📊 Our Scoring Methodology

1. Aggregation

We continuously monitor news APIs and RSS feeds from trusted sources. When new UFO/UAP content is detected, it enters our analysis queue. We also analyse video — transcribing UFO/UAP footage and interviews so the same scrutiny applies to what's claimed on camera.

2. Source Evaluation

Each source is assigned a base reliability score from our database. Unknown sources receive a default neutral score of 50.

3. Live Context Gathering & Fact-Check

Before any scoring happens, we gather up-to-date context about the story's subject from a live web search, so the analysis isn't based on outdated knowledge. This step:

• searches the web for recent, verified facts and developments on the subject — and for contested or thinly-covered topics, runs additional official, skeptical, and latest-news searches so multiple perspectives are represented;
• reads beyond headlines — it fetches and cleans the full text of the one or two most reliable sources, not just search snippets, for deeper grounding;
• pre-checks the article's own claims, named entities, and any websites it links to — and actually fetches those links to confirm they resolve (official .gov/.mil domains are treated as authoritative);
• flags recent changes — such as an agency being renamed or a new official site launching — so a story isn't penalised for using current terminology that older AI knowledge wouldn't recognise.

Crucially, every fact in the resulting "current-context brief" is tagged with a date and a confidence level. The analysts are told to lean on high-confidence, recent, multiply-sourced facts and to down-weight anything low-confidence or undated — so a rumour never carries the same weight as a documented fact. The brief is cached and reused across articles on the same subject, and is treated as authoritative over the AI's own training data for anything recent.

4. Multi-Agent AI Analysis

Our analysis pipeline has three stages:

Stage 1 — RAG Retrieval: The article is matched against our knowledge graph of linked cases, evidence, and sources, and combined with the live context brief from step 3.

Stage 2 — Five-Agent Debate: Five specialized AI analysts independently evaluate the story: an Evidence Analyst classifies and scores evidence types; a Source Investigator checks source credibility and finds corroborating coverage; a Scientific Skeptic demands empirical proof and identifies alternative explanations; a Historical Archivist compares against known cases and identifies patterns; and a Government Disclosure Analyst tracks official statements and policy context.

Stage 3 — Synthesis: A master Orchestrator reviews all agent outputs, identifies points of agreement and disagreement, runs cross-examinations on disputed areas, and produces a final consensus report with credibility verdict and confidence intervals.

Faster stories are triaged with a lighter single-specialist "Quick" pass; high-profile and contested stories get the full five-agent debate.

5. Score Calculation

Each of the five factors is scored 0-100, then weighted according to the percentages above to produce a final credibility score.

6. Editorial Selection

Scoring a story doesn't automatically publish it. An AI editor reviews everything that's been analysed and ranks it for publishing — weighing credibility, how new it is versus what we already cover, and topical relevance — so the front page stays signal, not noise. The same editor can also work the other way: it independently identifies trending or historically important stories we haven't covered yet, researches them, and feeds them into this very pipeline. Nothing is published automatically unless we've explicitly enabled it for a trusted lane; otherwise a human approves.

7. Report Generation

A detailed report is generated explaining the score breakdown, key findings, and any concerns — with every source we consulted (the original publisher, the live web-context sources, and related cases) listed on the article so you can check our work.

8. Ongoing Freshness

A score is a snapshot of what was known when we analysed it. Because the situation can change, published stories are automatically re-checked and re-scored once their underlying context ages — so an old verdict doesn't quietly go stale. When a re-check is in progress, the article shows a small "score may be outdated — refreshing" note.