Overview
Every article published on Stabilarity Research Hub undergoes automated reference quality evaluation. The system analyzes all outbound links and citations, classifies them against a database of 900+ academic sources, and computes a composite Trust Score that reflects the overall reliability of an article’s evidence base.
Trust Score
The Trust Score is a weighted composite metric (0-100) computed from six reference quality indicators. Each indicator measures what percentage of an article’s references meet a specific quality criterion.
| Badge | Metric | Weight | Description | Threshold |
|---|---|---|---|---|
| [t] | Trusted Sources | 25% | References pointing to sources verified as high-quality in our Sources Database (900+ classified sources) | 80% |
| [a] | DOI Coverage | 20% | References with a Digital Object Identifier, ensuring persistent citability and discoverability | 80% |
| [s] | Reviewed Sources | 20% | References from editorially reviewed journals, conferences, or curated academic repositories | 80% |
| [i] | Indexed | 15% | References with metadata indexed in academic databases (CrossRef, OpenAlex, Semantic Scholar) | 80% |
| [b] | CrossRef | 10% | References registered and discoverable via the CrossRef API | 80% |
| [l] | Academic Sources | 10% | References from journals, conference proceedings, or preprint servers (vs blogs, news sites, etc) | 80% |
Formula
The Trust Score is computed as:
Score = Sum(metric_percentage * weight) / Sum(weights) + min(ref_count, 10) * 0.5Where each metric percentage represents the fraction of references meeting that criterion. A small bonus (up to 5 points) is added for reference count, rewarding articles with broader evidence bases. The final score is capped at 100.
Score Labels
| Range | Label | Interpretation |
|---|---|---|
| 90-100 | Excellent | Outstanding reference quality across all dimensions |
| 75-89 | Strong | High-quality references with broad academic coverage |
| 60-74 | Good | Solid reference base with room for improvement |
| 40-59 | Fair | Acceptable references but gaps in coverage or verification |
| 0-39 | Developing | Early-stage article; references need enrichment |
Badge System
Each article displays quality badges — single-character monospace labels that indicate which quality thresholds the article meets. A badge appears when 80% or more of the article’s references satisfy that criterion.
Badge colors follow two groups:
- Grey (#555) — Academic/quality indicators: [s], [t], [a], [b], [i], [l]
- Green (#2e7d32) — Achieved badges (threshold met)
- Gradient — Badges approaching threshold show proportional darkness
Target vs Actual
Each article has a target badge set (default: stabil = all six academic badges). The quality bar on each article shows both target badges and achieved badges, making progress visible.
Verification
Articles that meet their badge targets and pass editorial review receive the [V] Verified mark on their title. Verification requires:
- All target badges achieved (80%+ on each targeted metric)
- Approval from designated reviewer(s)
- No outstanding quality issues
Verified articles display a green Trust Score badge and green [V] mark.
Source Classification
The evaluation system is backed by a database of 900+ classified reference sources across 16 types:
| Type | Examples | Trust Level |
|---|---|---|
| Journal | Nature, Science, IEEE Trans. | High |
| Preprint | arXiv, bioRxiv, medRxiv | Medium |
| Conference | NeurIPS, ICML, ACL | High |
| Repository | Zenodo, Figshare, Dryad | Medium |
| Government | WHO, FDA, Eurostat | High |
| Encyclopedia | Wikipedia, Stanford Encyclopedia | Medium |
| News | MIT Tech Review, Ars Technica | Low |
| Blog | Personal/corporate blogs | Low |
Each source has flags for trust level, editorial review status, indexing, open access, and pricing — enabling granular quality assessment at the source level.
Evaluation Process
When an article is published or updated, the system:
- Extracts all outbound URLs from article content
- Registers unknown URLs in the reference database with domain, hash, and metadata
- Resolves sources — maps each reference to its source in the Sources Database (auto-creates if missing)
- Validates — checks HTTP status, fetches content, extracts DOI metadata via CrossRef/OpenAlex/Semantic Scholar
- Flags — propagates source-level trust/review/index flags to each reference
- Computes — calculates badge percentages and Trust Score
- Caches — stores results as post meta for fast rendering
Reference Trust Analyzer Tool
Try our public tool to evaluate the reference quality of any article or paper by URL:
Reference Trust Analyzer — paste any article URL and get a full trust report.