Nearest Systems Matrix

Part of Inverse Innovation with the Encyclopedia of Abstractions · Nearest Systems Matrix · Last revised August 2026

Review date: 2026-08-03
Purpose: Compare the Encyclopedia of Abstractions (EoA) experiments with the nearest systems found across cognitive analogy, engineering design, technology-opportunity discovery, computational creativity, and automated science. This is a functional comparison, not a claim of historical priority.

How to read the matrix

• D — demonstrated as an explicit part of the reported system or study.
• P — partial, restricted, indirect, or human-mediated demonstration.
• N — not demonstrated in the inspected report.
• NR — not reported or could not be determined from the inspected source.

The coding asks whether the published workflow demonstrates a feature, not whether the authors could add it or whether an LLM might do it implicitly. Different rows also have very different evidence levels: a cognitive model, a design-support interface, a patent-mining prototype, and a wet-lab research agent are not interchangeable accomplishments.

A. Search direction and representation

System	Starts with reusable solution	Generates or identifies target problem/application	Explicit roles, relations, function, or mechanism	Crosses distant domains	Declared archetype × domain matrix	Multiple complete candidates per declared cell
EoA Experiments 1–6	D	D	D	D	D	D
AskNatureGPT	D	D	P	P	N	N
AskNatureNet	D	D	D	P	N	N
Solution-driven BID / SR.BID	D	D	D	P	N	N
Function-based technology-opportunity discovery, Yoon et al.	D	P	D	D	N	N
Technology-function network, Qiao et al.	D	P	D	P	N	N
InnoGPS	D	P	P	D	N	N
DANE / Idea-Inspire	P	N	D	P	N	N
SOLVENT / Kang et al.	N	N	D	D	N	N
ViMimic	P	P	D	D	N	P
Shen, Druckmann, and Zou	N	N	D	D	N	D
MOOSE-Star	N	N	P	P	N	P
SciMON / Scideator	N	N	P	P	N	D
AI co-scientist	N	N	N	P	N	D
AutoTRIZ	N	N	D	P	N	D
Deliberate exploratory search, Reinsberger et al.	P	D	N	P	N	P

Interpretation

AskNatureGPT, AskNatureNet, solution-driven bio-inspired design, and the patent opportunity-discovery lineage invalidate any claim that starting with a solution and looking for applications or problems is historically new. Shen, Druckmann, and Zou invalidate a claim that explicit LLM-mediated cross-domain solution transfer is new. The feature not demonstrated by those systems is the declared, auditable crossing of reusable solution archetypes with a broad domain set while retaining a fixed number of complete attempts for each cell.

That matrix feature should not be treated as intrinsically superior. It buys coverage accounting and a measurable denominator at substantial computational cost. Experiment 7 should test whether selective scrutiny can retain most of the yield without researching every candidate.

B. Scrutiny, repair, and evidence

System	Operational proposal or concept	Output-specific prior-art search	Critic separate from proposer	Critique–revision loop	Need, adopter, or usefulness evidence	Feasibility, deployability, or cost
EoA Experiments 1–6	D	D	D	D	P	P
AskNatureGPT	D	N	P	N	P	P
AskNatureNet	N	N	N	N	N	N
Solution-driven BID / SR.BID	D	N	P	P	P	P
Yoon et al.	P	N	N	N	N	N
Qiao et al.	P	N	N	N	N	N
InnoGPS	P	N	N	N	P	N
DANE / Idea-Inspire	D	N	N	P	P	P
SOLVENT / Kang et al.	P	N	P	N	P	N
ViMimic	D	N	P	P	P	P
Shen, Druckmann, and Zou	D	D	P	N	P	P
MOOSE-Star	D	P	P	P	N	N
SciMON / Scideator	D	D	P	P	P	P
AI co-scientist	D	D	D	D	P	P
AutoTRIZ	D	N	P	P	P	P
Reinsberger et al.	D	P	D	D	D	D

Interpretation

No individual scrutiny component is unprecedented. Automated science systems already combine literature search, generation, ranking, critique, and revision. Case-based reasoning and derivational analogy have long retained and repaired failed solution paths. Real organizational exploratory search supplies much stronger adopter and implementation grounding than the EoA experiments.

The EoA distinction is the application of a comparatively rich scrutiny bundle to solution-first, cross-domain target-problem proposals. Even there, the current evidence is bounded: web prior-art searches can miss terminological neighbors; model critics share blind spots; adopter pull was inferred rather than observed; cost estimates were preliminary; and most candidates were not implemented.

C. Validation, negative evidence, and reproducibility

System	Real implementation or empirical test	Preserves rejected/negative trajectories	Reports a falsifiable next test	Releases inspectable artifacts/data	Full cell/failure denominator	Resource or compute accounting
EoA Experiments 1–6	P	D	D	D	D	D
AskNatureGPT	P	N	P	P	P	N
AskNatureNet	N	N	N	P	D	N
Solution-driven BID / SR.BID	P	N	P	P	P	N
Yoon et al.	P	N	N	P	P	N
Qiao et al.	P	N	N	P	P	N
InnoGPS	P	N	N	D	P	N
DANE / Idea-Inspire	P	N	P	P	P	N
SOLVENT / Kang et al.	D	P	N	D	D	N
ViMimic	D	P	P	P	D	N
Shen, Druckmann, and Zou	D	P	D	D	D	P
MOOSE-Star	P	D	P	D	D	P
SciMON / Scideator	P	P	P	D	D	P
AI co-scientist	D	P	D	P	P	P
AutoTRIZ	P	P	P	D	D	P
Reinsberger et al.	D	D	P	P	D	N

Important coding qualifications

• EoA empirical validation is partial. “Strict” and “empirical-partner” are pipeline dispositions, not evidence that a proposal works in the world.
• AskNatureGPT implementation is partial. It reports generated concepts, case studies, classifier evaluation, and human ratings, not broad field deployment.
• Shen et al. is unusually strong for a preprint. It reports four biomedical implementations with quantitative results, but only after receiving the target research problems and selecting a tiny subset for implementation.
• MOOSE-Star reconstructs research trajectories from papers. Its negative sampling and temporal evaluation are valuable, but retrospective paper decomposition is not a faithful record of how the discovery actually arose.
• AI co-scientist has real experimental evidence in selected cases. It does not isolate analogy or the solution-to-problem direction.
• Reinsberger et al. has the strongest real organizational grounding in this table. It studies deliberate search for need–solution pairs through 306 interviews and 89 proposals across four projects, but it is a human process, not a scalable computational matrix.

What the matrix does—and does not—support

The inspected literature supports the following scoped statement:

We did not locate a reported system that combines a curated, domain-general ontology of solution structures with a predeclared solution-archetype × domain matrix, multiple complete proposals per cell, output-specific prior-art research, independent criticism, practical/opportunity and empirical-partner gates, iterative repair, falsifiable next tests, retained negative trajectories, and resource accounting.

This is a provisional integration gap, not a priority claim. It could be closed by an overlooked paper, dissertation, patent, commercial system, or later publication. More importantly, uniqueness of combination does not establish scientific importance. The program's value must be shown through comparisons: whether the ontology and matrix increase valid or useful transfer; whether the scrutiny reduces false positives; whether later proposals add enough yield to justify their cost; and whether external domain readers confirm that selected problems are real and worth pursuing.

Sources and scope

This matrix synthesizes the three field reviews in this directory. Detailed evidence and complete bibliographies are in:

• Analogical transfer and structural alignment
• Design-by-analogy and computational creativity
• LLM discovery and synthetic transfer training