Methodology | SI News & Analysis

Open Source Intelligence (OSINT) Methodology

SI News & Analysis uses Open Source Intelligence methodology adapted from intelligence analysis tradecraft. Every article goes through a structured process of collection, verification, analysis, and synthesis designed to produce comprehensive, multi-perspective coverage.

The Admiralty Code

Every source is graded using the NATO Admiralty Code (also called the NATO Evaluation System), a standard intelligence assessment framework used by military and intelligence organizations worldwide. It evaluates two independent dimensions:

Reliability (Source Quality)

[A1]A — Completely Reliable

[B1]B — Usually Reliable

[C1]C — Fairly Reliable

[D1]D — Not Usually Reliable

[E1]E — Unreliable

[F1]F — Cannot Be Judged

Credibility (Information Quality)

1 — Confirmed by other sources

2 — Probably True

3 — Possibly True

4 — Doubtfully True

5 — Improbable

6 — Cannot Be Judged

Example: A rating of [B2] means the source is “Usually Reliable” and the specific information is “Probably True.” Reliability and credibility are assessed independently — a highly reliable source can occasionally report unconfirmed information.

SI Structured Analysis Format

Every synthesized article follows our five-section structured format, designed to separate facts from analysis and make methodology transparent:

THE FACTS

Verified claims cross-referenced across multiple sources. Each fact is graded by the Admiralty Code. No interpretation or editorializing.

THE ANALYSIS

Original analytical framing synthesized from source material. Data-driven insights and contextual interpretation clearly labeled as analysis, not fact.

THE PERSPECTIVES

Every significant viewpoint mapped on a political/ideological spectrum. Includes evidence strength assessment and source attribution for each perspective.

THE ETHICS

Detected cognitive biases, logical fallacies, and media health assessment. Includes ethical scoring based on our philosophical framework.

THE METHODOLOGY

Full disclosure: sources analyzed, date range, AI models used, human review level, key assumptions, and any corrections.

The Role of AI

AI is integral to our process. Here is exactly what AI does and does not do:

AI Does:

Aggregate and categorize source articles at scale
Extract and cross-reference factual claims across sources
Detect perspective patterns and map viewpoint distribution
Draft synthesis text based on structured analysis
Flag potential biases and fallacies for human review
Generate Media Health Scores based on coverage analysis

AI Does Not:

Make editorial decisions about what to publish
Determine the final framing or headline
Add opinions or commentary not grounded in sources
Override human editorial judgment
Fabricate sources or quotes

Models Used

Synthesis Engine

Claude Opus 4.6 — primary analysis and synthesis

Bias Detection

Specialized classifiers trained on media analysis

Perspective Mapping

Embedding models with political spectrum calibration

Fact Extraction

NER + claim verification pipeline

Source Grading

Sources are evaluated on two axes using the Admiralty Code system described above. Reliability (A-F) assesses the source’s track record for accuracy, editorial standards, correction practices, and independence. Credibility (1-6) assesses the specific piece of information: is it confirmed by multiple sources, or is it a single uncorroborated claim?

We maintain an internal source database with rolling reliability scores updated based on correction frequency, retraction history, and cross-verification rates. New sources start at C3 (Fairly Reliable / Possibly True) until sufficient track record is established.

Bias Detection Methodology

Our system scans source coverage for 16 recognized cognitive biases:

Confirmation Bias

Anchoring Bias

Framing Bias

Selection Bias

Recency Bias

Survivorship Bias

Attribution Bias

Availability Bias

Bandwagon Effect

Authority Bias

Omission Bias

Negativity Bias

Status Quo Bias

Dunning-Kruger Effect

Halo Effect

In-Group Bias

Each detected bias is rated by severity (low / medium / high) and includes the specific source and evidence that triggered detection.

Fallacy Detection Methodology

We identify 12 common logical fallacies in source arguments and media framing:

False Dilemma

Straw Man

Ad Hominem

Appeal to Authority

Appeal to Emotion

Appeal to Tradition

Red Herring

Slippery Slope

Circular Reasoning

Hasty Generalization

Tu Quoque

False Equivalence

Fallacy detection uses both pattern-matching classifiers and LLM-based reasoning. Every flagged fallacy includes the evidence quote and an explanation of why it qualifies.

Human Review Levels

NoneAutomated processing only (used for data feeds, not published articles)

LightSpot-check of facts and framing (routine category updates)

StandardFull read-through with fact verification (most published articles)

DeepIndependent verification, expert consultation (major stories, investigations)

Ethical Framework

Our ethical analysis is grounded in three philosophical traditions that affirm the existence of objective truth and the moral obligation to seek it honestly:

C.S. Lewis

The Abolition of Man

The Tao — universal moral law recognized across cultures for millennia. Truth and ethical standards transcend cultural context.

Francis Schaeffer

He Is There and He Is Not Silent

True Truth — reality exists and can be known through honest, rigorous inquiry. Evidence-based reporting as moral imperative.

Tim Keller

Generous Justice

Fairness to all perspectives, including those we might personally disagree with. Justice requires hearing all voices.

When our ethics engine scores an article, it evaluates coverage through four lenses: Imago Dei (human dignity), Truth (correspondence to reality), Justice (fair representation), and Tone (proportional language). These scores inform the Media Health Score and ethical assessment note on each article.

Corrections Policy

When we get something wrong, we correct it promptly and transparently:

Immediate corrections — factual errors are corrected as soon as identified, with an inline notice on the article.
Public log — all corrections are logged in our permanent corrections page with the original text, corrected text, and reason for the change.
No silent edits — we never alter published content without a visible correction notice. Typos and formatting fixes are excluded.
Severity levels — corrections are classified as Minor (phrasing/context), Moderate (factual detail), or Major (fundamental claim). Major corrections trigger a re-review of the entire article.
Reader submissions — readers can flag errors via corrections@synthetic-insights.ai. We acknowledge all submissions within 24 hours.

Questions about our methodology? We welcome scrutiny. Contact us at methodology@synthetic-insights.ai. Read our Ethics Charter →