Data and code for: Quantifying cultural-intensity distributions in three dimensions — an anisotropic Gaussian spatial model with confidence weighting, demonstrated on the Neolithic cemetery–altar complex of Lingjiatan, China

This deposit contains the digitised evaluation corpus and the analysis code reproducing every number reported in the manuscript. | File | Description | |---|---| | `corpus_lingjiatan_85.csv` | The 85 cultural feature units (70 tombs, 15 altar/sacrificial units) digitised from the published excavation plans and reports: unit position (x, y in a local metric grid; z as elevation in m), recorded cultural intensity o, observation type (`count` = documented artefact count; `count_est` = transparent documented estimate; `structure` = structural unit recorded by area proxy), confidence weight α ∈ [0, 1] (grading rubric in Table 2 of the manuscript), estimated plan-reading uncertainty, and per-unit provenance notes (Chinese) giving the exact published source of every value. | | `3dgsm_analysis.py` | Main pipeline: α-weighted maximum-likelihood fit of the global influence covariance; anisotropic Gaussian-kernel regression; leave-one-out cross-validation against held-out recorded intensities; baselines (ordinary kriging on 3D coordinates, isotropic 2D KDE, IDW power 2) and ablations (isotropic 3D, anisotropic 2D, no-confidence); Wilcoxon signed-rank tests; percentile-bootstrap confidence intervals (2000 resamples). Two protocols: `--protocol counts` (primary; structural units set to the documented lower bound o = 0) and `--protocol composite` (sensitivity analysis, Section 4.6). | | `perturbation_simulation.py` | Coordinate-perturbation experiment (σ = 0.5 m horizontal / 0.2 m vertical, 100 replicates), confidence-grading perturbation (α ± 0.1, 100 replicates), and the controlled simulation study (axis ratio ρ ∈ {1, 2, 4, 8}, n ∈ {50, 100, 200}, 20 replicates; estimator recovery and prediction benchmarks). | | `supplementary_diagnostics.py` | Semivariogram fit of the recorded intensities (partial-sill collapse), short-lag directional semivariance check, bandwidth sensitivity (h × 0.8/1.0/1.2), and functional-zone centroids. | | `results_counts_primary.json` | Reference output of the primary counts protocol (as reported). | | `results_composite_sensitivity.json` | Reference output of the composite-proxy sensitivity analysis. | | `results_perturbation_simulation.json` | Reference output of the perturbation and simulation experiments. | | `results_supplementary_diagnostics.json` | Reference output of the supplementary diagnostics. | Positions were digitised from the published excavation plans of the Lingjiatan cemetery–altar complex (excavation briefs of the 1987, 1998, 2000 and 2007 campaigns and the 2016 tomb clearance, and the 2006 site report; full references in the manuscript), with an estimated plan-reading precision of ±0.5 m. Recorded intensities are the documented artefact counts published in the excavation inventories; for the 16 tombs of the 2000 campaign published without full inventories a transparent documented estimate (6.5 artefacts) is used and flagged `count_est`, and the eight structural units recorded by area proxy are set to the documented lower bound o = 0 under the primary protocol. Confidence weights follow the four-grade rubric of Table 2 of the manuscript. The corpus is an evaluation corpus assembled from published sources, not the excavation archive; the underlying excavation data belong to the Anhui Provincial Institute of Cultural Relics and Archaeology.