Editorial standards
Our editorial policy.
Real Narrative News is a news-aggregation and analysis platform, not a primary publisher. This document explains what we collect, how we label it, and how you can hold us accountable.
Last updated: July 11, 2026
Core principles
What we hold ourselves to.
Balance without false equivalence
Coverage is aggregated across the political and geographic spectrum. Where evidence differs in weight we say so — we do not manufacture a tie between well-supported findings and fringe claims.
Sourcing transparency
Every article on Real Narrative News is linked to its original publisher. Bias, factuality and credibility scores use the outlet as the unit of assessment — we never rewrite third-party reporting.
Model-assisted, human-reviewed
Bias tags, propaganda-technique labels and category assignments are produced by our engine. Editors sample outputs weekly and flag systematic errors to the model team for retraining.
Corrections logged publicly
Corrections are dated and preserved. If a story is deprecated or an article is removed, the URL returns an "archived by our team" notice rather than silently disappearing.
Methodology
How the engine works.
Ingestion
Aggregating ~25,000 articles per day
We ingest RSS, sitemap and API feeds from ~2,000 outlets across 146 countries. All ingestion is metadata-based: we store the headline, description, URL, image and publisher — not the full body of third-party articles.
Classification
Bias and factuality signal
Each outlet carries a bias score (left / center / right) and a credibility band. These are outlet-level, sourced from published methodology inspired by MBFC-style rubrics plus internal audits.
Narrative modelling
Propaganda technique detection
A dedicated model classifies rhetorical devices (loaded language, appeal to fear, name-calling, etc.). Detections are informational — presence of a technique is not, on its own, evidence of malicious intent.
Cross-outlet linking
Semantic story clustering
Articles are embedded into a vector space (Qdrant, MiniLM-L6-v2). On each read page we surface the closest matches from other outlets so you can compare how the same story is framed elsewhere.
Corrections & takedowns
How to request a correction.
We take corrections seriously and log every one. The fastest way to reach the editorial desk is by email.
Email editorial@realnarrativenews.com with the article URL.
Describe the correction and, where possible, cite an authoritative source.
We acknowledge within 3 business days and publish the correction on the affected page.
Model-level errors (e.g. a mis-classified outlet) are queued for the next weekly retrain cycle.
Independence & funding
No hidden sponsors.
Real Narrative News is operated by CoDataLab. Revenue comes from programmatic display ads and API/data-set licensing. No advertiser has influence over how outlets are classified or how stories are ranked. Sponsored placements — where they exist — are labelled and isolated from editorial surfaces.
We do not accept payment to raise or suppress the visibility of any outlet, story or narrative cluster.