Skip to content

Literature review protocol

Part of Inverse Innovation with the Encyclopedia of Abstractions · Literature review protocol · Last revised August 2026

Review type: Cross-disciplinary scoping review with a nearest-system and claim-boundary focus
Frozen before substantive searching: 2026-08-03
Primary cutoff: Sources discoverable through 2026-08-03

Review objective

Determine whether prior work has demonstrated a system materially comparable to the Encyclopedia of Abstractions inverse-innovation pipeline, identify the closest partial precedents, and establish language that accurately distinguishes the current work without claiming historical priority beyond the evidence.

Primary research question

Has prior research demonstrated a reproducible system that uses explicit, domain-general solution structures or mechanisms to formulate candidate problems in unrelated target domains, operationalize adapted solutions, scrutinize them for prior art and usefulness, and repeat the process across a systematic matrix?

Secondary questions

  1. What has cognitive science established about analogical retrieval, structural alignment, far transfer, and the difficulty of recognizing deep similarity?
  2. What computational systems perform design-by-analogy, case-based reasoning, conceptual blending, TRIZ-like transfer, biomimetic search, or patent-based technology transfer?
  3. What LLM research evaluates analogy, compositional generalization, cross-domain scientific ideation, problem finding, or solution transfer?
  4. Do existing systems work problem-to-solution, solution-to-problem, or both?
  5. Do they explicitly represent mechanisms and structural roles, or rely on surface retrieval and unconstrained generation?
  6. Do they incorporate prior-art search, independent criticism, negative cases, iterative revision, adopter evidence, falsifiable tests, or cost accounting?
  7. Has synthetic data or process supervision been used to teach analogical or cross-domain transfer, and how was genuine transfer distinguished from memorization?
  8. What terminology and baselines should the eventual report use?

Search clusters

A. Cognitive and computational analogy

  • analogical transfer; far transfer; remote analogy
  • structure mapping; structural alignment; relational similarity
  • analogical retrieval; analogical encoding; schema induction
  • case-based reasoning; derivational analogy; analogical problem solving
  • negative transfer; semantic distance; surface versus structural similarity

B. Design and computational creativity

  • design by analogy; analogical design; bio-inspired design; biomimetics
  • functional analogy; function-behavior-structure; SAPPhIRE
  • TRIZ; inventive principles; contradiction matrix
  • patent analogy; technology opportunity discovery; cross-domain patent search
  • conceptual blending; bisociation; computational creativity
  • problem finding; opportunity identification; reverse innovation search

C. LLM reasoning and discovery

  • large language model analogical reasoning and analogy benchmarks
  • LLM far transfer; cross-domain reasoning; relational reasoning
  • LLM scientific ideation; hypothesis generation; research agents
  • knowledge-graph-guided discovery; ontology-guided generation
  • automated invention; technology forecasting; innovation agents

D. Synthetic curricula and training

  • synthetic data for reasoning; self-distillation; process supervision
  • analogy curriculum; relational representation learning; contrastive structure
  • compositional generalization; domain generalization; systematicity
  • hard negatives; counterfactual data augmentation; label leakage
  • training versus evaluation on held-out domains and relation combinations

Source policy

Prioritize primary sources: original papers, proceedings, books or chapters that introduced a method, official project documentation, benchmark papers, and authoritative institutional reports. Use reviews and surveys to map terminology and citation networks, but verify central claims against original sources.

Include older foundational work when it defines a relevant theory or system. Include recent preprints when directly relevant, but label peer-review status and avoid treating recency as authority. Record corrections, retractions, or later versions when found.

Inclusion criteria

A source is included in the core evidence table if it contributes at least one of:

  • A theory or benchmark of cross-domain structural transfer.
  • A computational system that retrieves or adapts solutions across domains.
  • A system that generates problems, hypotheses, designs, inventions, or research opportunities using analogical or structured knowledge.
  • Evidence about LLM strengths or failures in analogical/far-transfer reasoning.
  • A training method plausibly relevant to learning abstraction-mediated transfer.
  • A close methodological precedent for systematic generation, scrutiny, or synthetic transfer curricula.

Sources that merely use “transfer” to mean ordinary transfer learning, domain adaptation, style transfer, or organizational technology transfer are excluded unless they address structural problem-solution transfer.

Evidence fields

For every core source, record:

  • Stable citation, year, venue/status, and URL/DOI.
  • Research field and task direction.
  • Input representation and knowledge source.
  • Whether structure, mechanism, function, or causal roles are explicit.
  • Source and target domain relationship.
  • Retrieval, mapping, adaptation, generation, and rejection capabilities.
  • Scale and evaluation design.
  • Human involvement.
  • Prior-art, novelty, feasibility, adopter, and empirical-test treatment.
  • Training use, if any.
  • Strongest demonstrated result.
  • Principal limitation.
  • Relationship to the Encyclopedia approach: foundational, adjacent, nearest partial precedent, possible close precedent, or contrast.
  • Confidence in the characterization.

Comparison dimensions for the nearest-system matrix

  1. Explicit reusable solution archetype.
  2. Explicit mechanisms or structural role mapping.
  3. Transfer across semantically distant domains.
  4. Solution-to-problem generation.
  5. Complete domain-specific proposal generation.
  6. Multiple distinct proposals per pairing.
  7. Systematic archetype-by-domain matrix.
  8. Independent criticism or adversarial review.
  9. External prior-art search.
  10. Practicality, cost, adopter, and impact assessment.
  11. Falsifiable empirical next step.
  12. Iterative repair with preserved trajectories.
  13. Negative examples and terminal classifications.
  14. Reproducible raw artifacts and resource accounting.
  15. Use as a synthetic training curriculum.

Claims policy

Absence of a discovered precedent is not proof that none exists. The integrated review will distinguish:

  • Established: directly demonstrated by authoritative sources.
  • Supported interpretation: synthesis consistent with multiple sources.
  • Provisional gap: no close precedent found within the scoped search.
  • Open: evidence incomplete or terminology too fragmented.

The final review must state its databases/search limitations, avoid “first” or “no one has done this” unless a defensible systematic basis exists, and provide the closest contrary cases even when they complicate the desired narrative.

Planned outputs

  • LITERATURE_REVIEW.md: accessible integrated synthesis.
  • SOURCE_INVENTORY.md: deduplicated annotated primary-source list.
  • NEAREST_SYSTEMS_MATRIX.md: feature comparison of the closest systems.
  • field_reviews/: specialist search memos retained as intermediate artifacts.
  • SEARCH_LOG.md: query families, search dates, and material gaps.