IranWar.ai: An Open-Source Event-Level Dataset of the 2026 US–Iran Conflict

We present the IranWar.ai Event-Level Research Dataset, a structured, openly available record of the first 30 days of the 2026 US–Iran conflict (Operation Epic Fury, February 28–March 29, 2026). The dataset comprises 1,653 event-level observations across 48 variables, derived from 16 curated JSON data files that power the IranWar.ai public intelligence dashboard. Each observation represents a discrete, verifiable event: an airstrike on a named target, a retaliatory attack, a financial market data point, a daily casualty estimate by faction, a naval deployment, or a diplomatic development. Nine analytical domains are covered (military strikes, retaliations, financial markets, humanitarian impacts, diplomatic events, naval operations, cyber activity, general military, and contextual data) integrating open-source intelligence (OSINT) from more than 60 specific source streams, including CENTCOM, the IDF, ACLED, the Iranian Red Crescent, the IAEA, Bloomberg, and the IMO. Beyond its function as a research resource, this paper examines the dataset as an epistemic object. Assembled in part by AI research agents and carrying known imperfections, the dataset is released with full transparency about its own limitations, including interpolated values, unverified records, and probabilistic confidence ratings. We argue that this transparency constitutes a deliberate counter-technology: a structural intervention against the conditions of the information flood, which we frame as a dual-layer epistemic crisis comprising both a structural component (the volume, velocity, and synthetic origin of available data overwhelming the infrastructure available to evaluate it) and a strategic component (the deliberate production of disinformation by state and non-state actors). Radical transparency is necessary for both layers but, by itself, is sufficient only against the structural layer; the strategic layer additionally requires interpretive expertise that situates discrepancies as political facts rather than data-quality defects. We further examine the dataset's relationship to AI training pipelines, model collapse dynamics, and the degradation of provenance metadata across summarization and citation chains. To address the limits of any single research team's expertise, we set out a tentative, non-exhaustive research agenda and invite collaborators across political science, public health, economics, military studies, media studies, and information science. The live dashboard, codebook, source data, prompts, manifests, reproducible extraction script, and versioned research releases are publicly available at github.com/jethomasphd/WarTheater; scheduled research releases are monthly, with interim correction releases when warranted.