GSLHub — Generative Search Lab Hub is an independent research initiative and open technological infrastructure focused on understanding how generative AI systems discover, select, cite and recommend information.
Our work combines Generative Search research, Generative Engine Optimization (GEO), AI evaluation, governed evidence, reproducible experimentation and software engineering.
The objective is not only to study generative systems, but to build the infrastructure required to make that research auditable, repeatable and useful in real-world environments.
Scientific problem
→ Hypothesis
→ Experiment
→ Controlled execution
→ Preserved research artifact
→ Evidence
→ Observation
→ Citation / Metric
→ Reproducibility review
→ Public dissemination
GSLHub treats metrics as traceable research outputs rather than isolated numbers. Scientific methodology lives in gslhub/research, benchmark specifications in gslhub/benchmarks, reusable implementations in gslhub/software, and the governed operational platform in gslhub/website.
Public website and core research platform for experiment governance, controlled executions, evidence provenance, scientific metrics, research artifacts and reproducibility controls.
Stack: Next.js · TypeScript · React · Payload CMS · MongoDB · Tailwind CSS · Node.js · GitHub Actions
License: AGPL-3.0-only
Canonical methodological layer containing the research model, protocols, codebooks, governance, reproducibility requirements and citation metadata.
License: CC BY 4.0 for original research documentation unless otherwise stated.
Reproducible benchmark and metric specifications for Generative Search and GEO. The current public baseline includes AIR, CR, MCP and RCR, a machine-readable benchmark definition and synthetic validation fixtures.
License: CC BY 4.0 for original benchmark specifications and documentation.
Reusable research software implementing independently testable parts of the GSLHub methodology. The first package, @gslhub/metrics-core v0.1.0, provides framework-independent deterministic AIR, CR, MCP and RCR calculations with exclusions, numerator/denominator data and SHA-256 audit checksums.
Stack: TypeScript · Node.js · npm workspaces · GitHub Actions
License: AGPL-3.0-only
Cross-project public technical and institutional documentation covering architecture, repository boundaries, governance and safe-publication standards.
License: CC BY 4.0 unless otherwise stated.
Approved GSLHub visual identity, light/dark logo variants, icon, palette and brand-usage guidance. Brand and trademark rights are handled separately from software and research-documentation licenses.
| Area | Status | Purpose |
|---|---|---|
website |
Public | Public site and core research platform |
research |
Public | Canonical protocols, methodology and codebooks |
benchmarks |
Public | Evaluation frameworks and reproducible metric specifications |
software |
Public | Reusable research software and deterministic metric implementations |
docs |
Public | Technical and institutional documentation |
branding |
Public | Approved visual identity and usage guidance |
| datasets | Preparing | Reviewed dataset releases with per-release licensing |
Repositories are opened progressively only when their contents are documented, licensed and ready for public reuse.
| Code | Metric | Primary question |
|---|---|---|
| AIR | Answer Inclusion Rate | How often is the evaluated target visibly included? |
| CR | Citation Rate | How often is the evaluated target explicitly cited? |
| MCP | Mean Citation Position | When cited, how early does the target appear? |
| RCR | Response Consistency Rate | How stable are controlled repetitions against a frozen baseline? |
The normative specifications are versioned in gslhub/benchmarks, while reusable deterministic implementations are versioned independently in gslhub/software. This separation allows calculations to be tested against a frozen specification without coupling them to the application database or CMS.
GSLHub uses licenses by output type:
- Research platform and original software: GNU AGPL-3.0-only unless a package states otherwise.
- Original research and benchmark documentation: Creative Commons Attribution 4.0 International (CC BY 4.0), unless otherwise stated.
- Datasets: licensed individually according to provenance, rights and research constraints.
- Publications: governed by their individual publication or publisher terms.
- Brand assets and trademarks: governed separately.
Transparent — methodology and technical decisions should be inspectable.
Reproducible — experiments should preserve enough context to be repeated.
Evidence-driven — conclusions should remain traceable to preserved evidence.
Technically rigorous — research infrastructure should be engineered with production-level care.
Open where possible — software, methods and outputs should be reusable whenever legal, ethical and methodological constraints allow it.
GSLHub is open to collaboration with researchers, developers, universities, AI practitioners and organizations interested in Generative Search, GEO, AI evaluation and reproducible research.
Website: gslhub.com
Research: github.com/gslhub/research
Benchmarks: github.com/gslhub/benchmarks
Software: github.com/gslhub/software
Email: research@gslhub.com
Open research · Reproducible evidence · Applied AI