Methodology
How Overcurrent works, what we get right, and what we get wrong.
We analyze how news outlets around the world cover the same stories. We don't produce journalism. We don't claim to be unbiased. We are transparent.
Every story goes through a six-stage pipeline before publication.
- 01Gather sources from RSS feeds across 50+ countries.
- 02AI triage deduplicates and categorizes incoming articles.
- 03Four AI models independently analyze each region’s coverage.
- 04Models cross-examine each other’s findings.
- 05A moderator synthesizes the debate into a unified analysis.
- 06Human reviews and publishes.
Analysis is structured as a three-round debate between four models:
Round 1 — Independent Analysis. Each model analyzes the source material separately. No model sees another's output.
Round 2 — Cross-Examination. Each model reviews the others' analyses and challenges claims, flags disagreements, and identifies gaps.
Round 3 — Moderator Synthesis. A moderator model synthesizes the debate into a single coherent analysis, noting where models agreed and disagreed.
Multi-model debate reduces the chance that any single model's biases, hallucinations, or blind spots survive into the final output. Disagreement is a feature, not a bug.
The consensus score is the percentage of outlets that agree on the core facts of a story.
Consensus does not equal truth. High consensus is notable, not proof.
We could be wrong. Our AI models hallucinate. Our source coverage has gaps. We miss non-English coverage. We may mischaracterize outlet positions. Flag errors and we'll fix them.
If you see something wrong, tell us. We will correct it publicly.