How We Evaluate Evidence

Published: June 2026

Transparency about how we gather, evaluate, and present information is central to our mission. This page explains the standards behind every article, tool, and breed profile on Guardians.dog.

Source Hierarchy

Not all sources carry equal weight. We evaluate evidence using the following hierarchy, from strongest to weakest:

Tier Source Type Examples How We Use It
1 Peer-reviewed research Published veterinary, behavioral, genetic studies with DOI Primary authority for health, biomechanics, and genetics claims
2 Government & extension sources USDA, state wildlife agencies, university extension bulletins Predator data, population statistics, regulatory context
3 Documented field reports Verified accounts from working ranchers, shepherds, and breed professionals Real-world validation of lab findings; behavioral data that doesn't exist in studies
4 Breeder & breed organization data Breed registries, kennel clubs, historical breed standards Breed classification, conformation, and lineage data — cross-checked against field evidence
5 Historical & cultural sources Archaeological records, indigenous oral histories, cultural documentation Heritage and classification context — always labeled as historical evidence
6 Anecdotal & informal sources Social media, forums, unverified owner accounts Not used as primary evidence. May inform research questions but never treated as authoritative.
Important distinction: Guardians.dog uses peer-reviewed sources. The site's own articles are not themselves peer-reviewed academic publications. We are an educational resource that references, synthesizes, and explains research — not a journal.

What Confidence Scores Mean

Our breed ontology includes a confidence score for key data points. This score reflects our editorial assessment of how well-supported a particular claim or trait value is:

  • High confidence (80–100%): Supported by multiple independent sources, including peer-reviewed research or extensive verified field data.
  • Moderate confidence (50–79%): Supported by credible sources but with limited independent verification, or where sources partially disagree.
  • Low confidence (below 50%): Based on limited evidence, single sources, or extrapolation from related breeds. Flagged accordingly.

Confidence scores are internal editorial assessments, not statistical probabilities. They help us be transparent about what we know well and what remains uncertain.

Verification States

Every data point in our breed database also carries a verification state:

  • Unreviewed: Data exists but no human editor has validated it yet.
  • Reviewed: A human editor has examined the data and believes it is reasonable.
  • Verified: Confirmed against multiple independent sources.
  • Contested: Active disagreement exists among sources. The claim is flagged and presented with context about the dispute.
  • Deprecated: Superseded by newer evidence. Retained for audit trail but not shown as current.

How We Handle Disputed Claims

Livestock guardian dog research is not a settled field. Breed classification, behavioral traits, and health data can be genuinely disputed among experts. When we encounter conflicting evidence:

  1. We flag the data point as "contested" in our ontology.
  2. We present the strongest available evidence from each side.
  3. We do not pick a winner unless the evidence is clearly one-sided.
  4. We note the nature of the disagreement (e.g., "breed registries classify this differently than genetic studies").
  5. We revisit contested claims when new evidence emerges.

Field Experience vs. Academic Evidence

LGD work sits at the intersection of science and practice. Academic studies provide controlled data, but they cannot replicate the full complexity of a working livestock operation. We value both:

  • Academic evidence takes precedence for health protocols, pharmacology, genetics, and biomechanics.
  • Field evidence takes precedence for deployment tactics, behavioral patterns in working environments, and breed suitability in specific terrain/climate combinations.
  • Where the two conflict, we present both and explain the gap — for example, a study conducted on kennel dogs may not reflect the behavior of range-deployed LGDs.

AI-Assisted Research

How We Use AI Tools

AI-assisted tools may be used to help locate sources, organize research notes, compare claims across papers, and prepare draft research summaries. Tools we may use include academic search engines, citation managers, and AI-based research assistants.

AI-generated material is not treated as final authority. All claims are reviewed against source quality, evidence strength, and editorial judgment before being promoted into Guardians.dog content or ontology data. No article or breed profile is published based solely on AI-generated output.

We disclose this because transparency builds trust. Using AI tools to accelerate research is no different from using a database or search engine — the human editorial review and accountability remain unchanged.

Updates & Corrections

Research evolves. Our content is not static. Here is how we handle updates:

  • Review cycle: High-risk content (health, pharmacology, triage) is reviewed at least annually. Breed profiles are reviewed when new evidence or field data becomes available.
  • Corrections: If we discover an error, we correct it promptly and note the change. Material corrections are logged.
  • "Last reviewed" dates: Pages that contain actionable health or safety information display a review date badge.
  • Reader feedback: If you believe something is inaccurate, contact us. We take corrections seriously.

Editorial Policy

Our sourcing tiers, editorial independence, and corrections policy.

Research Methodology

How we approach heritage classification and cultural research standards.

Mission & Ethics

Our editorial standards and ethical guidelines.

Contact

Report errors, suggest corrections, or ask questions.
Published: June 2026