1[2 {"id":"C1","text":"The estimation sample has 1800 firm-years from 300 firms after removing 40 exact duplicate rows and 300 rows with unobserved outcome (origin year 2022).","results_keys":["n_analysed","n_firms","n_input_rows","n_rows_after_dedupe","data_issues"],"status":"supported","caveat":"Duplicates were identical copies; one copy retained."},3 {"id":"C2","text":"rating_2025 violates the information-availability rule (built in 2025 from realised defaults) and is excluded from the predictors.","results_keys":["variables_excluded","predictors_used"],"status":"supported","caveat":"Based on the codebook; corroborated by all observed defaults falling in rating D."},4 {"id":"C3","text":"Baseline next-year default rate in the estimation sample is 0.121.","results_keys":["default_rate"],"status":"supported","caveat":""},5 {"id":"C4","text":"Conditional on cash ratio and log sales, leverage at end of year t is positively associated with default in t+1: coefficient 0.264 (firm-clustered SE 0.041, 95% CI 0.184 to 0.344).","results_keys":["beta_leverage","se_leverage_cluster","ci95_leverage_low","ci95_leverage_high","p_leverage_cluster"],"status":"supported","caveat":"Associational LPM estimate on synthetic data; not causal."},6 {"id":"C5","text":"The model has low explanatory power (R^2 = 0.035).","results_keys":["r2"],"status":"supported","caveat":"R^2 is naturally low for binary outcomes."}7]8