Aggregate results accompanying the manuscript

These files contain aggregate results only. No operation or patient identifiers, individual prompts, credentials or model weights are included.
VitalDB: source cohort, 1909 operations. MOVER: external scheduled-resection candidates, 1791 operations.
Endpoint: in-hospital death OR discharge >14 days after anesthesia end.
qwen14 = Qwen3-14B Q4_K_M; gemma12 = Gemma3-12B-it Q4_K_M.
common_logistic / p_common_logistic = logistic regression; common_boosting / p_common_boosting = histogram gradient boosting.
raw = original probability; source_calibrated = source-only logistic calibration; source_calibrated_fixed_map = unchanged baseline source map applied to a variant.
repeat_table = identical-text replay; reverse_order = reversed clinical field order; unit_change = hemoglobin/albumin g/dL to g/L.
numeric_change = at least one convertible value; both_missing_wording_only = both values missing, unit text changed only.
AUROC / auroc: area under receiver operating characteristic curve. AUPRC / auprc: average precision. Brier / brier: mean squared probability error.
mean_absolute_change_pp: percentage-point absolute change. Brier_difference: variant minus baseline error; positive indicates worse error.
ci_low/ci_high: percentile 95% patient-cluster bootstrap limits conditional on saved predictions, 1000 draws, no multiplicity correction.
Subgroup tables without confidence intervals are descriptive. Rows are analyses, not patients. Empty metric cells mean undefined or not estimated; do not impute.
Source analyses use patient-grouped folds; external outcomes were not used for model fitting or calibration.
Small subgroup cell counts remain aggregate data; authors must confirm provider-compliant release.

File mapping:
baseline_performance.csv: 共同变量双模型_内部外部比较.csv
baseline_cluster_intervals.csv: 双模型配对患者聚类区间.csv
paired_model_and_calibration_comparisons.csv: 同校准阶段及校准前后_配对比较.csv
representation_intervals.csv: 等价表达_患者聚类区间与固定校准迁移.csv
representation_point_estimates.csv: 等价表达与重放_完整配对结果.csv
representation_patient_equal_sensitivity.csv: 等价表达与重放_患者等权敏感性.csv
restricted_cohort_sensitivity.csv: 双模型重复手术敏感性.csv
anatomy_missingness_subgroups.csv: 双模型解剖与缺失分层结果.csv
predictor_distributions.csv: 两库共同变量分布与缺失.csv
subgroup_denominators.csv: 预定分层样本与事件数.csv
fixed_bin_calibration.csv: 完整模型比较_固定分箱校准.csv
calibration_intercepts_slopes.csv: 完整模型比较_校准截距斜率.csv
source_calibration_parameters.json: 语言模型源数据校准参数.json
