridge Map, Clusters, Timeline — is entirely deterministic and uses no AI. Gaptrium™ does include an optional AI Advisor feature on Scope and above, which uses a language model to suggest cluster labels and analytical term sets. This feature is clearly labelled as AI-assisted, is entirely optional, and does not affect the deterministic analysis results.