Our Methodology & Integrity Rules
The only serious risk we see is publishing unreliable data. Here is how we prevent it.
When methodology changes, people must know. Every version is documented.
Every Report Includes
Every finding includes how it was produced, even if it can't be fully reproduced.
Confidence Distribution
We don't hide low-confidence findings. We show the full distribution.
Critical Rule
If we ever publish "research" based on synthetic data without clear labeling, that credibility is nearly impossible to recover. We treat data integrity as our highest priority.
No Simulated Data in Public Reports
Seed data is used exclusively for development. No public report is ever published based on simulated or incomplete data.
Methodology Transparency
Every published finding includes methodology: number of queries, which AI models tested, time period covered, and significance criteria used.
Statistical Rigor
We require sufficient sample sizes before publishing. A single anomalous response is not a finding. Trends are confirmed across multiple queries and time periods.
Full Archive Access
Our raw data is verifiable. The AI Search Archive™ stores every response with full provenance: prompt, model, citations, entities, date.
Confidence Gates
Before any finding is published, it must pass evidence scoring, confidence evaluation, and editorial review. Low-confidence observations are flagged, never hidden.
Independent & Unbiased
We are an independent research center. We do not favor any AI model, platform, or company. Our only allegiance is to accurate, reproducible data.
Observatory DOI
OBS-2026-0042
Like academic DOI. Citable. Permanent. Never changes.
Permanent URLs
/research/2026/07/chatgpt-github-citations
Never /latest. Always permanent. This is an archive.
Three Data Modes
Everything allowed, simulated data visible
Everything allowed, simulated data clearly labeled
If isSimulated==true, API REFUSES. No option. Not even by mistake.