What is the NarrativEye confidence framework?

NarrativEye uses three confidence tiers: Assert (high confidence, suitable for direct publication), Suggest with caveat (medium confidence, requires editorial judgment), and Do not publish (low confidence, for intelligence purposes only). Every finding is scored against this framework.

How does NarrativEye preserve evidence?

NarrativEye creates tamper-evident evidence chains for every detected narrative. Sources are timestamped, hashed, and archived in a format that can survive editorial and legal scrutiny — including for court proceedings and parliamentary inquiries.

How does NarrativEye attribute disinformation campaigns?

Attribution uses behavioural fingerprinting (posting cadence, coordination timing, linguistic patterns), network topology analysis (amplification graph structure), and technical signals (infrastructure, account metadata) to identify actors behind coordinated campaigns.

Intelligence you can stake your reputation on.

Every claim has a confidence tier. Every source has an audit trail. Every classification has a versioned record. Built for people who publish, brief, and stake their reputation on accuracy.

How we collect

We monitor 140,000+ sources across 47 languages using a multi-tier source strategy. Tier 1: wire services (Reuters, AP, AFP), major newspapers, broadcast transcripts. Tier 2: GDELT, social platforms (Twitter/X, Facebook, Reddit, YouTube), Telegram channels. Tier 3: state media, fringe platforms, known disinformation networks. Tier 4: academic repositories, think tank publications, policy documents.

Every source is captured with Wayback Machine archiving at point of collection. Metadata preserved: timestamp, URL, raw HTML, HTTP headers. No content is processed without a tamper-evident archive.

How we classify

Each article is mapped to our narrative ontology — a structured taxonomy of claims, entities, stances, and narrative frames. We perform entity recognition (who/what/where), stance detection (support/oppose/neutral), source credibility scoring, and cross-reference verification.

Our ontology covers: geopolitical narratives, corporate reputation, electoral claims, health misinformation, climate narratives, financial manipulation, and coordination patterns. Classifications are versioned — if our ontology changes, historical classifications are preserved with their original context.

How we score confidence

Every narrative assessment has one of three confidence tiers:

Assert: Publishable, briefable. Multi-source verification, cross-platform evidence, behavioral fingerprints. Legal review standard.
Suggest with caveat: Act with awareness. Single-source or circumstantial evidence. Include confidence caveat in reporting.
Do not publish: Insufficient evidence. May be true, but we cannot verify. Internal monitoring only.

Confidence is a spectrum, not a binary. We show our work: evidence graph, source count, coordination metrics, behavioral indicators.

How we attribute

Attribution is built from behavioral fingerprints, not guesswork. We track: coordination timing patterns (accounts posting within seconds), amplification networks (who amplifies whom), linguistic fingerprints (shared phrases, translation artifacts), account behavior (creation date, post frequency, follower graphs).

We don't claim certainty where there is none. Attribution comes with a confidence score: High (multiple independent indicators), Medium (circumstantial evidence, one strong indicator), Low (speculative, single weak indicator). We separate technical attribution (these accounts are coordinated) from political attribution (this is a state-backed operation).

How we preserve evidence

Every source is archived via Wayback Machine at point of capture. If a post is deleted, we have the original. If a website goes offline, we have the snapshot. All archives are timestamped, content-hashed, and indexed for retrieval.

Our audit trail is ACPO-aligned (Association of Chief Police Officers) — built for legal admissibility. Chain of custody is preserved: who collected, when, how. Versioned classifications: if we change an assessment, the original is preserved with justification. Tamper-evident logs: any change to a record generates a cryptographic hash.

Our editorial standards

What we publish: Only Assert-tier assessments. Multi-source verification. Clear confidence caveats. Source citations included.

How we review: Internal peer review before publication. External expert review for high-stakes claims. Legal review for attribution of state actors.

How we correct: Mistakes happen. When we're wrong, we say so publicly. Corrections are versioned and timestamped. Original assessments remain visible with strikethrough and correction notice.

What we refuse to do

No surveillance of private individuals. We monitor public sources only. No scraping of private accounts, no unauthorized access.

No targeted political work. We do not build opposition research for campaigns. We do not target individuals for political purposes.

No facilitation of harm. We will not provide intelligence that could be used to target journalists, activists, or vulnerable groups. We will not assist authoritarian regimes.

Our customers sign an Acceptable Use Policy. Violations result in immediate contract termination.

See our methodology in practice.

Request a demo and we'll walk through a live narrative with full evidence chain.

Request a demo