Verification Methodology

How we verify a story.

Every headline enters the desk unverified. Here is the exact process that determines whether it earns a confirmed badge or stays flagged as developing.

Reliability Tiers

A / B / C / D

NATO admiralty system

Sources Monitored

198

Across 6 languages

Refresh Cycle

120s

Continuous polling

Corroboration Threshold

2 sources

Minimum for confirmation

Trust gate A/B chain before full confidence. Reliability stays visible from first source onward. Action gate Size against label, not headline heat. B and C items stay developing until corroborated. Invalidation Contradictions travel with the event. Denials downgrade confidence without hiding the chain.

The Rating System

Every source that enters the Intel Desk pipeline receives a reliability letter based on the NATO admiralty rating system. This letter is assigned at ingestion time and stays visible to the reader throughout the life of the story. The letter can be upgraded or downgraded as new sources confirm or contradict, but the original assignment is preserved in the verification chain.

Corroborated. At least two independent source tiers confirm the substance. Institutional or wire-service origin. This is the highest confidence the desk assigns.

Single credible source. Not yet confirmed by a second independent outlet, but the originating source has an established editorial track record. Most stories enter the wire at B.

OSINT, worth watching. Single open-source intelligence feed or social media account with a credible history. The desk surfaces it but does not yet stand on it.

State-aligned outlet. Press TV, TASS, Mehr News, IRNA, RT, and similar. Factored into the verification chain but weighted accordingly. A D-source denial does not automatically override an A or B confirmation.

Confidence Scoring

Beyond the per-source reliability letter, every story on the wire receives a composite confidence score. This score is recalculated each time a new source is added to the event chain. The formula is deterministic and visible.

Level	Criteria
HIGH	Three or more unique sources, including at least two rated A or B. No unresolved contradictions.
MEDIUM	Two or more sources with an average reliability score of 2.5 or above (where A=4, B=3, C=2, D=1).
LOW-MEDIUM	Two or more sources, but the average reliability score falls below 2.5. Often a B plus a D, or two C sources.
LOW	Single source only. No corroboration. The story is surfaced but flagged as developing.

When It Goes Wrong

Stories do not always survive scrutiny. When a contradicting source enters the chain, the desk does not discard it and does not pretend the contradiction did not happen. Instead, the story transitions through a state machine that tracks its lifecycle from first report to resolution.

Developing → Credible → Contested → Market-Moving

Contradictions are preserved in the event chain, never deleted. If a state media outlet issues a denial, that denial is filed as a new item in the same chain. The original story's reliability letter is appended with -contested (for example, A-contested), and the reader sees a CONTESTED badge on the feed item. The desk does not retract stories on the basis of a single state media denial.

This approach ensures that the full provenance of a story is always available. Readers can trace the exact sequence of claims, confirmations, and denials that led to the current confidence level.

See It In Action

The Trump / Iran / Brent case file is now published as one canonical record with source receipts, confirmation timing and market observations aligned. New case files remain withheld until the same reconciliation clears.

The live measurement status is documented on the lead-time proof page, and the published case file is available at /case-file/trump-iran-brent-2026-03-23.

Latest Alert Chain (Live)

Note The rating system and confidence formula described above are deterministic rules applied by the Intel Desk ingestion pipeline. They are not editorial judgments. The desk does not editorialize on whether a story is "true" or "false." It tracks sourcing, assigns reliability letters, and surfaces contradictions. The reader makes the final call.