The Ground Truth

What it does

Elections are happening everywhere. From Serbia to the United States and South Africa, political campaigns are shaping how millions of people choose their next leaders. But between sensational headlines, political propaganda, selective reporting and misleading claims, how do we know what to believe? The Ground Truth turns fact-checking into an interactive investigation. Set inside an immersive 3D newsroom, users become investigative editors, submitting political claims, news articles or statements for a team of AI journalists to investigate. These agents independently research evidence, challenge one another's findings and uncover missing context, while users decide which leads to pursue. Rather than claiming to know the absolute truth, The Ground Truth helps people distinguish verified facts from misleading narratives, examine the evidence for themselves and make more informed decisions. Because choosing tomorrow's leaders should start with understanding today's headlines.

How the agents run concurrently

The Ground Truth uses Mozaik to coordinate multiple specialised AI agents through an event-driven investigative workflow. Initial investigation: Three agents run concurrently. The Prosecutor searches for contradictions, misleading framing and omitted context. The Defender investigates evidence supporting the claim and tests counterarguments. The Investigator independently retrieves primary sources, checks quotations and establishes factual context. Shared evidence and cross-examination: Agents contribute findings, source references, evidence exhibits and assessments to a shared case record. Mozaik coordinates their work through events, allowing agents to examine and challenge one another's findings. Disagreements trigger further scrutiny and agents can revise their positions as new evidence emerges. Editor-directed parallel research: Rather than ending after one assessment, agents identify unresolved questions and investigative leads. The user acts as editor, deciding which leads to pursue and assigning follow-up research. Multiple agents can investigate these leads concurrently while retaining the existing case context. Evidence-based verdict: Once the research and cross-examination stages are complete, a Judge agent evaluates the accumulated evidence and produces a sourced assessment. The application visualises agent activity, evidence collection, disagreements and revisions inside an interactive 3D newsroom. This demonstrates parallel execution, shared state, event-driven coordination, iterative research and human-in-the-loop decision-making, rather than a single chatbot performing sequential tasks.