Abstract for: From Spreadsheet to Structural Critique: How Representation Shapes Evaluation of Startup Financial Projections

Startup financial projections are typically presented as spreadsheets that make numerical assumptions visible but obscure causal structure. This may cause evaluators to focus on parameter plausibility while overlooking feedback loops, endogeneity, and boundary choices. Whether changing the representational form of a projection model from a spreadsheet to a stock-and-flow diagram shifts this pattern has not been empirically tested. We conducted a between-subjects classroom experiment with 92 graduate students enrolled in an introductory system dynamics course. Students were randomly assigned to evaluate one of two startup financial models — either as an Excel spreadsheet or a pre-built Vensim model — and responded to identical open-ended critique prompts. Responses were scored using an AI-assisted, expert-reviewed rubric distinguishing structural critique from parametric critique. Structural critique scores did not differ significantly across conditions (t=−1.01t = -1.01 t=−1.01, p=.314p = .314 p=.314, d=−0.21d = -0.21 d=−0.21), contrary to the primary hypothesis. Parametric critique scores were significantly higher in the Vensim condition (t=2.68t = 2.68 t=2.68, p=.009p = .009 p=.009, d=0.54d = 0.54 d=0.54). Overall scores were low in both conditions, and case-level results showed opposing directions across the two venture cases. A single session of representational exposure, without scaffolding, appears insufficient to shift evaluative reasoning toward structural critique. The unexpected parametric advantage in the Vensim condition may reflect how visible flow equations contextualise numerical inputs, though a case-content confound prevents firm conclusions. The study contributes an AI-assisted rubric methodology for scoring open-ended model critiques and motivates more controlled future designs. AI was used in the assessment, design of the study and the editing of the paper.