Long-form podcasts and video interviews have become influential venues for economic narratives, yet such artifacts are typically treated as anecdote: quoted selectively, summarized impressionistically, and left “uncodable” for evaluation. This paper presents a replicable pipeline that converts a public long-form transcript into (i) a bounded set of normalized claims, (ii) a directed acyclic graph (DAG) of asserted support relations, and (iii) a falsification-oriented measurement plan that maps key downstream claims to potential falsifiers and indicator families. The method is intentionally procedural rather than substantive: it does not endorse a narrative’s predictions; it makes the narrative auditable and empirically vulnerable by specifying what observations would count against it. The pipeline combines lightweight claim coding, graph construction with topological ordering, optional rhetorical edge weights (emphasis × specificity × modal commitment), Toulmin decomposition for key claims, and a “stock vs evolvability” measurement architecture to separate current tightness from adaptive capacity. The protocol generalizes to comparative narrative evaluation and longitudinal scorecards.

This paper introduces a source-agnostic pipeline for converting long-form podcasts, video interviews, or written narratives into auditable test designs. It extracts and normalizes key claims, maps their asserted support relations as a directed acyclic graph (DAG), identifies load-bearing premises and a “core spine” using simple graph diagnostics, and links high-centrality downstream claims to potential falsifiers and measurable indicator families. The method separates fast “stock” metrics (state/tightness) from slower “evolvability” metrics (adaptive capacity), and supports optional rhetorical edge weighting plus Toulmin decomposition for key claims. The contribution is procedural rather than substantive: a replicable protocol and artifact set that makes narratives transparent, contestable, and empirically vulnerable.