← work mcsinger@home:~/work/newsplanetai
Case study 02 · intelligence platform

NewsPlanetAI

Live · operational since 2023

An independent public-interest intelligence system that tests global reporting against durable context and first-order evidence, then publishes what actually changed, with the receipts attached. A continuously updated world model, not a truth machine.

The NewsPlanetAI World Model, an hourly briefing with inline source citations
The World Model, each hour's briefing, written by Cortex from verified evidence, every claim cited inline.
219,039 articles archived
67 RSS feeds
12 categories
7 theaters tracked
9,122 Cortex briefings
The problem

News moves faster than verification.

Recurring events get sold as new. Declarations get confused with execution. The same wire, echoed across fifty outlets, looks like corroboration. NewsPlanetAI helps you tell a change in the world from a change in the story being told about it.

The wire, made visible

Every article, geocoded and placed on the world.

The live map, geocoded articles from the last 24 hours, colored by theater
Every dot is a geocoded article from the last 24 hours, the raw wire, placed on the world and colored by theater. The world, observed in context.
How it works

A pipeline that runs itself, a world model that doesn't.

Global wire 67 feeds → classified World model postgres + pgvector Cortex brief AI writes it Published hourly · map · API verified human OK reads the hour hourly revised on phase shift
The pipeline runs on its own. The world model doesn't change until I authorize it, and only on a genuine phase shift, not a repeat. Cortex writes each hour's brief from that evidence.
The hourly pipeline
Node.js/TypeScript. 67 configured RSS feeds in; each article extracted, embedded, classified across 12 categories and 7 theaters, geocoded, and stored in Postgres with pgvector.
The verified world model
One versioned model of what's happening. Same-origin repeats are collapsed before corroboration is counted; source tier, attribution, and cross-source convergence determine confidence. Contested attribution is preserved, and I authorize every change.
Cortex, the briefing
An AI writes each hourly broadcast from the hour's evidence and the curator-maintained world model, bounded to synthesis, never handed editorial authority.
Inside the platform
The hard part

The hard part isn't the code, it's time and epistemics: holding context across thousands of hourly cycles without letting the story overwrite the world.

Phase shift, not pattern
Claims are not events

A declaration isn't an execution. Announced measures stay separate from implemented ones, the gap between them is usually where the real story lives.

Repetition isn't corroboration

Same-origin reports get collapsed, so a single wire echoed across fifty outlets can't masquerade as fifty independent confirmations.

No post-cutoff contamination

Each verification cycle freezes a shared evidence cutoff, so later developments cannot silently enter an earlier assessment.

AI testimony is evidence, not authority

High-risk verification cycles can use a primary analyst and read-only adversarial counter-verifiers under one evidence cutoff. The curator resolves conflicts, and only the exact approved, hash-matched editorial artifact is promoted.

Where it stands

Public, free, and running, and it's the kind of system I build for clients.

NewsPlanetAI is free, ad-free, and under active development. If you need evidence-backed monitoring, analyst tooling, or a provenance system that abstains instead of guessing, get in touch.