The LLT Lab empirically investigates the argument structures actually found in a sample of medical malpractice decisions selected from the U.S. federal courts. These decisions adjudicate claims for compensation for injuries due to medical malpractice in (for example) federal hospitals and clinics, as well as certain federally supported health centers. Using such decisions has several advantages, including the application of state substantive law and written fact-finding decisions (FTCA cases are bench trials, without a jury). In annotating spans of text within these decisions for the semantic roles played in legal reasoning and argumentation, traditional formal logic proves inadequate because it can represent very little of the natural language patterns found in actual decisions, and it does not provide useful heuristics for interpreting actual sentences in text as elements of argument structures. The Lab addresses both of these problems by proceeding from actual texts to develop patterns that explain all or most of the text. In the case of medical malpractice decisions, we find useful patterns not only characterized by the nature of the inference (deductive vs. probabilistic), but also characterized by types of evidence (e.g., legal precedent, medical or scientific study) and by types of evidentiary discrepancies (e.g., expert vs. expert, fact witness vs. contemporaneous documentation).