Who did what to whom, when, and where? Event data can be an invaluable resource in establishing the "hard facts" of warfare. Policymakers, activists, journalists and scientists often rely on such data to track the dynamics of armed conflict as it unfolds. Yet because these data almost always rely on some combination of commercial and state-controlled media, they tend to be quite noisy, and can internalize the potential biases of their source material --- including editorial decisions on "newsworthiness", but also purposeful efforts to deceive and misrepresent. What are the implications for social science research that uses such data? We will examines this problem in the case of Russia's 2022 invasion of Ukraine, using VIINA (Violent Incident Information from News Articles) --- a near-real time territorial control and violent event tracking system, which scrapes and parses news reports from Ukrainian and Russian media, georeferences them, and classifies them into standard event categories (e.g. firefight, tank battle, artillery shelling) through machine learning. This resource is updated daily, and is freely available at github.com/zhukovyuri/VIINA. We will provide an overview of the project, and discuss potential applications for policy and social science. These applications include investigations of changes in territorial control and variation across media sources in the relative location, intensity and attribution of reported violent events.