Credibility Scoring

Credibility Score Methodology

Understanding the "Cred" badge shown on person profiles, how it is calculated, and what it does and does not tell you.

What Is the Credibility Score?

The Credibility Score is a 0-100 composite metric assigned to each person mentioned in UFO/UAP video analysis. It is generated by AI (GPT-4o-mini) during the Program Intelligence extraction pipeline, based on verifiable indicators present in the video transcript.

The score is not a judgment of whether someone is telling the truth. It measures the density of verifiable credibility indicators found in the source material, such as security clearances held, sworn testimony given, documented evidence cited, or career risks taken.

When a person appears across multiple videos, their individual per-video credibility scores are averaged into an avg_credibility_score displayed on the Persons of Interest directory.

Scoring Rubric

The AI evaluates six factors. Each factor has a maximum point value. The AI sums all applicable factors to produce a score from 0 to 100.

Clearance Level Held

+20 pts

The person held or holds a documented security clearance (e.g., TS/SCI, Q clearance, SAP access). Higher clearance levels indicate access to classified UAP-related programs.

Top Secret / SCISAP-level accessQ clearance (DOE)

Sworn Testimony

+20 pts

The person has testified under oath, such as before a congressional committee, in a court of law, or in a signed affidavit. Perjury penalties make sworn statements carry more weight than informal interviews.

Congressional hearing testimonyICIG formal complaintSigned affidavit or deposition

Career Sacrifice / Whistleblower Risk

+15 pts

The person faced or risked significant professional consequences for coming forward, including termination, loss of clearance, legal threats, or social ostracism.

Lost security clearanceForced retirementNDA violation riskHarassment or intimidation reported

Corroborating Witnesses

+15 pts

Other named individuals independently confirm or corroborate the person's claims. The AI checks whether the transcript mentions additional witnesses supporting the account.

Named colleagues confirming the accountMultiple military witnesses to the same eventIndependent civilian corroboration

Documented Evidence

+15 pts

The person's claims are backed by physical documentation: FOIA releases, official reports, radar data, photographs, or other verifiable records referenced in the transcript.

FOIA-released documentsRadar or sensor dataOfficial government reportsPublished photographs or video

Peer Review / Academic Credentials

+15 pts

The person holds relevant academic credentials (PhD, professorship) or their work has been published in peer-reviewed journals, indicating methodological rigor.

PhD in relevant fieldPeer-reviewed publicationsAcademic appointmentProfessional scientific credentials

Score Interpretation

70-100

High — multiple strong indicators

40-69

Moderate — some verifiable factors

0-39

Low — limited verifiable indicators

How It's Calculated

Video Analysis — When a Tier 2 (Program/Research) video is processed through our pipeline, the transcript is sent to GPT-4o-mini in three parallel passes. Pass 1 (Network Mapping) extracts every named person mentioned.

Per-Person Scoring — For each named person, the AI identifies which of the six rubric factors are supported by evidence in the transcript. It produces a list of credibility_indicators (verifiable facts) and sums the corresponding point values.

Entity Resolution— Person mentions are fuzzy-matched across all videos into canonical profiles (handling aliases like "David Grusch" / "Grusch" / "David Charles Grusch"). This uses Levenshtein distance, variant rules, and LLM verification for ambiguous cases.

Averaging— The canonical person profile's displayed score is the average of all per-video credibility scores across every video in which they appear. A person mentioned in 5 videos will have 5 individual scores averaged together.

Strengths

Consistent application — Every person is evaluated against the same six-factor rubric. No human bias, favoritism, or reputation effects influence the score.

Evidence-anchored— The rubric is grounded in verifiable, objective facts (clearances, sworn testimony, documentation), not subjective assessments of "trustworthiness."

Multi-source averaging — Scores improve with more data. A person mentioned across many videos has a more robust average than someone appearing in a single interview.

Transparent rubric— The exact factors and point values are published here. Researchers can independently assess whether the AI's scoring aligns with their own reading of the source material.

Scalable — The same analysis has been applied uniformly across hundreds of videos, something no human team could achieve with the same consistency.

Limitations & Weaknesses

Transcript-dependent — The AI can only score indicators mentioned in the video transcript. If a person holds a TS/SCI clearance but the video never states this, it will not be counted. Background knowledge is intentionally excluded to prevent hallucination.

Not a truth detector — A high credibility score means the person has many verifiable indicators of institutional credibility. It does not mean their claims are true. Someone can be highly credentialed and still make inaccurate statements.

AI interpretation errors— GPT-4o-mini may occasionally misattribute indicators (e.g., confusing a journalist's report about sworn testimony with the journalist themselves having testified under oath).

Institutional bias in rubric design — The rubric inherently favors people with government/military backgrounds (clearances, congressional testimony). Civilian researchers, journalists, or experiencers without institutional affiliations will score lower even if their claims are well-supported by other forms of evidence.

Video selection bias — The score reflects what channels in our corpus say about a person. If the corpus is skewed toward channels that favor certain figures, those figures may have inflated or deflated scores.

No cross-verification— The AI does not independently verify whether a claimed credential is real. If a transcript says "he held a TS/SCI clearance," the AI takes that at face value.

Explore Persons of Interest

See the credibility scores in context alongside video appearances, roles, and extracted claims.

Browse Persons Full Methodology