扩大数据与逐项模块实验（独立终审通过）

完成模型实例：26；训练教师建筑：20000；新测试城市：Groningen, Maastricht；原生测试：200栋×12条件×19流程。

数据为自动3DBAG教师屋面与模拟点云观测；所有流程共用教师足迹和地面。本轮DL输入为点云特征，RGB仅用于另附质量预览。结果不能作为真实AHN/3DEP、美国屋顶精度或标注节省的证明。

源验证预先选定：graph_20000_long，固定部署流程：graph_20000_long_s20261031。该流程在新测试城市的限幅损失 0.5123m；raw建筑宏均值 1.0940m。指标包括原生失败、LoD1与缺覆盖的10m罚分，不是通常成功屋顶MAE。

固定部署流程在新城市是否通过同样的损伤、失败、覆盖门槛：通过。这是测试后的诊断，不用于换模型或调阈值。

1. 增加数据与逐项模块

规模比较使用同10城市、哈希嵌套3500/10000/20000建筑、相同24000更新。long另用48000更新及相应cosine周期。所有模型先以校准BCE选checkpoint，再按实际原生重建选方法。

数据组成审查：各规模10城等额；新增16,500栋中80.7%含多个近似共面组。共面组不能直接作为屋脊类型或平屋顶标签。

Family	源loss(m)	新城市loss(m)	3seed SD	各seed clean损伤>2cm栋数	覆盖	共同含边域3D F1
raw	1.2461	1.0940	—	[0]	0.985	0.5411725439075514
geometry	1.2518	1.1041	—	[1]	0.984	0.5449900013276858
graph_10000	0.6863	0.5153	0.0033	[1, 1, 0]	0.987	0.6603495959430744
global_3500	0.7121	0.5536	0.0041	[2, 2, 2]	0.986	0.6490683605437774
graph_3500	0.6842	0.5161	0.0017	[1, 2, 0]	0.987	0.6586336892563921
graph_20000	0.6837	0.5147	0.0003	[0, 0, 0]	0.987	0.6600515861610803
graph_20000_long	0.6833	0.5128	0.0007	[0, 0, 0]	0.987	0.6619114530791954

三seed均值是分别重建后误差的平均，预测概率集成另列。主方案必须胜过通过安全门槛的raw/geometry基线；若未胜过，不归因于DL。

次要集成对照：每组使用三个checkpoint的概率均值，阈值仅用源校准集预先固定。增加了约三倍模型前向计算，不与单模型预算等同；完整报告预定的两个集成，不按测试结果改变主方案。

概率集成	新城市loss(m)	共同含边域3D F1	覆盖	clean损伤栋数	新增失败	原生LoD2
graph_10000_mean3	0.5152	0.6603	0.9869	0	0	2369/2400
graph_20000_long_mean3	0.5136	0.6608	0.9869	0	0	2369/2400

模型预算和代价（同一H100）：时间为各seed训练与校准墙钟均值，不含数据准备或原生重建；显存为高于公共驻留数据的峰值。MAC仅计卷积、线性和注意力矩阵乘法，排除了kNN/PCA/归一化/反向传播等，不能作为完整FLOPs或端到端效率证明。

Family	参数	训练秒	更新步数	每cloud前向百万MAC	额外峰值GB
graph_10000	59,141	263.3	24,000	159.5	2.007
global_3500	50,757	55.6	24,000	25.5	0.180
graph_3500	59,141	263.1	24,000	159.5	2.007
graph_20000	59,141	263.3	24,000	159.5	1.996
graph_20000_long	59,141	526.0	48,000	159.5	1.996

2. 筛选、组合与扩容兼容性

先在固定50栋源开发子集上比较全部首seed配置，最多选择3个模块；该50栋包含于完整200栋验证。仅入选模块与规模对照复验3seed，不能声称所有模块均经多seed验证。没有模块入选仅表示本轮50栋、首seed未达到预设改善要求；淘汰模块缺少3seed复验，不能据此判断它们普遍无效。

{
  "top_modules": [],
  "combination": null,
  "large_data_transfer": null
}

EMA是在相同有标签模拟观测上做双视图一致性正则，和独立监督AI、无标签目标域适配均为不同设置。法线/残差仅作辅助训练目标；原生重建使用分类概率筛点。

全部首轮配置及安全门槛（源50栋）

通过要求：逐栋clean损伤≤2cm、无新增原生失败、平均覆盖不低于raw减0.005。这里只作源开发筛选。结构F1另列为次要诊断，小幅升降不能替代多seed复验；主筛选继续严格使用预设限幅损失及门槛。

配置	源50 loss(m)	共同含边域3D F1	完整门槛	clean损伤栋数	覆盖
raw	1.1959	0.4663	通过	0	0.9800
geometry	1.2086	0.4744	未通过	1	0.9800
global_3500_s20261031	0.6395	0.5188	未通过	1	0.9800
graph_3500_s20261031	0.6147	0.5365	未通过	2	0.9800
graph_10000_s20261031	0.6133	0.5371	通过	0	0.9800
graph_20000_s20261031	0.6210	0.5333	通过	0	0.9800
graph_20000_long_s20261031	0.6160	0.5365	通过	0	0.9800
graph_10000_wide_s20261031	0.6227	0.5422	通过	0	0.9800
graph_10000_multiscale_s20261031	0.6247	0.5344	通过	0	0.9800
graph_10000_attention_s20261031	0.6178	0.5394	通过	0	0.9800
graph_10000_geometry_s20261031	0.6168	0.5390	通过	0	0.9800
graph_10000_masked_s20261031	0.6164	0.5394	通过	0	0.9800
graph_10000_normal_s20261031	0.6153	0.5384	通过	0	0.9800
graph_10000_residual_s20261031	0.6165	0.5398	通过	0	0.9800
graph_10000_reweight_s20261031	0.6194	0.5346	通过	0	0.9800
graph_10000_density_s20261031	0.6214	0.5281	通过	0	0.9800
graph_10000_strongaug_s20261031	0.6261	0.5347	通过	0	0.9800
graph_10000_consistency_s20261031	0.6319	0.5334	通过	0	0.9783

完整源200栋各family是否全部3seed通过门槛

{
  "graph_10000": false,
  "global_3500": false,
  "graph_3500": false,
  "graph_20000": false,
  "graph_20000_long": true
}

3. 屋面质量与失败案例

部署流程clean损伤>2cm：0/200；新增原生失败：0；成功LoD2：2369/2400。原生成功不能替代几何有效性或水密性认证。

以下案例按最大改善、中位差异、最大损伤及最大clean损伤事后选择，使用相同坐标范围与颜色，不代表随机抽样质量。

figures/reconstruction/test/full/graph_10000_s20261031/largest_clean_damage.png

figures/reconstruction/test/full/graph_10000_s20261031/largest_damage.png

figures/reconstruction/test/full/graph_10000_s20261031/largest_improvement.png

figures/reconstruction/test/full/graph_10000_s20261031/median_delta.png

figures/reconstruction/test/full/graph_20000_long_s20261031/largest_clean_damage.png

figures/reconstruction/test/full/graph_20000_long_s20261031/largest_damage.png

figures/reconstruction/test/full/graph_20000_long_s20261031/largest_improvement.png

figures/reconstruction/test/full/graph_20000_long_s20261031/median_delta.png

4. 空间相关性与研究边界

主损失以物理BAG建筑为宏平均单位，12个污染条件保持配对。辅助区间在每个固定城市内部按相邻建筑组重采样，再以固定城市权重合并；训练seed变化和逐城结果另报。两个测试城市不足以代表美国或全国分布。

{
  "raw": {
    "mean_parent_delta_m": 0.0,
    "parent_buildings": 200,
    "adjacency_clusters": 150,
    "fixed_city_parent_weights": {
      "Groningen": 0.5,
      "Maastricht": 0.5
    },
    "bootstrap_design": "Resample whole adjacency groups separately within each fixed city, then combine using observed fixed parent-city weights.",
    "cluster_bootstrap_95_m": [
      0.0,
      0.0
    ],
    "scope": "Conditional on the fixed cities and trained seeds; not a national or cross-city confidence interval."
  },
  "geometry": {
    "mean_parent_delta_m": 0.010071141820144028,
    "parent_buildings": 200,
    "adjacency_clusters": 150,
    "fixed_city_parent_weights": {
      "Groningen": 0.5,
      "Maastricht": 0.5
    },
    "bootstrap_design": "Resample whole adjacency groups separately within each fixed city, then combine using observed fixed parent-city weights.",
    "cluster_bootstrap_95_m": [
      -0.009639178305537694,
      0.030930368962604354
    ],
    "scope": "Conditional on the fixed cities and trained seeds; not a national or cross-city confidence interval."
  },
  "graph_10000": {
    "mean_parent_delta_m": -0.5787460191924895,
    "parent_buildings": 200,
    "adjacency_clusters": 150,
    "fixed_city_parent_weights": {
      "Groningen": 0.5,
      "Maastricht": 0.5
    },
    "bootstrap_design": "Resample whole adjacency groups separately within each fixed city, then combine using observed fixed parent-city weights.",
    "cluster_bootstrap_95_m": [
      -0.6083021777882281,
      -0.5491802950942336
    ],
    "scope": "Conditional on the fixed cities and trained seeds; not a national or cross-city confidence interval."
  },
  "global_3500": {
    "mean_parent_delta_m": -0.5404633790862304,
    "parent_buildings": 200,
    "adjacency_clusters": 150,
    "fixed_city_parent_weights": {
      "Groningen": 0.5,
      "Maastricht": 0.5
    },
    "bootstrap_design": "Resample whole adjacency groups separately within each fixed city, then combine using observed fixed parent-city weights.",
    "cluster_bootstrap_95_m": [
      -0.5724274999021602,
      -0.5093952541087531
    ],
    "scope": "Conditional on the fixed cities and trained seeds; not a national or cross-city confidence interval."
  },
  "graph_3500": {
    "mean_parent_delta_m": -0.5779167226466622,
    "parent_buildings": 200,
    "adjacency_clusters": 150,
    "fixed_city_parent_weights": {
      "Groningen": 0.5,
      "Maastricht": 0.5
    },
    "bootstrap_design": "Resample whole adjacency groups separately within each fixed city, then combine using observed fixed parent-city weights.",
    "cluster_bootstrap_95_m": [
      -0.6081782217537546,
      -0.5481118305756805
    ],
    "scope": "Conditional on the fixed cities and trained seeds; not a national or cross-city confidence interval."
  },
  "graph_20000": {
    "mean_parent_delta_m": -0.5793525053865807,
    "parent_buildings": 200,
    "adjacency_clusters": 150,
    "fixed_city_parent_weights": {
      "Groningen": 0.5,
      "Maastricht": 0.5
    },
    "bootstrap_design": "Resample whole adjacency groups separately within each fixed city, then combine using observed fixed parent-city weights.",
    "cluster_bootstrap_95_m": [
      -0.6092562398950536,
      -0.5499015724574148
    ],
    "scope": "Conditional on the fixed cities and trained seeds; not a national or cross-city confidence interval."
  },
  "graph_20000_long": {
    "mean_parent_delta_m": -0.581224560525294,
    "parent_buildings": 200,
    "adjacency_clusters": 150,
    "fixed_city_parent_weights": {
      "Groningen": 0.5,
      "Maastricht": 0.5
    },
    "bootstrap_design": "Resample whole adjacency groups separately within each fixed city, then combine using observed fixed parent-city weights.",
    "cluster_bootstrap_95_m": [
      -0.6105223924045632,
      -0.552251675332294
    ],
    "scope": "Conditional on the fixed cities and trained seeds; not a national or cross-city confidence interval."
  }
}

Automatic3DBAG Roofer-lineage teachers, not independent human GT
Synthetic point observations, no real AHN/3DEP training claim
Exact teacher roof-projection footprint and teacher ground are common priors; registration and footprint errors are not tested
No optical CNN or spatial image fusion tested in this ladder
No US precision, annotation saving, new architecture or SOTA claim
Only screened modules were repeated; screening50 is part of source200, so module selection is exploratory.
Three-seed family means average separately reconstructed errors, not the probability-mean ensemble.
Source/final use identical12 simulated corruption mechanisms; new cities do not establish real-point-cloud accuracy.
Native LoD2 status and edge-incidence diagnostics do not certify watertightness or complete solid validity.
Terrain input changed before this round; old-round absolute scores cannot be subtracted to attribute model improvement.
Bootstrap intervals omit teacher bias, new-city uncertainty, multiple-comparison selection effects and population seed uncertainty.

5. 按城市和污染条件核查

Family	Groningen	Maastricht
raw	1.2187	0.9694
geometry	1.2147	0.9935
graph_10000	0.6373	0.3933
global_3500	0.6823	0.4248
graph_3500	0.6399	0.3924
graph_20000	0.6363	0.3931
graph_20000_long	0.6340	0.3916

各城市都是100栋；值为三seed分别重建后的loss均值，基线只有一条确定性流程。

模拟条件	raw	geometry	graph_20000_long
clean	0.1100	0.1088	0.1099
sparse	0.5336	0.5379	0.5336
very_sparse	2.4510	2.4510	2.4510
noise	0.2582	0.2470	0.2590
hole	0.2429	0.2445	0.2515
scattered_tree	0.2335	0.1864	0.1769
dense_canopy	1.9311	2.0829	0.3838
flat_canopy	1.8641	1.8794	0.3670
low_canopy	0.2864	0.2874	0.2315
tree_multiple	0.5429	0.5148	0.2986
tilted_canopy	1.9830	1.9807	0.2536
occlusion_heavy	2.6916	2.7287	0.8374

本表继续报告源预选family；未根据新测试城市重新选择模型。

固定部署流程的误差分解（事后诊断）：缺覆盖罚分已经包含在主损失中，以下不会重复相加。该分解可帮助判断下一步应优先研究稀疏/遮挡下的结构补全，还是更可靠的污染点判别，但不能证明具体新模块的因果效果。

{
  "deployment_method": "graph_20000_long_s20261031",
  "mean_primary_loss_m": 0.5123316077996095,
  "strict_missing_coverage_penalty_component_m": 0.13099867838435816,
  "remaining_covered_cell_capped_error_component_m": 0.3813329294152513,
  "native_failed_or_non_LoD2_slots": 31,
  "native_LoD2_slots_failing_edge_incidence_diagnostic": 3,
  "condition_contribution_to_overall_loss_m": {
    "clean": 0.009166530906391659,
    "dense_canopy": 0.031720982533240884,
    "flat_canopy": 0.030145140355079068,
    "hole": 0.021010743486904612,
    "low_canopy": 0.019409288166246152,
    "noise": 0.021517935502312225,
    "occlusion_heavy": 0.06888393539233927,
    "scattered_tree": 0.015041680655041934,
    "sparse": 0.04447029535806123,
    "tilted_canopy": 0.021612862156877164,
    "tree_multiple": 0.025102776742394122,
    "very_sparse": 0.20424943654472114
  },
  "largest_condition_contributions": [
    "very_sparse",
    "occlusion_heavy",
    "sparse"
  ],
  "interpretation": "Post-test diagnosis of the fixed deployment method; 10*(1-coverage) is a component of loss, not added again. Edge incidence is an additional diagnostic, not complete solid validation. No causal attribution or retuning."
}

6. 可复核记录

训练、校准、源验证与最终测试的ID/来源/数据、代码、模型与原生命令均有内容哈希；独立AI监督逐阶段审查。最终方法在测试打开前冻结。

实现修正记录：首批推理遗漏加载训练state，已在任何源选型和最终测试开启之前发现。旧预测及该批评分已完整作废归档；16个有效训练检查点保留不改。修正后完整校准BCE/阈值复现、权重扰动/重新加载测试、实际源导出概率与独立加载forward逐点回放通过后，才重新评估和继续实验。科学数据、模型、超参和选型规则没有因这次修复而改变。

{
  "inference_correction_amendment.json": "dc0b2b92b8944222bef757398aea59e4a590912dc345d177cd3ccb24961831d0",
  "inference_correction_calibration_replay.json": "2a5e72eb9cb7e36f3485a5c645a5e66d58155a54ebbaebf059377d21b18c7ed1",
  "inference_correction_source_forward_replay.json": "c38e2631f4a1506c1c63427bb7b18e1b33083d2de9e2f003ec9a13deb55900bf"
}

{
  "data_counts": {
    "train": 20000,
    "calibration": 500,
    "validation": 500,
    "test": 600
  },
  "adjacency_component_counts": {
    "train": 8541,
    "calibration": 391,
    "validation": 385,
    "test": 360
  },
  "selection_sha256": "ffcbd35a50cb5d64b3f82c1380a57223a667f4c7b3035b059cafd285e78f83d5",
  "final_unlock_sha256": "06f0463a0cdd4be0afef2406b55de7378e6672dca53ac871d87cd68291ddd489",
  "source_summary_sha256": "6df09ee503fdfc228b6f8a0cb8bc858e6fc31c80ce90c082f5e257f2e70bf043",
  "test_summary_sha256": "543e4ef1ec46af71cf50154e617a90b4329344e1ec6514e37656fb4e5f949eaf",
  "elapsed_hours": 5.1262901975
}

模型预算、校准和全部记录

[
  {
    "method": "global_3500",
    "seed": 20261031,
    "sha256": "8a50b6e826176a041b9a5f8b3622955ff5423ce2bf0a79eabb9d3d86f30c1f3d",
    "settings": {
      "kind": "global",
      "width": 64,
      "buildings": 3500,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 50757,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 36.57142857142857,
    "elapsed_seconds": 56.318222761154175,
    "best_step": 23000,
    "best_calibration_BCE": 0.09751668733539004,
    "cutoff": 0.13753245770931244,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 4.794271744,
    "common_static_gpu_data_GB": 4.61456384,
    "peak_above_common_data_GB": 0.179707904,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.20755248060077428,
        "calibration_BCE": 0.17732745721990412,
        "elapsed_seconds": 2.637404203414917,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.15737022562325,
        "calibration_BCE": 0.16086677833325935,
        "elapsed_seconds": 4.942474365234375,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.14190271811187266,
        "calibration_BCE": 0.16285628958037404,
        "elapsed_seconds": 7.34491491317749,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.13290542291849852,
        "calibration_BCE": 0.14024283988259056,
        "elapsed_seconds": 9.699925661087036,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.1255174596644938,
        "calibration_BCE": 0.13846131642659507,
        "elapsed_seconds": 12.005761861801147,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.12105061615258456,
        "calibration_BCE": 0.12815652285489168,
        "elapsed_seconds": 14.309646368026733,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.1177331726513803,
        "calibration_BCE": 0.12821921830610797,
        "elapsed_seconds": 16.61907386779785,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.11294576336070895,
        "calibration_BCE": 0.13029558341286399,
        "elapsed_seconds": 18.93684220314026,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.10803376430645585,
        "calibration_BCE": 0.11878555607651219,
        "elapsed_seconds": 21.321228504180908,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.10565980772674084,
        "calibration_BCE": 0.11679085129246568,
        "elapsed_seconds": 23.623656511306763,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.10243098124116659,
        "calibration_BCE": 0.11694200055671461,
        "elapsed_seconds": 25.93861198425293,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.10100993597134948,
        "calibration_BCE": 0.11178947240656073,
        "elapsed_seconds": 28.269538164138794,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.09775661374628544,
        "calibration_BCE": 0.10808552921179569,
        "elapsed_seconds": 30.692551136016846,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.09569044718332588,
        "calibration_BCE": 0.1114038544625947,
        "elapsed_seconds": 32.99470257759094,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.09343532528355718,
        "calibration_BCE": 0.1080512617631392,
        "elapsed_seconds": 35.29401659965515,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.09100622834078967,
        "calibration_BCE": 0.10422049583666253,
        "elapsed_seconds": 37.60445213317871,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.08786197135411203,
        "calibration_BCE": 0.10267868009625059,
        "elapsed_seconds": 39.913875579833984,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.08791670129448176,
        "calibration_BCE": 0.10157663814660274,
        "elapsed_seconds": 42.21714949607849,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.0869833719022572,
        "calibration_BCE": 0.09976994198885832,
        "elapsed_seconds": 44.53237700462341,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.08580694073624909,
        "calibration_BCE": 0.10025006230672201,
        "elapsed_seconds": 46.840434551239014,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.08521859879419208,
        "calibration_BCE": 0.09850513707941229,
        "elapsed_seconds": 49.14965867996216,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.08506563155539335,
        "calibration_BCE": 0.09835082096909031,
        "elapsed_seconds": 51.60789346694946,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.08498601767607034,
        "calibration_BCE": 0.09751668733539004,
        "elapsed_seconds": 53.910781145095825,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.08349884132295847,
        "calibration_BCE": 0.0976761836196437,
        "elapsed_seconds": 56.20901393890381,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950307741820371,
    "calibration_contamination_rejection": 0.8452434040053417,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 25460736,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "global_3500",
    "seed": 20261101,
    "sha256": "392dcfb622f5b40555cefe878f872afed82f1b5409c2983f4117e0d33a7a965e",
    "settings": {
      "kind": "global",
      "width": 64,
      "buildings": 3500,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 50757,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 36.57142857142857,
    "elapsed_seconds": 55.20407319068909,
    "best_step": 22000,
    "best_calibration_BCE": 0.10145714374889027,
    "cutoff": 0.11596072465181351,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 4.827826176,
    "common_static_gpu_data_GB": 4.648118272,
    "peak_above_common_data_GB": 0.179707904,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.2089384358525276,
        "calibration_BCE": 0.1872454092314749,
        "elapsed_seconds": 2.277921676635742,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.15919816817343235,
        "calibration_BCE": 0.1647674591064453,
        "elapsed_seconds": 4.560402154922485,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.1457636426500976,
        "calibration_BCE": 0.16103779209021366,
        "elapsed_seconds": 6.982209920883179,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.13634442433714866,
        "calibration_BCE": 0.14884807836359198,
        "elapsed_seconds": 9.262949228286743,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.12715990643575786,
        "calibration_BCE": 0.13824791134920988,
        "elapsed_seconds": 11.551110029220581,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.12259981879964471,
        "calibration_BCE": 0.14046658329819187,
        "elapsed_seconds": 13.825275182723999,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.11800562307611108,
        "calibration_BCE": 0.13559488044507575,
        "elapsed_seconds": 16.09875726699829,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.11293481920659543,
        "calibration_BCE": 0.12662417343602037,
        "elapsed_seconds": 18.39189648628235,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.11071749233081937,
        "calibration_BCE": 0.12810031127929689,
        "elapsed_seconds": 20.66331386566162,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.1072705483995378,
        "calibration_BCE": 0.1286088726159298,
        "elapsed_seconds": 22.94138526916504,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.10373095356673002,
        "calibration_BCE": 0.11900217109448982,
        "elapsed_seconds": 25.225608587265015,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.10096012700162828,
        "calibration_BCE": 0.11428457609234434,
        "elapsed_seconds": 27.586474418640137,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.10027393011376262,
        "calibration_BCE": 0.11114744701963482,
        "elapsed_seconds": 29.90825080871582,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.09638360834494233,
        "calibration_BCE": 0.11044816048362038,
        "elapsed_seconds": 32.18381094932556,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.09568309844657778,
        "calibration_BCE": 0.11133006138657078,
        "elapsed_seconds": 34.46447801589966,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.0941103385463357,
        "calibration_BCE": 0.10951024974476208,
        "elapsed_seconds": 36.73933434486389,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.0918976506665349,
        "calibration_BCE": 0.10584193984522965,
        "elapsed_seconds": 39.028244733810425,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.0896586826518178,
        "calibration_BCE": 0.10562175565777403,
        "elapsed_seconds": 41.31535720825195,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.08921312993392348,
        "calibration_BCE": 0.10342842841870857,
        "elapsed_seconds": 43.591989040374756,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.08740001617744565,
        "calibration_BCE": 0.103557941829797,
        "elapsed_seconds": 45.868192195892334,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.08803998601436615,
        "calibration_BCE": 0.10296597451874705,
        "elapsed_seconds": 48.255335569381714,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.08528399816527962,
        "calibration_BCE": 0.10145714374889027,
        "elapsed_seconds": 50.553410053253174,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.08553951030224562,
        "calibration_BCE": 0.10172979736328125,
        "elapsed_seconds": 52.83232831954956,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.0853587368130684,
        "calibration_BCE": 0.10163000705603398,
        "elapsed_seconds": 55.10445690155029,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950329975492934,
    "calibration_contamination_rejection": 0.8297117347879186,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 25460736,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "global_3500",
    "seed": 20261102,
    "sha256": "5994d16274e670db9fc8718e0f1ace842bc00dd26f2a92b9a1a02dd39c2926f4",
    "settings": {
      "kind": "global",
      "width": 64,
      "buildings": 3500,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 50757,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 36.57142857142857,
    "elapsed_seconds": 55.35028028488159,
    "best_step": 23000,
    "best_calibration_BCE": 0.0958559595743815,
    "cutoff": 0.14033624529838562,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 4.827826176,
    "common_static_gpu_data_GB": 4.648118272,
    "peak_above_common_data_GB": 0.179707904,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.20846603649854661,
        "calibration_BCE": 0.1788548455440637,
        "elapsed_seconds": 2.273207426071167,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.15515569366514684,
        "calibration_BCE": 0.16312097496263908,
        "elapsed_seconds": 4.552597999572754,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.14024589574337007,
        "calibration_BCE": 0.14675011998956855,
        "elapsed_seconds": 6.834739685058594,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.13049405920132995,
        "calibration_BCE": 0.13934910999644887,
        "elapsed_seconds": 9.119912385940552,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.12479532635211944,
        "calibration_BCE": 0.13548634259773024,
        "elapsed_seconds": 11.405663967132568,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.11889717220515013,
        "calibration_BCE": 0.13294963993881687,
        "elapsed_seconds": 13.878011226654053,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.11546258643269539,
        "calibration_BCE": 0.1234297524654504,
        "elapsed_seconds": 16.148821592330933,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.10959879403561354,
        "calibration_BCE": 0.12012746554287997,
        "elapsed_seconds": 18.578587293624878,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.10638720502704382,
        "calibration_BCE": 0.12371486603707978,
        "elapsed_seconds": 20.86366868019104,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.10407043457776308,
        "calibration_BCE": 0.11786705155806108,
        "elapsed_seconds": 23.134666442871094,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.10040268225595356,
        "calibration_BCE": 0.11202296817663944,
        "elapsed_seconds": 25.420761585235596,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.09900821807980538,
        "calibration_BCE": 0.11141183462431936,
        "elapsed_seconds": 27.6998028755188,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.09529703962057828,
        "calibration_BCE": 0.10626305370908795,
        "elapsed_seconds": 29.986641883850098,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.09288721657171846,
        "calibration_BCE": 0.10572285600142045,
        "elapsed_seconds": 32.28241848945618,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.09124612294137478,
        "calibration_BCE": 0.10361651971990411,
        "elapsed_seconds": 34.56127309799194,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.09043407595157624,
        "calibration_BCE": 0.10139043902772846,
        "elapsed_seconds": 36.830538511276245,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.0874794134888798,
        "calibration_BCE": 0.09989677419951468,
        "elapsed_seconds": 39.27737355232239,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.08776865745335817,
        "calibration_BCE": 0.09718309141216856,
        "elapsed_seconds": 41.555553674697876,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.08520146591961383,
        "calibration_BCE": 0.0965614326245857,
        "elapsed_seconds": 43.84484601020813,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.08502086754143239,
        "calibration_BCE": 0.09683225042169745,
        "elapsed_seconds": 46.130871295928955,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.08263604864478111,
        "calibration_BCE": 0.09692931856097597,
        "elapsed_seconds": 48.413480281829834,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.08422204268351197,
        "calibration_BCE": 0.09605137060916785,
        "elapsed_seconds": 50.68825650215149,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.08297696720436215,
        "calibration_BCE": 0.0958559595743815,
        "elapsed_seconds": 52.97921633720398,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.08282885035127402,
        "calibration_BCE": 0.09592232582785866,
        "elapsed_seconds": 55.25368857383728,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.995017989820313,
    "calibration_contamination_rejection": 0.8467726687334335,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 25460736,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_attention",
    "seed": 20261031,
    "sha256": "2f6b431f0ba232ce20ead2d2132debbdfe52010bcbf68d94ffbb83550d3d52ec",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false,
      "attention": true
    },
    "parameters": 75910,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 286.99126529693604,
    "best_step": 22000,
    "best_calibration_BCE": 0.039272872947924066,
    "cutoff": 0.4012185335159302,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.678286848,
    "common_static_gpu_data_GB": 4.648741888,
    "peak_above_common_data_GB": 2.02954496,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14079779489710928,
        "calibration_BCE": 0.10896479154644591,
        "elapsed_seconds": 11.93373417854309,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.08342219720594585,
        "calibration_BCE": 0.0845788681492661,
        "elapsed_seconds": 23.876986503601074,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.06913726244866848,
        "calibration_BCE": 0.08694290799227627,
        "elapsed_seconds": 35.82264447212219,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.061220570674166086,
        "calibration_BCE": 0.06108115763808742,
        "elapsed_seconds": 47.76000380516052,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.05601302357390523,
        "calibration_BCE": 0.056996163709235916,
        "elapsed_seconds": 59.68847298622131,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.053132771169766786,
        "calibration_BCE": 0.06247941681832978,
        "elapsed_seconds": 71.62587547302246,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05097756564617157,
        "calibration_BCE": 0.052456716317841504,
        "elapsed_seconds": 83.56420755386353,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.04627880330290645,
        "calibration_BCE": 0.052467534521854285,
        "elapsed_seconds": 95.49041795730591,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.04423473285045475,
        "calibration_BCE": 0.048548329856178976,
        "elapsed_seconds": 107.41616702079773,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.04310004386864603,
        "calibration_BCE": 0.05385237223307292,
        "elapsed_seconds": 119.34491229057312,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04103714543301612,
        "calibration_BCE": 0.047368838292902164,
        "elapsed_seconds": 131.26804161071777,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04084595323074609,
        "calibration_BCE": 0.042864721471613104,
        "elapsed_seconds": 143.20372557640076,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.03980322431586683,
        "calibration_BCE": 0.04450601910677823,
        "elapsed_seconds": 155.14022088050842,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.03878038150910288,
        "calibration_BCE": 0.044041873353900335,
        "elapsed_seconds": 167.07319903373718,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.03700945640169084,
        "calibration_BCE": 0.041458928287390506,
        "elapsed_seconds": 179.00009202957153,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.03645027374010533,
        "calibration_BCE": 0.04103615863684452,
        "elapsed_seconds": 190.9352524280548,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.03574734592344612,
        "calibration_BCE": 0.04073897783684008,
        "elapsed_seconds": 202.87994122505188,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.03525598699972034,
        "calibration_BCE": 0.040508431475090256,
        "elapsed_seconds": 214.8167028427124,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.03516445247456432,
        "calibration_BCE": 0.04008445238055605,
        "elapsed_seconds": 226.75455927848816,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.03455250523798168,
        "calibration_BCE": 0.040309135020862925,
        "elapsed_seconds": 238.6821985244751,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03437355443462729,
        "calibration_BCE": 0.04006529064756451,
        "elapsed_seconds": 250.6087772846222,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.03480195820331573,
        "calibration_BCE": 0.039272872947924066,
        "elapsed_seconds": 262.5632152557373,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03403330743405968,
        "calibration_BCE": 0.03947147540468158,
        "elapsed_seconds": 274.48850107192993,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03386821833299473,
        "calibration_BCE": 0.039409975803259646,
        "elapsed_seconds": 286.43536853790283,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950057613004031,
    "calibration_contamination_rejection": 0.9623689095486291,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 201490432,
      "attention_extra_MACs": 41943040,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_consistency",
    "seed": 20261031,
    "sha256": "4b66b5f48756ec1421eaca61803317c8ff4caa9994ed19d724e4d6971353d3d4",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": true,
      "consistency": true
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 391.356258392334,
    "best_step": 22000,
    "best_calibration_BCE": 0.045207305307099314,
    "cutoff": 0.3812195360660553,
    "extra_EMA_teacher_forwards": 24000,
    "peak_allocated_gpu_GB": 6.965275648,
    "common_static_gpu_data_GB": 4.68229632,
    "peak_above_common_data_GB": 2.282979328,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14347954071685673,
        "calibration_BCE": 0.11059381686123934,
        "elapsed_seconds": 16.28844976425171,
        "student_sample_exposures": 64000,
        "consistency_confident_fraction": 0.7809160390691577
      },
      {
        "step": 2000,
        "training_loss": 0.09008058872632682,
        "calibration_BCE": 0.08972659304069751,
        "elapsed_seconds": 32.57401156425476,
        "student_sample_exposures": 128000,
        "consistency_confident_fraction": 0.9112215800285339
      },
      {
        "step": 3000,
        "training_loss": 0.07711075095273555,
        "calibration_BCE": 0.0775044003573331,
        "elapsed_seconds": 48.86705827713013,
        "student_sample_exposures": 192000,
        "consistency_confident_fraction": 0.9237184746265411
      },
      {
        "step": 4000,
        "training_loss": 0.07022237353026867,
        "calibration_BCE": 0.06801601856116092,
        "elapsed_seconds": 65.15945982933044,
        "student_sample_exposures": 256000,
        "consistency_confident_fraction": 0.9294807569384574
      },
      {
        "step": 5000,
        "training_loss": 0.06461646543629468,
        "calibration_BCE": 0.0646577475345496,
        "elapsed_seconds": 81.44523406028748,
        "student_sample_exposures": 320000,
        "consistency_confident_fraction": 0.9351209453344345
      },
      {
        "step": 6000,
        "training_loss": 0.05906399249844253,
        "calibration_BCE": 0.060185337436560425,
        "elapsed_seconds": 97.73230719566345,
        "student_sample_exposures": 384000,
        "consistency_confident_fraction": 0.9395903243422509
      },
      {
        "step": 7000,
        "training_loss": 0.057178218809887764,
        "calibration_BCE": 0.06247885078661369,
        "elapsed_seconds": 114.0237820148468,
        "student_sample_exposures": 448000,
        "consistency_confident_fraction": 0.9408173334598541
      },
      {
        "step": 8000,
        "training_loss": 0.054144629949703815,
        "calibration_BCE": 0.05557084161007043,
        "elapsed_seconds": 130.30861353874207,
        "student_sample_exposures": 512000,
        "consistency_confident_fraction": 0.9443634033203125
      },
      {
        "step": 9000,
        "training_loss": 0.052264436826109886,
        "calibration_BCE": 0.05735267123598041,
        "elapsed_seconds": 146.63287043571472,
        "student_sample_exposures": 576000,
        "consistency_confident_fraction": 0.9447608340978623
      },
      {
        "step": 10000,
        "training_loss": 0.05066599368490279,
        "calibration_BCE": 0.054846282982103756,
        "elapsed_seconds": 162.9162197113037,
        "student_sample_exposures": 640000,
        "consistency_confident_fraction": 0.9457368503212928
      },
      {
        "step": 11000,
        "training_loss": 0.05031520941946656,
        "calibration_BCE": 0.056790845651337594,
        "elapsed_seconds": 179.2327320575714,
        "student_sample_exposures": 704000,
        "consistency_confident_fraction": 0.9457584342360497
      },
      {
        "step": 12000,
        "training_loss": 0.04800832948833704,
        "calibration_BCE": 0.05023678900978782,
        "elapsed_seconds": 195.511461019516,
        "student_sample_exposures": 768000,
        "consistency_confident_fraction": 0.9469507742524147
      },
      {
        "step": 13000,
        "training_loss": 0.04643465357273817,
        "calibration_BCE": 0.054420730128432765,
        "elapsed_seconds": 211.79125547409058,
        "student_sample_exposures": 832000,
        "consistency_confident_fraction": 0.9488028225898743
      },
      {
        "step": 14000,
        "training_loss": 0.044933604747988284,
        "calibration_BCE": 0.05082409857547644,
        "elapsed_seconds": 228.08149766921997,
        "student_sample_exposures": 896000,
        "consistency_confident_fraction": 0.9497149918675423
      },
      {
        "step": 15000,
        "training_loss": 0.04431978460075334,
        "calibration_BCE": 0.04802269446633079,
        "elapsed_seconds": 244.38714814186096,
        "student_sample_exposures": 960000,
        "consistency_confident_fraction": 0.9501782185435295
      },
      {
        "step": 16000,
        "training_loss": 0.04446165152918547,
        "calibration_BCE": 0.04720228569724343,
        "elapsed_seconds": 260.65567421913147,
        "student_sample_exposures": 1024000,
        "consistency_confident_fraction": 0.9499180282354355
      },
      {
        "step": 17000,
        "training_loss": 0.042994133766740564,
        "calibration_BCE": 0.0474896567142371,
        "elapsed_seconds": 276.9207887649536,
        "student_sample_exposures": 1088000,
        "consistency_confident_fraction": 0.950605259001255
      },
      {
        "step": 18000,
        "training_loss": 0.04234619366750121,
        "calibration_BCE": 0.046396457556522254,
        "elapsed_seconds": 293.1919467449188,
        "student_sample_exposures": 1152000,
        "consistency_confident_fraction": 0.9509630829095841
      },
      {
        "step": 19000,
        "training_loss": 0.04167182790208608,
        "calibration_BCE": 0.04611282202980735,
        "elapsed_seconds": 309.45648860931396,
        "student_sample_exposures": 1216000,
        "consistency_confident_fraction": 0.9512960576415062
      },
      {
        "step": 20000,
        "training_loss": 0.041398721202276646,
        "calibration_BCE": 0.04574625704216235,
        "elapsed_seconds": 325.71951031684875,
        "student_sample_exposures": 1280000,
        "consistency_confident_fraction": 0.9518512914180756
      },
      {
        "step": 21000,
        "training_loss": 0.04129244608990848,
        "calibration_BCE": 0.04551995306304007,
        "elapsed_seconds": 341.9960913658142,
        "student_sample_exposures": 1344000,
        "consistency_confident_fraction": 0.9511278522610664
      },
      {
        "step": 22000,
        "training_loss": 0.0410475689843297,
        "calibration_BCE": 0.045207305307099314,
        "elapsed_seconds": 358.25934767723083,
        "student_sample_exposures": 1408000,
        "consistency_confident_fraction": 0.9515074676275254
      },
      {
        "step": 23000,
        "training_loss": 0.039964158335700634,
        "calibration_BCE": 0.04530713598077948,
        "elapsed_seconds": 374.5638380050659,
        "student_sample_exposures": 1472000,
        "consistency_confident_fraction": 0.952040900349617
      },
      {
        "step": 24000,
        "training_loss": 0.04024467528052628,
        "calibration_BCE": 0.045227962077747694,
        "elapsed_seconds": 390.8304600715637,
        "student_sample_exposures": 1536000,
        "consistency_confident_fraction": 0.9527077386379242
      }
    ],
    "calibration_roof_retention": 0.9950063171422171,
    "calibration_contamination_rejection": 0.954061410800224,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_density",
    "seed": 20261031,
    "sha256": "7e470a1900b658cc8676eed6122d36645c83efc953e035618e95ad09fcd7a239",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": true,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 268.32639384269714,
    "best_step": 22000,
    "best_calibration_BCE": 0.04629608882557262,
    "cutoff": 0.33895185589790344,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.678081024,
    "common_static_gpu_data_GB": 4.68229632,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.15342139476910233,
        "calibration_BCE": 0.10471392013087417,
        "elapsed_seconds": 11.156249284744263,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.09781123831495643,
        "calibration_BCE": 0.09093291820179332,
        "elapsed_seconds": 22.323222398757935,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.08547902940399944,
        "calibration_BCE": 0.08062115665782582,
        "elapsed_seconds": 33.48090672492981,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.07649469399265944,
        "calibration_BCE": 0.07622644579338304,
        "elapsed_seconds": 44.63830232620239,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.07277417497709393,
        "calibration_BCE": 0.06878901986786813,
        "elapsed_seconds": 55.80813145637512,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.06739094337075949,
        "calibration_BCE": 0.06641572311863755,
        "elapsed_seconds": 66.96772170066833,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.06486477123387158,
        "calibration_BCE": 0.06971825892130534,
        "elapsed_seconds": 78.1257266998291,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.061406486112624406,
        "calibration_BCE": 0.06501417749578303,
        "elapsed_seconds": 89.28182601928711,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05950192556157708,
        "calibration_BCE": 0.05721567355069247,
        "elapsed_seconds": 100.43788528442383,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.057303029529750346,
        "calibration_BCE": 0.05414238960959695,
        "elapsed_seconds": 111.58127975463867,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.05548087550885975,
        "calibration_BCE": 0.05404541309241093,
        "elapsed_seconds": 122.74050331115723,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.05429064064845443,
        "calibration_BCE": 0.055055619025952886,
        "elapsed_seconds": 133.8969886302948,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.05305988154187798,
        "calibration_BCE": 0.05123421882860588,
        "elapsed_seconds": 145.05546617507935,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.051978127898648384,
        "calibration_BCE": 0.05436936585397431,
        "elapsed_seconds": 156.20902562141418,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04989285219740123,
        "calibration_BCE": 0.049598890431722005,
        "elapsed_seconds": 167.37228107452393,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.050026378103531895,
        "calibration_BCE": 0.05159573865948301,
        "elapsed_seconds": 178.53257083892822,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.04836606372334063,
        "calibration_BCE": 0.04864303600426876,
        "elapsed_seconds": 189.69884514808655,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.047333757039159534,
        "calibration_BCE": 0.048498964852997754,
        "elapsed_seconds": 200.85780501365662,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.04702067128382623,
        "calibration_BCE": 0.04811051621870561,
        "elapsed_seconds": 212.01589345932007,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.046916916138492526,
        "calibration_BCE": 0.04693837816642992,
        "elapsed_seconds": 223.17104315757751,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.04539786383509636,
        "calibration_BCE": 0.04680336638941909,
        "elapsed_seconds": 234.32987785339355,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.04431480517517775,
        "calibration_BCE": 0.04629608882557262,
        "elapsed_seconds": 245.48365473747253,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.04525232817325741,
        "calibration_BCE": 0.046432228435169566,
        "elapsed_seconds": 256.64417457580566,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.04497275410965085,
        "calibration_BCE": 0.046381479251745975,
        "elapsed_seconds": 267.8008358478546,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950057613004031,
    "calibration_contamination_rejection": 0.9515225459522646,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_geometry",
    "seed": 20261031,
    "sha256": "6e97b133c47a45fcac002536a4d6aeebc6156ff22fc5bce78e890f1f90979bba",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false,
      "geometry": true
    },
    "parameters": 59653,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 278.5217077732086,
    "best_step": 23000,
    "best_calibration_BCE": 0.04244986667055072,
    "cutoff": 0.38769033551216125,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.707710464,
    "common_static_gpu_data_GB": 4.68229632,
    "peak_above_common_data_GB": 2.025414144,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.12559009171277286,
        "calibration_BCE": 0.08980729282263554,
        "elapsed_seconds": 11.590149641036987,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.0741802172102034,
        "calibration_BCE": 0.06930464167739406,
        "elapsed_seconds": 23.178777933120728,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.06354264403507114,
        "calibration_BCE": 0.06261461812799628,
        "elapsed_seconds": 34.75884747505188,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.05920573180541396,
        "calibration_BCE": 0.05723552983601888,
        "elapsed_seconds": 46.34397602081299,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.05561204681172967,
        "calibration_BCE": 0.056110564145174895,
        "elapsed_seconds": 57.92337989807129,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.052144771734252575,
        "calibration_BCE": 0.053705870587898025,
        "elapsed_seconds": 69.50725030899048,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05162766819167882,
        "calibration_BCE": 0.0512230442162716,
        "elapsed_seconds": 81.08135342597961,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.04908399455621838,
        "calibration_BCE": 0.05136321394949248,
        "elapsed_seconds": 92.65919733047485,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.04590835695713758,
        "calibration_BCE": 0.050610662471886836,
        "elapsed_seconds": 104.23534774780273,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.04597023826092481,
        "calibration_BCE": 0.048079371411872636,
        "elapsed_seconds": 115.83110761642456,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.0437256855359301,
        "calibration_BCE": 0.048080474645441226,
        "elapsed_seconds": 127.40196800231934,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04273115465417504,
        "calibration_BCE": 0.04572357799645626,
        "elapsed_seconds": 138.98657274246216,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04248739892337471,
        "calibration_BCE": 0.047734420290860266,
        "elapsed_seconds": 150.55743026733398,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04149944255501032,
        "calibration_BCE": 0.04550819979580966,
        "elapsed_seconds": 162.14535403251648,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.039420791001059115,
        "calibration_BCE": 0.04457126425540808,
        "elapsed_seconds": 173.71797347068787,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.03923243035562336,
        "calibration_BCE": 0.04372115989453865,
        "elapsed_seconds": 185.29711890220642,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.03839238944184035,
        "calibration_BCE": 0.04565456166122899,
        "elapsed_seconds": 196.8700795173645,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.037605269075371324,
        "calibration_BCE": 0.04331085327610825,
        "elapsed_seconds": 208.46603417396545,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.03780084730964154,
        "calibration_BCE": 0.042664771802497635,
        "elapsed_seconds": 220.03963208198547,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.037416189024224875,
        "calibration_BCE": 0.04332550642996123,
        "elapsed_seconds": 231.61728525161743,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03684259382355958,
        "calibration_BCE": 0.04303110238277551,
        "elapsed_seconds": 243.18978309631348,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.037443784174509345,
        "calibration_BCE": 0.04256086509011008,
        "elapsed_seconds": 254.7692174911499,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03669949032738805,
        "calibration_BCE": 0.04244986667055072,
        "elapsed_seconds": 266.34169578552246,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03657087077107281,
        "calibration_BCE": 0.042477402820009176,
        "elapsed_seconds": 277.94119095802307,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950074288258454,
    "calibration_contamination_rejection": 0.9588810919378258,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 163741696,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_masked",
    "seed": 20261031,
    "sha256": "b02e9aabab110a1b7b97c05b1a3863e3c8e2529918c18a943af0faaab0393232",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false,
      "masked_context": true
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 298.5306570529938,
    "best_step": 24000,
    "best_calibration_BCE": 0.04357742554635713,
    "cutoff": 0.37570643424987793,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.679391744,
    "common_static_gpu_data_GB": 4.68229632,
    "peak_above_common_data_GB": 1.997095424,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14915769396349787,
        "calibration_BCE": 0.11680113002892696,
        "elapsed_seconds": 12.407689332962036,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.08962303475663067,
        "calibration_BCE": 0.10731630216656309,
        "elapsed_seconds": 24.82796335220337,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07503771865926682,
        "calibration_BCE": 0.07366112925211589,
        "elapsed_seconds": 37.237884283065796,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06757743767276406,
        "calibration_BCE": 0.06764802689985795,
        "elapsed_seconds": 49.65121102333069,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06299342162534595,
        "calibration_BCE": 0.06356834081013997,
        "elapsed_seconds": 62.06977105140686,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.05981725293211639,
        "calibration_BCE": 0.07185377493193655,
        "elapsed_seconds": 74.48625230789185,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05751648820750415,
        "calibration_BCE": 0.06181882895267371,
        "elapsed_seconds": 86.89487290382385,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05287890908680856,
        "calibration_BCE": 0.05672815854621656,
        "elapsed_seconds": 99.29938745498657,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05040375733654946,
        "calibration_BCE": 0.055932479927756566,
        "elapsed_seconds": 111.7143907546997,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.04903226094506681,
        "calibration_BCE": 0.05343160941384056,
        "elapsed_seconds": 124.12316823005676,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04707875270023942,
        "calibration_BCE": 0.05509906935258345,
        "elapsed_seconds": 136.55536341667175,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.045818052975460884,
        "calibration_BCE": 0.05079648157755534,
        "elapsed_seconds": 148.95394039154053,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04458913388196379,
        "calibration_BCE": 0.04873997405081084,
        "elapsed_seconds": 161.3692855834961,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.043794044803828,
        "calibration_BCE": 0.0486021608988444,
        "elapsed_seconds": 173.77677249908447,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04089263048488647,
        "calibration_BCE": 0.04695142309015447,
        "elapsed_seconds": 186.1957585811615,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.041458316111005845,
        "calibration_BCE": 0.04587470430316347,
        "elapsed_seconds": 198.59840512275696,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.039590300575830045,
        "calibration_BCE": 0.04706147047100645,
        "elapsed_seconds": 211.0123815536499,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.039485452909953894,
        "calibration_BCE": 0.04583856628880356,
        "elapsed_seconds": 223.42130374908447,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.039055374471470714,
        "calibration_BCE": 0.04466718583540483,
        "elapsed_seconds": 235.84403109550476,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.0386184239089489,
        "calibration_BCE": 0.04503269579338305,
        "elapsed_seconds": 248.25301432609558,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.038090398591011765,
        "calibration_BCE": 0.04406472498575846,
        "elapsed_seconds": 260.67244052886963,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.03858016131538898,
        "calibration_BCE": 0.04364332552823153,
        "elapsed_seconds": 273.09670853614807,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.037897119467146696,
        "calibration_BCE": 0.043592350815281726,
        "elapsed_seconds": 285.51400923728943,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03738294875528663,
        "calibration_BCE": 0.04357742554635713,
        "elapsed_seconds": 297.9270167350769,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950085405094735,
    "calibration_contamination_rejection": 0.9562732555858,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_multiscale",
    "seed": 20261031,
    "sha256": "a4ae7361608c1ef771d45880ab889ac43d93885065e73fc7db97f75913aaca91",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false,
      "multiscale": true
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 493.4796311855316,
    "best_step": 24000,
    "best_calibration_BCE": 0.04075763744007457,
    "cutoff": 0.40450742840766907,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 8.8633216,
    "common_static_gpu_data_GB": 4.648741888,
    "peak_above_common_data_GB": 4.214579712,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14200173185020684,
        "calibration_BCE": 0.12736412455240886,
        "elapsed_seconds": 20.510374546051025,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.0862429644819349,
        "calibration_BCE": 0.08792607468113754,
        "elapsed_seconds": 41.0235641002655,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07066762835532427,
        "calibration_BCE": 0.07410197892622514,
        "elapsed_seconds": 61.54063630104065,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06363491295464337,
        "calibration_BCE": 0.061869636304450755,
        "elapsed_seconds": 82.04961919784546,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.058650530932471155,
        "calibration_BCE": 0.05929709814823035,
        "elapsed_seconds": 102.55989694595337,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.055178457966074346,
        "calibration_BCE": 0.060417441905628554,
        "elapsed_seconds": 123.07431626319885,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05391194884944707,
        "calibration_BCE": 0.05345986034508907,
        "elapsed_seconds": 143.59501719474792,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.049284576893784106,
        "calibration_BCE": 0.05214703184185606,
        "elapsed_seconds": 164.11274528503418,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.04675794020295143,
        "calibration_BCE": 0.049820011208274145,
        "elapsed_seconds": 184.6460428237915,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.047275212704204025,
        "calibration_BCE": 0.05065404912775213,
        "elapsed_seconds": 205.1538200378418,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04446133913472295,
        "calibration_BCE": 0.04797463878284801,
        "elapsed_seconds": 225.66698503494263,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04340863099880517,
        "calibration_BCE": 0.04671653282859109,
        "elapsed_seconds": 246.1774251461029,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04255878416728228,
        "calibration_BCE": 0.046868527776544745,
        "elapsed_seconds": 266.69128036499023,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04166699749603867,
        "calibration_BCE": 0.04449668817231149,
        "elapsed_seconds": 287.20150899887085,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.03929543607495725,
        "calibration_BCE": 0.04381247429125237,
        "elapsed_seconds": 307.71694588661194,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.03902534386981279,
        "calibration_BCE": 0.04236313959757487,
        "elapsed_seconds": 328.22877502441406,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.037944388776086274,
        "calibration_BCE": 0.044902223529237686,
        "elapsed_seconds": 348.77664065361023,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.03735668348055333,
        "calibration_BCE": 0.04170494105021159,
        "elapsed_seconds": 369.2846870422363,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.03713811789266765,
        "calibration_BCE": 0.04168206507364909,
        "elapsed_seconds": 389.7931044101715,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.03716867581382394,
        "calibration_BCE": 0.04142642216537938,
        "elapsed_seconds": 410.3106083869934,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03642250947188586,
        "calibration_BCE": 0.041369962125836,
        "elapsed_seconds": 430.8373193740845,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.0369208496902138,
        "calibration_BCE": 0.041091478012547346,
        "elapsed_seconds": 451.4603352546692,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.0362952182078734,
        "calibration_BCE": 0.0407888701467803,
        "elapsed_seconds": 471.9710900783539,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03591945499693975,
        "calibration_BCE": 0.04075763744007457,
        "elapsed_seconds": 492.4863154888153,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.995005205458589,
    "calibration_contamination_rejection": 0.9594685388862529,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 316833792,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_normal",
    "seed": 20261031,
    "sha256": "667a8ac833ad887cbb9fed1f096a9f42a6166d5ac6d574e13a77516039cc6cff",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.2,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 269.94947481155396,
    "best_step": 24000,
    "best_calibration_BCE": 0.04374683981230765,
    "cutoff": 0.36840569972991943,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.677658112,
    "common_static_gpu_data_GB": 4.68229632,
    "peak_above_common_data_GB": 1.995361792,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.1598443098552525,
        "calibration_BCE": 0.13515732153690221,
        "elapsed_seconds": 11.217906951904297,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.0946943873949349,
        "calibration_BCE": 0.09409297159368342,
        "elapsed_seconds": 22.44236183166504,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07941704102046787,
        "calibration_BCE": 0.07847112341216116,
        "elapsed_seconds": 33.66924428939819,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.07221506769023836,
        "calibration_BCE": 0.06844483459935044,
        "elapsed_seconds": 44.88313961029053,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06671654240600765,
        "calibration_BCE": 0.06374475000554865,
        "elapsed_seconds": 56.10254693031311,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.06337178150191904,
        "calibration_BCE": 0.062493905292857775,
        "elapsed_seconds": 67.39547848701477,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.06094248083420098,
        "calibration_BCE": 0.05733302475900361,
        "elapsed_seconds": 78.61763644218445,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.055983290577307346,
        "calibration_BCE": 0.05670176937218868,
        "elapsed_seconds": 89.82790803909302,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05346318176575005,
        "calibration_BCE": 0.05342205648711233,
        "elapsed_seconds": 101.04787135124207,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.0520410985853523,
        "calibration_BCE": 0.0543270722360322,
        "elapsed_seconds": 112.26635313034058,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.0495013756621629,
        "calibration_BCE": 0.05502815632675633,
        "elapsed_seconds": 123.51215815544128,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04850552033260465,
        "calibration_BCE": 0.05109113751035748,
        "elapsed_seconds": 134.74225878715515,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04739096110314131,
        "calibration_BCE": 0.050579129120797824,
        "elapsed_seconds": 145.9616436958313,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04623532546497881,
        "calibration_BCE": 0.04782956577647816,
        "elapsed_seconds": 157.175062417984,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04396254420559853,
        "calibration_BCE": 0.046968656065969755,
        "elapsed_seconds": 168.3927776813507,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.043125222412869334,
        "calibration_BCE": 0.04614274072358102,
        "elapsed_seconds": 179.6048903465271,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.04207840815652162,
        "calibration_BCE": 0.04702143566247189,
        "elapsed_seconds": 190.85082149505615,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.0415380035629496,
        "calibration_BCE": 0.04638876527728456,
        "elapsed_seconds": 202.07237720489502,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.041405605405569076,
        "calibration_BCE": 0.045630219592470114,
        "elapsed_seconds": 213.30733180046082,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.04131412764266133,
        "calibration_BCE": 0.045122007936419864,
        "elapsed_seconds": 224.51835298538208,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.04069911265745759,
        "calibration_BCE": 0.04423292719523112,
        "elapsed_seconds": 235.74223375320435,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.04102229991555214,
        "calibration_BCE": 0.04379241638183594,
        "elapsed_seconds": 246.97277641296387,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.040296631373465064,
        "calibration_BCE": 0.043754970758611504,
        "elapsed_seconds": 258.1936571598053,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.04003739979676902,
        "calibration_BCE": 0.04374683981230765,
        "elapsed_seconds": 269.41324615478516,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950090963512875,
    "calibration_contamination_rejection": 0.9560140178636195,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_residual",
    "seed": 20261031,
    "sha256": "cc58bda0a6c11a36e4cb62a932c9b628a105d1141882b08dd7817a4a2d1091dd",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.15,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 267.0233829021454,
    "best_step": 23000,
    "best_calibration_BCE": 0.04375845260620117,
    "cutoff": 0.38954657316207886,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.678213632,
    "common_static_gpu_data_GB": 4.68229632,
    "peak_above_common_data_GB": 1.995917312,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.1489017443433404,
        "calibration_BCE": 0.13884984010638612,
        "elapsed_seconds": 11.114728689193726,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.09294274890050291,
        "calibration_BCE": 0.09319908132842093,
        "elapsed_seconds": 22.255451917648315,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.0773144510705024,
        "calibration_BCE": 0.08713686715328332,
        "elapsed_seconds": 33.365328550338745,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06936378679983318,
        "calibration_BCE": 0.0696554779515122,
        "elapsed_seconds": 44.46015214920044,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06504128590784967,
        "calibration_BCE": 0.06369580528952859,
        "elapsed_seconds": 55.563743352890015,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.061129980757832526,
        "calibration_BCE": 0.07091689480868253,
        "elapsed_seconds": 66.66251850128174,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05837712265923619,
        "calibration_BCE": 0.05655638349128492,
        "elapsed_seconds": 77.76716446876526,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05399504652060568,
        "calibration_BCE": 0.05435249682339755,
        "elapsed_seconds": 88.86346864700317,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05125307021010667,
        "calibration_BCE": 0.056755964961196434,
        "elapsed_seconds": 99.9641797542572,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.050364856030792,
        "calibration_BCE": 0.055684831561464254,
        "elapsed_seconds": 111.0631217956543,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.048435165393166245,
        "calibration_BCE": 0.049137401118423,
        "elapsed_seconds": 122.16547179222107,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.0461996101308614,
        "calibration_BCE": 0.05068956499966708,
        "elapsed_seconds": 133.2569670677185,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.045659272705204784,
        "calibration_BCE": 0.04915464965357925,
        "elapsed_seconds": 144.36568927764893,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04441947518009692,
        "calibration_BCE": 0.049255334773208155,
        "elapsed_seconds": 155.46783137321472,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.042767526216804984,
        "calibration_BCE": 0.046323308540113045,
        "elapsed_seconds": 166.57032227516174,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.04184536119829863,
        "calibration_BCE": 0.045744337000991356,
        "elapsed_seconds": 177.6657257080078,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.04057081034500152,
        "calibration_BCE": 0.046613786500872986,
        "elapsed_seconds": 188.77206325531006,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.04025931069906801,
        "calibration_BCE": 0.04456974884957978,
        "elapsed_seconds": 199.86257219314575,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.04012284270115197,
        "calibration_BCE": 0.044298862688469165,
        "elapsed_seconds": 210.97758412361145,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.03979307186137885,
        "calibration_BCE": 0.0441510547059955,
        "elapsed_seconds": 222.0735456943512,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03941989950090647,
        "calibration_BCE": 0.044218760588674834,
        "elapsed_seconds": 233.17741298675537,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.0398115267502144,
        "calibration_BCE": 0.04399246715198864,
        "elapsed_seconds": 244.2943136692047,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03887187848705798,
        "calibration_BCE": 0.04375845260620117,
        "elapsed_seconds": 255.40057468414307,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.0383989245602861,
        "calibration_BCE": 0.04378336181640625,
        "elapsed_seconds": 266.499192237854,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.995011319718544,
    "calibration_contamination_rejection": 0.9569118365803453,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_reweight",
    "seed": 20261031,
    "sha256": "078f902e47afa3f2084f4cc540bcf05ef06e66f20ba79d23a38ad79395d668fa",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 4.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 262.66431879997253,
    "best_step": 22000,
    "best_calibration_BCE": 0.0553661708947384,
    "cutoff": 0.6774740219116211,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.678081024,
    "common_static_gpu_data_GB": 4.68229632,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.0889079767446965,
        "calibration_BCE": 0.1378901078195283,
        "elapsed_seconds": 10.924351453781128,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.053982272890396414,
        "calibration_BCE": 0.1265634143713749,
        "elapsed_seconds": 21.84115719795227,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.04511092993244529,
        "calibration_BCE": 0.09372377337831439,
        "elapsed_seconds": 32.77284646034241,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.04015584226232022,
        "calibration_BCE": 0.08549160045276989,
        "elapsed_seconds": 43.68840789794922,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.03781356783304363,
        "calibration_BCE": 0.10141193325158321,
        "elapsed_seconds": 54.64139103889465,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.03489242836833,
        "calibration_BCE": 0.07392003030487986,
        "elapsed_seconds": 65.54973340034485,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.033888550239615144,
        "calibration_BCE": 0.07836516866972952,
        "elapsed_seconds": 76.49131488800049,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.03131311614438891,
        "calibration_BCE": 0.07602272838245738,
        "elapsed_seconds": 87.41021823883057,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.02976840754505247,
        "calibration_BCE": 0.06515738719593395,
        "elapsed_seconds": 98.32892656326294,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.02850838768295944,
        "calibration_BCE": 0.06659075347438002,
        "elapsed_seconds": 109.25545954704285,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.027398986886721103,
        "calibration_BCE": 0.06345990034161192,
        "elapsed_seconds": 120.18169260025024,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.026466205445118247,
        "calibration_BCE": 0.06829159989790483,
        "elapsed_seconds": 131.10151290893555,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.025544150836300104,
        "calibration_BCE": 0.06294404523444898,
        "elapsed_seconds": 142.01555252075195,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.025103567099198698,
        "calibration_BCE": 0.06121722615559896,
        "elapsed_seconds": 152.93692755699158,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.023592448331881316,
        "calibration_BCE": 0.06376398886478309,
        "elapsed_seconds": 163.8509693145752,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.023583692308049648,
        "calibration_BCE": 0.059782154915549536,
        "elapsed_seconds": 174.79351353645325,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.02254008774925023,
        "calibration_BCE": 0.05657219444043709,
        "elapsed_seconds": 185.70974445343018,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.02217139959242195,
        "calibration_BCE": 0.0604864252726237,
        "elapsed_seconds": 196.6256229877472,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.022272797911427916,
        "calibration_BCE": 0.05667808109630238,
        "elapsed_seconds": 207.5402069091797,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.021917934517376125,
        "calibration_BCE": 0.05652129461115057,
        "elapsed_seconds": 218.45766592025757,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.021404276233632117,
        "calibration_BCE": 0.05728911474979285,
        "elapsed_seconds": 229.37365531921387,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.021943920836318284,
        "calibration_BCE": 0.0553661708947384,
        "elapsed_seconds": 240.29371190071106,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.021533224690705538,
        "calibration_BCE": 0.05641378090598367,
        "elapsed_seconds": 251.21297335624695,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.021292537048459053,
        "calibration_BCE": 0.05606792630282315,
        "elapsed_seconds": 262.13421607017517,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.995011319718544,
    "calibration_contamination_rejection": 0.953340777499117,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000",
    "seed": 20261031,
    "sha256": "04b58e8bd152d7dbee0c81fcca4f664a77b492f08b68740141da8fe376ec8ac9",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 262.7463080883026,
    "best_step": 24000,
    "best_calibration_BCE": 0.042420385441635594,
    "cutoff": 0.378459095954895,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.644526592,
    "common_static_gpu_data_GB": 4.648741888,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.1442119685821235,
        "calibration_BCE": 0.11358046616062974,
        "elapsed_seconds": 10.927356958389282,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.08928089263476431,
        "calibration_BCE": 0.10058449771765507,
        "elapsed_seconds": 21.84804058074951,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07449563357792795,
        "calibration_BCE": 0.07723904421257251,
        "elapsed_seconds": 32.792290449142456,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06667722562141716,
        "calibration_BCE": 0.0660538183038885,
        "elapsed_seconds": 43.73528337478638,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06284724522195756,
        "calibration_BCE": 0.06600410736546372,
        "elapsed_seconds": 54.66299915313721,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.058314336916431786,
        "calibration_BCE": 0.06874469241517962,
        "elapsed_seconds": 65.58318305015564,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05661145215295255,
        "calibration_BCE": 0.056614469840309836,
        "elapsed_seconds": 76.52413487434387,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.051762072475627065,
        "calibration_BCE": 0.05662657875292229,
        "elapsed_seconds": 87.44714903831482,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.04939063512627035,
        "calibration_BCE": 0.05332834111993963,
        "elapsed_seconds": 98.37351322174072,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.048901288052089514,
        "calibration_BCE": 0.054049947172222716,
        "elapsed_seconds": 109.29401016235352,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04694223840720951,
        "calibration_BCE": 0.04968834667783795,
        "elapsed_seconds": 120.2209403514862,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04540178445354104,
        "calibration_BCE": 0.04857201815518466,
        "elapsed_seconds": 131.1435227394104,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04407645619940013,
        "calibration_BCE": 0.048959527333577474,
        "elapsed_seconds": 142.06887078285217,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04339809553697705,
        "calibration_BCE": 0.04716143320257014,
        "elapsed_seconds": 153.01220154762268,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04125480163190514,
        "calibration_BCE": 0.04588085655443596,
        "elapsed_seconds": 163.9373848438263,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.040821254137903454,
        "calibration_BCE": 0.0445484545621005,
        "elapsed_seconds": 174.84967637062073,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.039611719680018725,
        "calibration_BCE": 0.04514558956261837,
        "elapsed_seconds": 185.77092146873474,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.03894875430501998,
        "calibration_BCE": 0.043616126274340083,
        "elapsed_seconds": 196.6861629486084,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.038805090645328166,
        "calibration_BCE": 0.04352850136034416,
        "elapsed_seconds": 207.61284708976746,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.03869785012397915,
        "calibration_BCE": 0.04404327956690933,
        "elapsed_seconds": 218.53025698661804,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03810030459240079,
        "calibration_BCE": 0.04283179799860174,
        "elapsed_seconds": 229.45337986946106,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.03838834354281426,
        "calibration_BCE": 0.04265350667780096,
        "elapsed_seconds": 240.38245511054993,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03787957506999373,
        "calibration_BCE": 0.04247308146158854,
        "elapsed_seconds": 251.30679273605347,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03734083339665085,
        "calibration_BCE": 0.042420385441635594,
        "elapsed_seconds": 262.2255594730377,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950063171422171,
    "calibration_contamination_rejection": 0.9562137973926393,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000",
    "seed": 20261101,
    "sha256": "5be4438748570b7aab91e2f8f497a61d8b79823fc4146ec7890b921267ee7b76",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 263.8584792613983,
    "best_step": 21000,
    "best_calibration_BCE": 0.043445868382309426,
    "cutoff": 0.3720487952232361,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.644526592,
    "common_static_gpu_data_GB": 4.615187456,
    "peak_above_common_data_GB": 2.029339136,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14461731031909586,
        "calibration_BCE": 0.11948922956062086,
        "elapsed_seconds": 11.39081335067749,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.08983098790794611,
        "calibration_BCE": 0.08808362114646218,
        "elapsed_seconds": 22.347418546676636,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07680220873095095,
        "calibration_BCE": 0.09122481719508316,
        "elapsed_seconds": 33.301284313201904,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06811016169935465,
        "calibration_BCE": 0.07372967459938742,
        "elapsed_seconds": 44.248682260513306,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06131513993442059,
        "calibration_BCE": 0.06485533518935695,
        "elapsed_seconds": 55.204139709472656,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.05807496721856296,
        "calibration_BCE": 0.061793795776367184,
        "elapsed_seconds": 66.14520907402039,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05454369547776878,
        "calibration_BCE": 0.06172657503070254,
        "elapsed_seconds": 77.10208487510681,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05265444313082844,
        "calibration_BCE": 0.05631991054650509,
        "elapsed_seconds": 88.05505847930908,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05037749468907714,
        "calibration_BCE": 0.053562396402070014,
        "elapsed_seconds": 99.01198649406433,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.04845072791166603,
        "calibration_BCE": 0.05056733070142341,
        "elapsed_seconds": 109.96149182319641,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04600423827301711,
        "calibration_BCE": 0.04901381086869673,
        "elapsed_seconds": 120.91588521003723,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04501877027004957,
        "calibration_BCE": 0.05234867153745709,
        "elapsed_seconds": 131.86777687072754,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04448807512409985,
        "calibration_BCE": 0.048378232343269115,
        "elapsed_seconds": 142.8254964351654,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.0424214002341032,
        "calibration_BCE": 0.04921507968324604,
        "elapsed_seconds": 153.77518367767334,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04102804062515497,
        "calibration_BCE": 0.04731053612448952,
        "elapsed_seconds": 164.73401021957397,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.04099289469048381,
        "calibration_BCE": 0.045754569105668504,
        "elapsed_seconds": 175.6902220249176,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.03997335045738146,
        "calibration_BCE": 0.044814800123734906,
        "elapsed_seconds": 186.64734864234924,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.038733664683066306,
        "calibration_BCE": 0.045389825115781844,
        "elapsed_seconds": 197.60661244392395,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.039316649904474615,
        "calibration_BCE": 0.044524429644960346,
        "elapsed_seconds": 208.5512499809265,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.038541474672034386,
        "calibration_BCE": 0.04385884119669596,
        "elapsed_seconds": 219.51275968551636,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03827536433283239,
        "calibration_BCE": 0.043445868382309426,
        "elapsed_seconds": 230.46277785301208,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.037370732589624825,
        "calibration_BCE": 0.043857461779045336,
        "elapsed_seconds": 241.42033004760742,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03732392032071948,
        "calibration_BCE": 0.04365215320009174,
        "elapsed_seconds": 252.36731362342834,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03741202101763338,
        "calibration_BCE": 0.04368577372233073,
        "elapsed_seconds": 263.33128571510315,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950146547694285,
    "calibration_contamination_rejection": 0.9552422505163944,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000",
    "seed": 20261102,
    "sha256": "8fb91978c8890b31b79b79183989fe00105e46b7ee9535437bb68903c85b9ab1",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 263.42905139923096,
    "best_step": 23000,
    "best_calibration_BCE": 0.04192320565888376,
    "cutoff": 0.38722676038742065,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.644526592,
    "common_static_gpu_data_GB": 4.648741888,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14416678776219488,
        "calibration_BCE": 0.11492733242150509,
        "elapsed_seconds": 10.955032587051392,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.09294067241810262,
        "calibration_BCE": 0.10483690030647047,
        "elapsed_seconds": 21.91873836517334,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07913596569001674,
        "calibration_BCE": 0.0870488629659017,
        "elapsed_seconds": 32.87649583816528,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.07052646017633378,
        "calibration_BCE": 0.07360258724328243,
        "elapsed_seconds": 43.82298159599304,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.0660256467424333,
        "calibration_BCE": 0.07051236364191228,
        "elapsed_seconds": 54.78288197517395,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.05999265631474555,
        "calibration_BCE": 0.0662945990822532,
        "elapsed_seconds": 65.72816371917725,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05591445362009108,
        "calibration_BCE": 0.06732825033014471,
        "elapsed_seconds": 76.69189548492432,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05325228310748935,
        "calibration_BCE": 0.05370315043131511,
        "elapsed_seconds": 87.65927505493164,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.0514290036149323,
        "calibration_BCE": 0.05335019165963838,
        "elapsed_seconds": 98.63767457008362,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.04758490396011621,
        "calibration_BCE": 0.05133161586414684,
        "elapsed_seconds": 109.5872437953949,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04776739033218473,
        "calibration_BCE": 0.053479114786783855,
        "elapsed_seconds": 120.54631423950195,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04578509217966348,
        "calibration_BCE": 0.05005175057613488,
        "elapsed_seconds": 131.49722003936768,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04406928723026067,
        "calibration_BCE": 0.05058495162039092,
        "elapsed_seconds": 142.4472210407257,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04287132216896862,
        "calibration_BCE": 0.046647366078694665,
        "elapsed_seconds": 153.38748383522034,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04146186808869243,
        "calibration_BCE": 0.04569129846746271,
        "elapsed_seconds": 164.34749627113342,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.042007073475047944,
        "calibration_BCE": 0.044553613604921286,
        "elapsed_seconds": 175.28821873664856,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.04003108937293291,
        "calibration_BCE": 0.044385753446636776,
        "elapsed_seconds": 186.23711490631104,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.03978685466898605,
        "calibration_BCE": 0.0434303888725512,
        "elapsed_seconds": 197.17841863632202,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.03941368347313255,
        "calibration_BCE": 0.04358932409575491,
        "elapsed_seconds": 208.13797521591187,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.038878512826748195,
        "calibration_BCE": 0.043430235822273025,
        "elapsed_seconds": 219.08571028709412,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.0377011575512588,
        "calibration_BCE": 0.04217658638231682,
        "elapsed_seconds": 230.03638792037964,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.038052395259961486,
        "calibration_BCE": 0.04200612876198508,
        "elapsed_seconds": 240.99468207359314,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03791183888306841,
        "calibration_BCE": 0.04192320565888376,
        "elapsed_seconds": 251.9486598968506,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03762849050201476,
        "calibration_BCE": 0.04200064771247632,
        "elapsed_seconds": 262.90405535697937,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950057613004031,
    "calibration_contamination_rejection": 0.9578500868684202,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_strongaug",
    "seed": 20261031,
    "sha256": "d9055ed4c59149dd733d79406181c3714fae2420ab9005333f4645f085211109",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": true,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 264.9828462600708,
    "best_step": 24000,
    "best_calibration_BCE": 0.04638752730398467,
    "cutoff": 0.3557748794555664,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.678605312,
    "common_static_gpu_data_GB": 4.68229632,
    "peak_above_common_data_GB": 1.996308992,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14234500170126557,
        "calibration_BCE": 0.11131591255881569,
        "elapsed_seconds": 11.022831678390503,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.08872522016614676,
        "calibration_BCE": 0.08870981137824781,
        "elapsed_seconds": 22.035576820373535,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07556087595969438,
        "calibration_BCE": 0.07433311406915838,
        "elapsed_seconds": 33.067256450653076,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06954493826813996,
        "calibration_BCE": 0.07249374921394117,
        "elapsed_seconds": 44.07817792892456,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06386692102067172,
        "calibration_BCE": 0.0643293081110174,
        "elapsed_seconds": 55.10232424736023,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.05886380803026259,
        "calibration_BCE": 0.06234538659760446,
        "elapsed_seconds": 66.11381363868713,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05778309366479516,
        "calibration_BCE": 0.07092714933915571,
        "elapsed_seconds": 77.1388943195343,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05392186171747744,
        "calibration_BCE": 0.057572010733864525,
        "elapsed_seconds": 88.15097045898438,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.052318017331883314,
        "calibration_BCE": 0.05613794953317353,
        "elapsed_seconds": 99.17696833610535,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.050874591081403193,
        "calibration_BCE": 0.056385959001020954,
        "elapsed_seconds": 110.18995332717896,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.05053350214101374,
        "calibration_BCE": 0.05534063646721117,
        "elapsed_seconds": 121.21483206748962,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.048956733463332056,
        "calibration_BCE": 0.054873087195194126,
        "elapsed_seconds": 132.22582817077637,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04664019992388785,
        "calibration_BCE": 0.055443812676632044,
        "elapsed_seconds": 143.24986720085144,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04483584296610206,
        "calibration_BCE": 0.050593997770367244,
        "elapsed_seconds": 154.25991678237915,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04474907707422972,
        "calibration_BCE": 0.04991284676754113,
        "elapsed_seconds": 165.28841614723206,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.04448338163644076,
        "calibration_BCE": 0.049368219456528174,
        "elapsed_seconds": 176.29729533195496,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.043475323140621185,
        "calibration_BCE": 0.048088870401093456,
        "elapsed_seconds": 187.31907629966736,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.042703497365117075,
        "calibration_BCE": 0.047757640422474254,
        "elapsed_seconds": 198.33211064338684,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.04190188309829682,
        "calibration_BCE": 0.04783804760557232,
        "elapsed_seconds": 209.3621335029602,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.04147733259852976,
        "calibration_BCE": 0.04680343401359789,
        "elapsed_seconds": 220.37988686561584,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.04161255034524947,
        "calibration_BCE": 0.04714831586895567,
        "elapsed_seconds": 231.39522647857666,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.041241956558078524,
        "calibration_BCE": 0.04645342890421549,
        "elapsed_seconds": 242.42303204536438,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.040317751845344904,
        "calibration_BCE": 0.04657122548421224,
        "elapsed_seconds": 253.44174909591675,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.040227447868324814,
        "calibration_BCE": 0.04638752730398467,
        "elapsed_seconds": 264.4573380947113,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950002028822621,
    "calibration_contamination_rejection": 0.9519137808632616,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_10000_wide",
    "seed": 20261031,
    "sha256": "4498ffd0014a26fcca783a7c68ad8a5158ec195f347ed4d3fc25821dddc9a63b",
    "settings": {
      "kind": "graph",
      "width": 128,
      "buildings": 10000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 142085,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 452.6724486351013,
    "best_step": 24000,
    "best_calibration_BCE": 0.039470191516298235,
    "cutoff": 0.42155182361602783,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 8.550091264,
    "common_static_gpu_data_GB": 4.648741888,
    "peak_above_common_data_GB": 3.901349376,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.13784694117307664,
        "calibration_BCE": 0.10079711705988104,
        "elapsed_seconds": 18.826173782348633,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.0812601877823472,
        "calibration_BCE": 0.08164474704626835,
        "elapsed_seconds": 37.65299201011658,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.06745482021942735,
        "calibration_BCE": 0.06889818536007043,
        "elapsed_seconds": 56.487112045288086,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06036396371759474,
        "calibration_BCE": 0.05863995446869821,
        "elapsed_seconds": 75.32374787330627,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.05546467475034297,
        "calibration_BCE": 0.056866214613481,
        "elapsed_seconds": 94.14716744422913,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.05246183938160539,
        "calibration_BCE": 0.05951379891597863,
        "elapsed_seconds": 113.01114273071289,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05136717659328133,
        "calibration_BCE": 0.05020585211551551,
        "elapsed_seconds": 131.83955907821655,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.046880560480058196,
        "calibration_BCE": 0.05485493878451261,
        "elapsed_seconds": 150.65797066688538,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.04493074419721961,
        "calibration_BCE": 0.047435290920373165,
        "elapsed_seconds": 169.47432494163513,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.044399969477206466,
        "calibration_BCE": 0.052712006239457564,
        "elapsed_seconds": 188.28216433525085,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04297287147678435,
        "calibration_BCE": 0.04700851893569484,
        "elapsed_seconds": 207.1067533493042,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.041171515244059265,
        "calibration_BCE": 0.0460900863647461,
        "elapsed_seconds": 225.92948865890503,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04056608775723725,
        "calibration_BCE": 0.04647413410995946,
        "elapsed_seconds": 244.7483012676239,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.039945362457074224,
        "calibration_BCE": 0.04709718664920691,
        "elapsed_seconds": 263.5522196292877,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.03787841889541596,
        "calibration_BCE": 0.04153532437411222,
        "elapsed_seconds": 282.37824153900146,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.03723347050976008,
        "calibration_BCE": 0.04434983171405214,
        "elapsed_seconds": 301.187842130661,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.036199922319501636,
        "calibration_BCE": 0.04197743014711322,
        "elapsed_seconds": 320.0043923854828,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.03595361762307584,
        "calibration_BCE": 0.0425095390551018,
        "elapsed_seconds": 338.81338119506836,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.03572427855338901,
        "calibration_BCE": 0.041117650373054274,
        "elapsed_seconds": 357.6288709640503,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.03532207162026316,
        "calibration_BCE": 0.03999850993300929,
        "elapsed_seconds": 376.5134015083313,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03486033781524748,
        "calibration_BCE": 0.03996069634177468,
        "elapsed_seconds": 395.32791662216187,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.03510761597752571,
        "calibration_BCE": 0.03965328676628344,
        "elapsed_seconds": 414.1634726524353,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03452347042132169,
        "calibration_BCE": 0.03953258703983191,
        "elapsed_seconds": 432.98789644241333,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.034267897407989946,
        "calibration_BCE": 0.039470191516298235,
        "elapsed_seconds": 451.80022716522217,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950124314021721,
    "calibration_contamination_rejection": 0.9613545527733085,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 583172096,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_20000_long",
    "seed": 20261031,
    "sha256": "0899d4f194dbad868ebf63f267dc6a54e7723491ab96c91347dfe323088d56b1",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 20000,
      "steps": 48000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 48000,
    "sample_exposures": 3072000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 525.0704326629639,
    "best_step": 46000,
    "best_calibration_BCE": 0.037348759252374825,
    "cutoff": 0.43206343054771423,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.645486592,
    "common_static_gpu_data_GB": 4.649701888,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14288573547452688,
        "calibration_BCE": 0.10663205469304865,
        "elapsed_seconds": 10.91603708267212,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.08988228173181415,
        "calibration_BCE": 0.09135254384821112,
        "elapsed_seconds": 21.852630138397217,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07699122696742415,
        "calibration_BCE": 0.07411275560783617,
        "elapsed_seconds": 32.79812455177307,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.0670053810980171,
        "calibration_BCE": 0.07060909088597153,
        "elapsed_seconds": 43.73158311843872,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06359456430375576,
        "calibration_BCE": 0.06534048804080847,
        "elapsed_seconds": 54.6480278968811,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.058925032367929814,
        "calibration_BCE": 0.06219345705436938,
        "elapsed_seconds": 65.57867884635925,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05575297846086323,
        "calibration_BCE": 0.05552361151955344,
        "elapsed_seconds": 76.51995134353638,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05338199113495648,
        "calibration_BCE": 0.06254831071333451,
        "elapsed_seconds": 87.44117999076843,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05104586965497583,
        "calibration_BCE": 0.056744908812551786,
        "elapsed_seconds": 98.36248826980591,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.04981583339348435,
        "calibration_BCE": 0.05864311823989406,
        "elapsed_seconds": 109.28024339675903,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04826351169403643,
        "calibration_BCE": 0.052036503531716086,
        "elapsed_seconds": 120.21416234970093,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04681813441310078,
        "calibration_BCE": 0.0507306052236846,
        "elapsed_seconds": 131.15544509887695,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04562482407782227,
        "calibration_BCE": 0.05680282350019975,
        "elapsed_seconds": 142.07310485839844,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04532378862891346,
        "calibration_BCE": 0.04839414381547408,
        "elapsed_seconds": 153.00777173042297,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04402727210521698,
        "calibration_BCE": 0.048109902445475264,
        "elapsed_seconds": 163.9321539402008,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.04292255485150963,
        "calibration_BCE": 0.044636502491344106,
        "elapsed_seconds": 174.87387657165527,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.04299745534639805,
        "calibration_BCE": 0.04551637455333363,
        "elapsed_seconds": 185.7994418144226,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.04132895869389176,
        "calibration_BCE": 0.04492824209964637,
        "elapsed_seconds": 196.73140740394592,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.041430957665666936,
        "calibration_BCE": 0.045662467794707326,
        "elapsed_seconds": 207.64906287193298,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.040718441675417126,
        "calibration_BCE": 0.045896109170624706,
        "elapsed_seconds": 218.56905579566956,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.040587417070753876,
        "calibration_BCE": 0.042007923727324516,
        "elapsed_seconds": 229.49495649337769,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.03961204582033679,
        "calibration_BCE": 0.045628400536739465,
        "elapsed_seconds": 240.41463923454285,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03914000938739628,
        "calibration_BCE": 0.041334020672422464,
        "elapsed_seconds": 251.34203338623047,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03875367568433285,
        "calibration_BCE": 0.04242676481767134,
        "elapsed_seconds": 262.26169538497925,
        "student_sample_exposures": 1536000
      },
      {
        "step": 25000,
        "training_loss": 0.03765910149365664,
        "calibration_BCE": 0.04232136870875503,
        "elapsed_seconds": 273.17863845825195,
        "student_sample_exposures": 1600000
      },
      {
        "step": 26000,
        "training_loss": 0.03758688362501562,
        "calibration_BCE": 0.04223990822994348,
        "elapsed_seconds": 284.0989758968353,
        "student_sample_exposures": 1664000
      },
      {
        "step": 27000,
        "training_loss": 0.03773049069615081,
        "calibration_BCE": 0.041150802311752784,
        "elapsed_seconds": 295.01934790611267,
        "student_sample_exposures": 1728000
      },
      {
        "step": 28000,
        "training_loss": 0.037048603439703585,
        "calibration_BCE": 0.04092325800115412,
        "elapsed_seconds": 305.9653973579407,
        "student_sample_exposures": 1792000
      },
      {
        "step": 29000,
        "training_loss": 0.036415971040725706,
        "calibration_BCE": 0.03919966169415098,
        "elapsed_seconds": 316.8843333721161,
        "student_sample_exposures": 1856000
      },
      {
        "step": 30000,
        "training_loss": 0.03579280798882246,
        "calibration_BCE": 0.04021108342950994,
        "elapsed_seconds": 327.81325674057007,
        "student_sample_exposures": 1920000
      },
      {
        "step": 31000,
        "training_loss": 0.03656659733783454,
        "calibration_BCE": 0.03989876588763613,
        "elapsed_seconds": 338.726722240448,
        "student_sample_exposures": 1984000
      },
      {
        "step": 32000,
        "training_loss": 0.0352510489099659,
        "calibration_BCE": 0.04028707152857925,
        "elapsed_seconds": 349.65415954589844,
        "student_sample_exposures": 2048000
      },
      {
        "step": 33000,
        "training_loss": 0.03559268483426422,
        "calibration_BCE": 0.038650829569498696,
        "elapsed_seconds": 360.57207894325256,
        "student_sample_exposures": 2112000
      },
      {
        "step": 34000,
        "training_loss": 0.034726914554834364,
        "calibration_BCE": 0.03815421662764116,
        "elapsed_seconds": 371.5057632923126,
        "student_sample_exposures": 2176000
      },
      {
        "step": 35000,
        "training_loss": 0.03394595278147608,
        "calibration_BCE": 0.039735581461588544,
        "elapsed_seconds": 382.4217257499695,
        "student_sample_exposures": 2240000
      },
      {
        "step": 36000,
        "training_loss": 0.03503098382987082,
        "calibration_BCE": 0.038766286237312084,
        "elapsed_seconds": 393.3546133041382,
        "student_sample_exposures": 2304000
      },
      {
        "step": 37000,
        "training_loss": 0.034565600630361584,
        "calibration_BCE": 0.037808057287967566,
        "elapsed_seconds": 404.2735331058502,
        "student_sample_exposures": 2368000
      },
      {
        "step": 38000,
        "training_loss": 0.03483885816950351,
        "calibration_BCE": 0.03856028315226237,
        "elapsed_seconds": 415.201682806015,
        "student_sample_exposures": 2432000
      },
      {
        "step": 39000,
        "training_loss": 0.03434809660259634,
        "calibration_BCE": 0.0374253630435828,
        "elapsed_seconds": 426.12023544311523,
        "student_sample_exposures": 2496000
      },
      {
        "step": 40000,
        "training_loss": 0.033923412117175755,
        "calibration_BCE": 0.03786840563687411,
        "elapsed_seconds": 437.06481432914734,
        "student_sample_exposures": 2560000
      },
      {
        "step": 41000,
        "training_loss": 0.03447652484010905,
        "calibration_BCE": 0.037599951056278114,
        "elapsed_seconds": 447.9879689216614,
        "student_sample_exposures": 2624000
      },
      {
        "step": 42000,
        "training_loss": 0.03338775422330946,
        "calibration_BCE": 0.03796807232481061,
        "elapsed_seconds": 458.9245195388794,
        "student_sample_exposures": 2688000
      },
      {
        "step": 43000,
        "training_loss": 0.03349443023931235,
        "calibration_BCE": 0.03770690543434837,
        "elapsed_seconds": 469.9022183418274,
        "student_sample_exposures": 2752000
      },
      {
        "step": 44000,
        "training_loss": 0.03414507790282369,
        "calibration_BCE": 0.03759942578402432,
        "elapsed_seconds": 480.8608524799347,
        "student_sample_exposures": 2816000
      },
      {
        "step": 45000,
        "training_loss": 0.03345363231142983,
        "calibration_BCE": 0.03755358042861476,
        "elapsed_seconds": 491.7864570617676,
        "student_sample_exposures": 2880000
      },
      {
        "step": 46000,
        "training_loss": 0.033303928177803754,
        "calibration_BCE": 0.037348759252374825,
        "elapsed_seconds": 502.7005527019501,
        "student_sample_exposures": 2944000
      },
      {
        "step": 47000,
        "training_loss": 0.0338088398873806,
        "calibration_BCE": 0.03735476305412524,
        "elapsed_seconds": 513.6179528236389,
        "student_sample_exposures": 3008000
      },
      {
        "step": 48000,
        "training_loss": 0.03362353353621438,
        "calibration_BCE": 0.037375614628647316,
        "elapsed_seconds": 524.5355114936829,
        "student_sample_exposures": 3072000
      }
    ],
    "calibration_roof_retention": 0.9950007587240762,
    "calibration_contamination_rejection": 0.9638815259826359,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_20000_long",
    "seed": 20261101,
    "sha256": "13cc96011f29cdf232d6636dd5336503eaa1354ed57f56b9f640da44b739afcb",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 20000,
      "steps": 48000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 48000,
    "sample_exposures": 3072000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 526.6132478713989,
    "best_step": 45000,
    "best_calibration_BCE": 0.03683700714111328,
    "cutoff": 0.45084288716316223,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.645486592,
    "common_static_gpu_data_GB": 4.649701888,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14400894006341697,
        "calibration_BCE": 0.11012377522786458,
        "elapsed_seconds": 10.959383726119995,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.09265443543344735,
        "calibration_BCE": 0.0951572578198982,
        "elapsed_seconds": 21.916066646575928,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.0760243616644293,
        "calibration_BCE": 0.08020212499445135,
        "elapsed_seconds": 32.87214207649231,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06842040880210698,
        "calibration_BCE": 0.0697271355600068,
        "elapsed_seconds": 43.83313322067261,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.062162517732009294,
        "calibration_BCE": 0.066114493375836,
        "elapsed_seconds": 54.78930377960205,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.05835020266938955,
        "calibration_BCE": 0.06138746310147372,
        "elapsed_seconds": 65.74040913581848,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05641315468214452,
        "calibration_BCE": 0.0615469589695786,
        "elapsed_seconds": 76.69901871681213,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05329497506096959,
        "calibration_BCE": 0.05751758792761601,
        "elapsed_seconds": 87.66286873817444,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05219130248762667,
        "calibration_BCE": 0.05331236537586559,
        "elapsed_seconds": 98.6217474937439,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.04958526301942766,
        "calibration_BCE": 0.04994811038392963,
        "elapsed_seconds": 109.57458853721619,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.047718880188651384,
        "calibration_BCE": 0.049137185715184065,
        "elapsed_seconds": 120.53559398651123,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.047247396122664216,
        "calibration_BCE": 0.05250357090343129,
        "elapsed_seconds": 131.48750805854797,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04652871971018612,
        "calibration_BCE": 0.04859225114764589,
        "elapsed_seconds": 142.4646918773651,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.045114806218072775,
        "calibration_BCE": 0.048745335272586704,
        "elapsed_seconds": 153.42461848258972,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04437570920493454,
        "calibration_BCE": 0.04726603731097597,
        "elapsed_seconds": 164.4038589000702,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.0435068437140435,
        "calibration_BCE": 0.046882987976074215,
        "elapsed_seconds": 175.4221851825714,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.0426026165811345,
        "calibration_BCE": 0.04684801464658795,
        "elapsed_seconds": 186.3808147907257,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.041525493824854494,
        "calibration_BCE": 0.04782751545761571,
        "elapsed_seconds": 197.34083080291748,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.04165248896926641,
        "calibration_BCE": 0.046686941690155956,
        "elapsed_seconds": 208.2975115776062,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.04123760489281267,
        "calibration_BCE": 0.045722033830122515,
        "elapsed_seconds": 219.2581808567047,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.04122429393325001,
        "calibration_BCE": 0.04267613682602391,
        "elapsed_seconds": 230.20925307273865,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.039838781165890395,
        "calibration_BCE": 0.04290766546075994,
        "elapsed_seconds": 241.1868190765381,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03928023054357618,
        "calibration_BCE": 0.04276575155547171,
        "elapsed_seconds": 252.13726711273193,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.0394664687840268,
        "calibration_BCE": 0.041808754429672704,
        "elapsed_seconds": 263.1055507659912,
        "student_sample_exposures": 1536000
      },
      {
        "step": 25000,
        "training_loss": 0.03813523243553937,
        "calibration_BCE": 0.04312865364768288,
        "elapsed_seconds": 274.0572202205658,
        "student_sample_exposures": 1600000
      },
      {
        "step": 26000,
        "training_loss": 0.038002657934091987,
        "calibration_BCE": 0.040020515164462006,
        "elapsed_seconds": 285.01635575294495,
        "student_sample_exposures": 1664000
      },
      {
        "step": 27000,
        "training_loss": 0.03765468009794131,
        "calibration_BCE": 0.04085634335604581,
        "elapsed_seconds": 295.96808791160583,
        "student_sample_exposures": 1728000
      },
      {
        "step": 28000,
        "training_loss": 0.03644514210522175,
        "calibration_BCE": 0.042346451383648495,
        "elapsed_seconds": 306.94270062446594,
        "student_sample_exposures": 1792000
      },
      {
        "step": 29000,
        "training_loss": 0.03727701268903911,
        "calibration_BCE": 0.0396341484301018,
        "elapsed_seconds": 317.90061497688293,
        "student_sample_exposures": 1856000
      },
      {
        "step": 30000,
        "training_loss": 0.03640360812470317,
        "calibration_BCE": 0.04031932289817117,
        "elapsed_seconds": 328.85310983657837,
        "student_sample_exposures": 1920000
      },
      {
        "step": 31000,
        "training_loss": 0.03632498892955482,
        "calibration_BCE": 0.03974931055704753,
        "elapsed_seconds": 339.81102299690247,
        "student_sample_exposures": 1984000
      },
      {
        "step": 32000,
        "training_loss": 0.03616737641394138,
        "calibration_BCE": 0.039125617934718276,
        "elapsed_seconds": 350.7670111656189,
        "student_sample_exposures": 2048000
      },
      {
        "step": 33000,
        "training_loss": 0.03585398575849831,
        "calibration_BCE": 0.04016082950938832,
        "elapsed_seconds": 361.72899889945984,
        "student_sample_exposures": 2112000
      },
      {
        "step": 34000,
        "training_loss": 0.035513473505154255,
        "calibration_BCE": 0.038144875196977096,
        "elapsed_seconds": 372.6812033653259,
        "student_sample_exposures": 2176000
      },
      {
        "step": 35000,
        "training_loss": 0.03510814919415861,
        "calibration_BCE": 0.038155342448841445,
        "elapsed_seconds": 383.6380739212036,
        "student_sample_exposures": 2240000
      },
      {
        "step": 36000,
        "training_loss": 0.03461709386669099,
        "calibration_BCE": 0.037857988993326826,
        "elapsed_seconds": 394.5922660827637,
        "student_sample_exposures": 2304000
      },
      {
        "step": 37000,
        "training_loss": 0.03453211886808276,
        "calibration_BCE": 0.037409004003351384,
        "elapsed_seconds": 405.5481770038605,
        "student_sample_exposures": 2368000
      },
      {
        "step": 38000,
        "training_loss": 0.03440155649185181,
        "calibration_BCE": 0.03868581524473248,
        "elapsed_seconds": 416.49704360961914,
        "student_sample_exposures": 2432000
      },
      {
        "step": 39000,
        "training_loss": 0.0350863098250702,
        "calibration_BCE": 0.03791126681241122,
        "elapsed_seconds": 427.45160722732544,
        "student_sample_exposures": 2496000
      },
      {
        "step": 40000,
        "training_loss": 0.0339346006270498,
        "calibration_BCE": 0.03739382802789862,
        "elapsed_seconds": 438.3980894088745,
        "student_sample_exposures": 2560000
      },
      {
        "step": 41000,
        "training_loss": 0.03450629975972697,
        "calibration_BCE": 0.03718557625972863,
        "elapsed_seconds": 449.3539009094238,
        "student_sample_exposures": 2624000
      },
      {
        "step": 42000,
        "training_loss": 0.03388644312601537,
        "calibration_BCE": 0.037223826090494794,
        "elapsed_seconds": 460.3114581108093,
        "student_sample_exposures": 2688000
      },
      {
        "step": 43000,
        "training_loss": 0.034140037395991384,
        "calibration_BCE": 0.036944295594186494,
        "elapsed_seconds": 471.26722478866577,
        "student_sample_exposures": 2752000
      },
      {
        "step": 44000,
        "training_loss": 0.034095500560943036,
        "calibration_BCE": 0.037004944772431346,
        "elapsed_seconds": 482.2301790714264,
        "student_sample_exposures": 2816000
      },
      {
        "step": 45000,
        "training_loss": 0.034345763575285675,
        "calibration_BCE": 0.03683700714111328,
        "elapsed_seconds": 493.1849181652069,
        "student_sample_exposures": 2880000
      },
      {
        "step": 46000,
        "training_loss": 0.03375078681576997,
        "calibration_BCE": 0.03696156882083777,
        "elapsed_seconds": 504.1385052204132,
        "student_sample_exposures": 2944000
      },
      {
        "step": 47000,
        "training_loss": 0.03393845059443265,
        "calibration_BCE": 0.03698477325439453,
        "elapsed_seconds": 515.0981261730194,
        "student_sample_exposures": 3008000
      },
      {
        "step": 48000,
        "training_loss": 0.03317107624467462,
        "calibration_BCE": 0.03698377391930782,
        "elapsed_seconds": 526.0501909255981,
        "student_sample_exposures": 3072000
      }
    ],
    "calibration_roof_retention": 0.9950029820913326,
    "calibration_contamination_rejection": 0.9646687524600828,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_20000_long",
    "seed": 20261102,
    "sha256": "ad2a7f523b694e5a7b7b6c28116b87c31f533d24ccbfb47fb05a4b1e7d6e33d0",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 20000,
      "steps": 48000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 48000,
    "sample_exposures": 3072000,
    "equivalent_training_epochs": 12.8,
    "elapsed_seconds": 526.3583545684814,
    "best_step": 43000,
    "best_calibration_BCE": 0.037310676297274505,
    "cutoff": 0.4330222010612488,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.645486592,
    "common_static_gpu_data_GB": 4.649701888,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14600371288880706,
        "calibration_BCE": 0.10699012469667377,
        "elapsed_seconds": 10.945021629333496,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.0919544935412705,
        "calibration_BCE": 0.08717123422333689,
        "elapsed_seconds": 21.91150975227356,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07879571634531021,
        "calibration_BCE": 0.08350349412397905,
        "elapsed_seconds": 32.853522300720215,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.07126625701971352,
        "calibration_BCE": 0.07104884761463512,
        "elapsed_seconds": 43.81513714790344,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06407297712564468,
        "calibration_BCE": 0.07777293294270833,
        "elapsed_seconds": 54.75468349456787,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.06068794989958406,
        "calibration_BCE": 0.06956223530624853,
        "elapsed_seconds": 65.70992040634155,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05719640835188329,
        "calibration_BCE": 0.059107509636156484,
        "elapsed_seconds": 76.66134452819824,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05455209216848016,
        "calibration_BCE": 0.05527207126039447,
        "elapsed_seconds": 87.61708402633667,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05292865718249232,
        "calibration_BCE": 0.05849813292532256,
        "elapsed_seconds": 98.57449269294739,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.04985620373580605,
        "calibration_BCE": 0.05463247687599876,
        "elapsed_seconds": 109.52929019927979,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04865563050750643,
        "calibration_BCE": 0.052856536888353754,
        "elapsed_seconds": 120.48012399673462,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04776377185340971,
        "calibration_BCE": 0.05295496652776545,
        "elapsed_seconds": 131.4374659061432,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.047020517888478934,
        "calibration_BCE": 0.04972200934670188,
        "elapsed_seconds": 142.39711260795593,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04524190576002002,
        "calibration_BCE": 0.04800577797167229,
        "elapsed_seconds": 153.36064338684082,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.0442129880450666,
        "calibration_BCE": 0.04699563307328658,
        "elapsed_seconds": 164.3335292339325,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.04488227933458984,
        "calibration_BCE": 0.04627803060358221,
        "elapsed_seconds": 175.27648496627808,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.04359927797596902,
        "calibration_BCE": 0.046669832980994023,
        "elapsed_seconds": 186.2358660697937,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.04239877405017614,
        "calibration_BCE": 0.04632655762181138,
        "elapsed_seconds": 197.18196535110474,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.041800629059784114,
        "calibration_BCE": 0.04308612804990826,
        "elapsed_seconds": 208.1402223110199,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.04086409065965563,
        "calibration_BCE": 0.045131033371434065,
        "elapsed_seconds": 219.08595085144043,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03987252361420542,
        "calibration_BCE": 0.042971377447879676,
        "elapsed_seconds": 230.0478868484497,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.04046444169152528,
        "calibration_BCE": 0.04292362511374734,
        "elapsed_seconds": 240.9968614578247,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03967911087255925,
        "calibration_BCE": 0.04181933242335464,
        "elapsed_seconds": 251.95659613609314,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.038674652182497084,
        "calibration_BCE": 0.04310907909508908,
        "elapsed_seconds": 262.9141535758972,
        "student_sample_exposures": 1536000
      },
      {
        "step": 25000,
        "training_loss": 0.03870199600979686,
        "calibration_BCE": 0.04880653351292465,
        "elapsed_seconds": 273.8682656288147,
        "student_sample_exposures": 1600000
      },
      {
        "step": 26000,
        "training_loss": 0.03817018520180136,
        "calibration_BCE": 0.04208195941809452,
        "elapsed_seconds": 284.82796359062195,
        "student_sample_exposures": 1664000
      },
      {
        "step": 27000,
        "training_loss": 0.03786323356535286,
        "calibration_BCE": 0.041487061494769474,
        "elapsed_seconds": 295.7814943790436,
        "student_sample_exposures": 1728000
      },
      {
        "step": 28000,
        "training_loss": 0.03747109745349735,
        "calibration_BCE": 0.04090166357791785,
        "elapsed_seconds": 306.7384397983551,
        "student_sample_exposures": 1792000
      },
      {
        "step": 29000,
        "training_loss": 0.036735354974865915,
        "calibration_BCE": 0.04118718807336056,
        "elapsed_seconds": 317.69139885902405,
        "student_sample_exposures": 1856000
      },
      {
        "step": 30000,
        "training_loss": 0.03712100812327117,
        "calibration_BCE": 0.03878761723836263,
        "elapsed_seconds": 328.6477298736572,
        "student_sample_exposures": 1920000
      },
      {
        "step": 31000,
        "training_loss": 0.03641202921885997,
        "calibration_BCE": 0.0399912620082046,
        "elapsed_seconds": 339.5987298488617,
        "student_sample_exposures": 1984000
      },
      {
        "step": 32000,
        "training_loss": 0.03563737741857767,
        "calibration_BCE": 0.039777735369133226,
        "elapsed_seconds": 350.5549576282501,
        "student_sample_exposures": 2048000
      },
      {
        "step": 33000,
        "training_loss": 0.035780455375788736,
        "calibration_BCE": 0.04014132528594046,
        "elapsed_seconds": 361.5033440589905,
        "student_sample_exposures": 2112000
      },
      {
        "step": 34000,
        "training_loss": 0.035746434629429134,
        "calibration_BCE": 0.03848984344944809,
        "elapsed_seconds": 372.4567017555237,
        "student_sample_exposures": 2176000
      },
      {
        "step": 35000,
        "training_loss": 0.03453750449931249,
        "calibration_BCE": 0.03916256239919951,
        "elapsed_seconds": 383.41654658317566,
        "student_sample_exposures": 2240000
      },
      {
        "step": 36000,
        "training_loss": 0.0351005119998008,
        "calibration_BCE": 0.037866981321392636,
        "elapsed_seconds": 394.3662066459656,
        "student_sample_exposures": 2304000
      },
      {
        "step": 37000,
        "training_loss": 0.03454652946814895,
        "calibration_BCE": 0.0384487787882487,
        "elapsed_seconds": 405.32563638687134,
        "student_sample_exposures": 2368000
      },
      {
        "step": 38000,
        "training_loss": 0.03489812805876136,
        "calibration_BCE": 0.038297389845414594,
        "elapsed_seconds": 416.2707359790802,
        "student_sample_exposures": 2432000
      },
      {
        "step": 39000,
        "training_loss": 0.03422501621162519,
        "calibration_BCE": 0.037969224710175484,
        "elapsed_seconds": 427.2317855358124,
        "student_sample_exposures": 2496000
      },
      {
        "step": 40000,
        "training_loss": 0.03456805548723787,
        "calibration_BCE": 0.03756833410552054,
        "elapsed_seconds": 438.1782035827637,
        "student_sample_exposures": 2560000
      },
      {
        "step": 41000,
        "training_loss": 0.03429787206184119,
        "calibration_BCE": 0.03759500147501628,
        "elapsed_seconds": 449.1357169151306,
        "student_sample_exposures": 2624000
      },
      {
        "step": 42000,
        "training_loss": 0.03420531074143946,
        "calibration_BCE": 0.037918478139241533,
        "elapsed_seconds": 460.08338713645935,
        "student_sample_exposures": 2688000
      },
      {
        "step": 43000,
        "training_loss": 0.033509743517264726,
        "calibration_BCE": 0.037310676297274505,
        "elapsed_seconds": 471.0448868274689,
        "student_sample_exposures": 2752000
      },
      {
        "step": 44000,
        "training_loss": 0.03359613016527146,
        "calibration_BCE": 0.037869802370938385,
        "elapsed_seconds": 481.99587082862854,
        "student_sample_exposures": 2816000
      },
      {
        "step": 45000,
        "training_loss": 0.0339873111564666,
        "calibration_BCE": 0.03785188346631599,
        "elapsed_seconds": 492.9583201408386,
        "student_sample_exposures": 2880000
      },
      {
        "step": 46000,
        "training_loss": 0.03395424574334174,
        "calibration_BCE": 0.0377517314564098,
        "elapsed_seconds": 503.91840171813965,
        "student_sample_exposures": 2944000
      },
      {
        "step": 47000,
        "training_loss": 0.03356793379969895,
        "calibration_BCE": 0.03764287322073272,
        "elapsed_seconds": 514.8745768070221,
        "student_sample_exposures": 3008000
      },
      {
        "step": 48000,
        "training_loss": 0.03376448848284781,
        "calibration_BCE": 0.03765057084054658,
        "elapsed_seconds": 525.8321888446808,
        "student_sample_exposures": 3072000
      }
    ],
    "calibration_roof_retention": 0.9950040937749608,
    "calibration_contamination_rejection": 0.9645557818930776,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_20000",
    "seed": 20261031,
    "sha256": "f87a713f5a91553152c1ce0c344d00731978a52cf9a7a2a3262ca55a67c21ab3",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 20000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 6.4,
    "elapsed_seconds": 262.67980337142944,
    "best_step": 23000,
    "best_calibration_BCE": 0.04245153350830078,
    "cutoff": 0.37983831763267517,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.645486592,
    "common_static_gpu_data_GB": 4.649701888,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14522750191017986,
        "calibration_BCE": 0.11733135935465495,
        "elapsed_seconds": 10.902752876281738,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.09155054268985986,
        "calibration_BCE": 0.08925211456761216,
        "elapsed_seconds": 21.811307191848755,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07693150382302702,
        "calibration_BCE": 0.07425432045676492,
        "elapsed_seconds": 32.74940586090088,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06675494218431413,
        "calibration_BCE": 0.07335308412494082,
        "elapsed_seconds": 43.666648864746094,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06391467861458659,
        "calibration_BCE": 0.06466202961314808,
        "elapsed_seconds": 54.59442210197449,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.05826516239158809,
        "calibration_BCE": 0.06150434970277729,
        "elapsed_seconds": 65.50063729286194,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05569660765491426,
        "calibration_BCE": 0.05684113163803563,
        "elapsed_seconds": 76.42725610733032,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.052657965905964374,
        "calibration_BCE": 0.05525661875406901,
        "elapsed_seconds": 87.34729552268982,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.0510215880041942,
        "calibration_BCE": 0.057493525117816344,
        "elapsed_seconds": 98.27431535720825,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.049590154059231284,
        "calibration_BCE": 0.05342095228252989,
        "elapsed_seconds": 109.19516587257385,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04697867139428854,
        "calibration_BCE": 0.054278437851414535,
        "elapsed_seconds": 120.13139462471008,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.045203985806554554,
        "calibration_BCE": 0.04917877234256629,
        "elapsed_seconds": 131.05152344703674,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04431086658779532,
        "calibration_BCE": 0.04971827700065844,
        "elapsed_seconds": 141.98023462295532,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.043825947823002934,
        "calibration_BCE": 0.048016791372588186,
        "elapsed_seconds": 152.90461230278015,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.042240935306064785,
        "calibration_BCE": 0.04693482666015625,
        "elapsed_seconds": 163.83100509643555,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.04114499049354345,
        "calibration_BCE": 0.044622120782100795,
        "elapsed_seconds": 174.77021074295044,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.039797676289454104,
        "calibration_BCE": 0.04536541336521958,
        "elapsed_seconds": 185.68695735931396,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.03964097294770181,
        "calibration_BCE": 0.044115884838682236,
        "elapsed_seconds": 196.6092779636383,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.03948789996560663,
        "calibration_BCE": 0.04359621205185399,
        "elapsed_seconds": 207.5297064781189,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.038709194309078154,
        "calibration_BCE": 0.042688749047481656,
        "elapsed_seconds": 218.4547791481018,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.038724900670349595,
        "calibration_BCE": 0.04306293778852983,
        "elapsed_seconds": 229.3770990371704,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.03822980948304758,
        "calibration_BCE": 0.04268588439479019,
        "elapsed_seconds": 240.30358409881592,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.038151573865674436,
        "calibration_BCE": 0.04245153350830078,
        "elapsed_seconds": 251.20200157165527,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03847554187243804,
        "calibration_BCE": 0.04249243247292259,
        "elapsed_seconds": 262.1327142715454,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950024262495185,
    "calibration_contamination_rejection": 0.9573934479449465,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_20000",
    "seed": 20261101,
    "sha256": "f7f55cbed7a34b812324ecd0e081ea810ec514d7bbf812285a8b02a59e1bdb7a",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 20000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 6.4,
    "elapsed_seconds": 263.61546087265015,
    "best_step": 22000,
    "best_calibration_BCE": 0.04231256260727391,
    "cutoff": 0.3821414113044739,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.645486592,
    "common_static_gpu_data_GB": 4.649701888,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.1422319802865386,
        "calibration_BCE": 0.1069587885538737,
        "elapsed_seconds": 10.959046363830566,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.09225962433218957,
        "calibration_BCE": 0.1019959743384159,
        "elapsed_seconds": 21.921534776687622,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.0761469120644033,
        "calibration_BCE": 0.07890036056980942,
        "elapsed_seconds": 32.88292479515076,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06867539270594716,
        "calibration_BCE": 0.07248284012765595,
        "elapsed_seconds": 43.84139370918274,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06184017188474536,
        "calibration_BCE": 0.06775879701556581,
        "elapsed_seconds": 54.81467604637146,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.058394565652124586,
        "calibration_BCE": 0.06069024228182706,
        "elapsed_seconds": 65.77752709388733,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.055550687190145255,
        "calibration_BCE": 0.057558899480646304,
        "elapsed_seconds": 76.72868728637695,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.053078704185783865,
        "calibration_BCE": 0.05925056778878877,
        "elapsed_seconds": 87.69219923019409,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05196895165555179,
        "calibration_BCE": 0.05447180397727273,
        "elapsed_seconds": 98.66697454452515,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.048825101720169184,
        "calibration_BCE": 0.05048565560543176,
        "elapsed_seconds": 109.63169145584106,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04729501200653612,
        "calibration_BCE": 0.05349825326168176,
        "elapsed_seconds": 120.58017826080322,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.046276217933744195,
        "calibration_BCE": 0.05145558187311346,
        "elapsed_seconds": 131.5447142124176,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.044724313624203205,
        "calibration_BCE": 0.05219251517093543,
        "elapsed_seconds": 142.49612760543823,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.043613444704562424,
        "calibration_BCE": 0.0473371113170277,
        "elapsed_seconds": 153.46063351631165,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04205425668694079,
        "calibration_BCE": 0.04769669885346384,
        "elapsed_seconds": 164.4245319366455,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.04194052974041551,
        "calibration_BCE": 0.04681234674164743,
        "elapsed_seconds": 175.38945722579956,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.040962710620835424,
        "calibration_BCE": 0.04390335686423562,
        "elapsed_seconds": 186.34147930145264,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.03930382405687124,
        "calibration_BCE": 0.047117876573042436,
        "elapsed_seconds": 197.3233666419983,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.039813084095716475,
        "calibration_BCE": 0.043541961854876896,
        "elapsed_seconds": 208.27272582054138,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.039609447692520915,
        "calibration_BCE": 0.043048753634366124,
        "elapsed_seconds": 219.23872590065002,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03968385188747198,
        "calibration_BCE": 0.042661668950861154,
        "elapsed_seconds": 230.19330477714539,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.038269766669720415,
        "calibration_BCE": 0.04231256260727391,
        "elapsed_seconds": 241.15620946884155,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.038249254334717986,
        "calibration_BCE": 0.042506114427971114,
        "elapsed_seconds": 252.1030490398407,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03860908236261457,
        "calibration_BCE": 0.04245246931133848,
        "elapsed_seconds": 263.07074213027954,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950013145658904,
    "calibration_contamination_rejection": 0.9567560561142644,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_20000",
    "seed": 20261102,
    "sha256": "9f7f29660aba0325fbd6d395a3645d9641286e5ce9d5acdec0835c4fde39d237",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 20000,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 6.4,
    "elapsed_seconds": 263.4929783344269,
    "best_step": 21000,
    "best_calibration_BCE": 0.042494518233790544,
    "cutoff": 0.38537469506263733,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.645486592,
    "common_static_gpu_data_GB": 4.649701888,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14544901887327433,
        "calibration_BCE": 0.1111414243293531,
        "elapsed_seconds": 10.95844841003418,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.09095430112816394,
        "calibration_BCE": 0.0881647685657848,
        "elapsed_seconds": 21.919862985610962,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07808787966705859,
        "calibration_BCE": 0.07879682346690785,
        "elapsed_seconds": 32.88379740715027,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06984682818502187,
        "calibration_BCE": 0.06988125543305368,
        "elapsed_seconds": 43.834948778152466,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06383003518544138,
        "calibration_BCE": 0.0651308210661917,
        "elapsed_seconds": 54.79511833190918,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.059388297447934744,
        "calibration_BCE": 0.0658114919720274,
        "elapsed_seconds": 65.74337553977966,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05620525631960482,
        "calibration_BCE": 0.05883680958603368,
        "elapsed_seconds": 76.70569944381714,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05280439297389239,
        "calibration_BCE": 0.05611024190729315,
        "elapsed_seconds": 87.66040754318237,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05134014326054603,
        "calibration_BCE": 0.0682095991192442,
        "elapsed_seconds": 98.61944007873535,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.04781197534967214,
        "calibration_BCE": 0.06114575847278942,
        "elapsed_seconds": 109.57710814476013,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04760048608947545,
        "calibration_BCE": 0.05101057121970437,
        "elapsed_seconds": 120.54629588127136,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.04561367421783507,
        "calibration_BCE": 0.04933223638823538,
        "elapsed_seconds": 131.50236177444458,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04388695937674492,
        "calibration_BCE": 0.04771775350859671,
        "elapsed_seconds": 142.455650806427,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.042757773743011054,
        "calibration_BCE": 0.05014772794318922,
        "elapsed_seconds": 153.4232587814331,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04260260339453816,
        "calibration_BCE": 0.04560281168619792,
        "elapsed_seconds": 164.37511539459229,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.041940444122068585,
        "calibration_BCE": 0.046896468422629616,
        "elapsed_seconds": 175.3381245136261,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.04097172822523862,
        "calibration_BCE": 0.04442830792051373,
        "elapsed_seconds": 186.28718495368958,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.04019584659673273,
        "calibration_BCE": 0.04379854747887814,
        "elapsed_seconds": 197.24086022377014,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.03979396320600063,
        "calibration_BCE": 0.04378900379990086,
        "elapsed_seconds": 208.189110994339,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.03840256032161415,
        "calibration_BCE": 0.04424080875281132,
        "elapsed_seconds": 219.14892029762268,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03857200311217457,
        "calibration_BCE": 0.042494518233790544,
        "elapsed_seconds": 230.10492086410522,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.039075943347997966,
        "calibration_BCE": 0.042724974476207385,
        "elapsed_seconds": 241.06251668930054,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03859121047751978,
        "calibration_BCE": 0.04249962491122159,
        "elapsed_seconds": 252.02327227592468,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.0383735465053469,
        "calibration_BCE": 0.042502519850297404,
        "elapsed_seconds": 262.97210669517517,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.995005205458589,
    "calibration_contamination_rejection": 0.956362442875541,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_3500",
    "seed": 20261031,
    "sha256": "24fc29d3ead72ffc125079f00543742e3c9b5df9b2ea868ee0b9ab468b3bb32d",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 3500,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 36.57142857142857,
    "elapsed_seconds": 262.59739685058594,
    "best_step": 24000,
    "best_calibration_BCE": 0.04456953786214193,
    "cutoff": 0.3090237081050873,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.643902976,
    "common_static_gpu_data_GB": 4.61456384,
    "peak_above_common_data_GB": 2.029339136,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.1430865341462195,
        "calibration_BCE": 0.10458761827873461,
        "elapsed_seconds": 11.054250240325928,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.09171079206466674,
        "calibration_BCE": 0.09497252167210435,
        "elapsed_seconds": 21.963170528411865,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07679801635816694,
        "calibration_BCE": 0.1177859072598544,
        "elapsed_seconds": 32.88545489311218,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06965915161930025,
        "calibration_BCE": 0.07172674958200166,
        "elapsed_seconds": 43.80142331123352,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06283434389531613,
        "calibration_BCE": 0.07079901310313831,
        "elapsed_seconds": 54.71302342414856,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.058212706796824935,
        "calibration_BCE": 0.05950949986775716,
        "elapsed_seconds": 65.62171697616577,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.055663093596696855,
        "calibration_BCE": 0.061313027422355885,
        "elapsed_seconds": 76.54187297821045,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05344337132014334,
        "calibration_BCE": 0.057379346165512546,
        "elapsed_seconds": 87.45989465713501,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.05022246492654085,
        "calibration_BCE": 0.05776017407503995,
        "elapsed_seconds": 98.37077784538269,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.049533047826960684,
        "calibration_BCE": 0.0574172653429436,
        "elapsed_seconds": 109.29056310653687,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04646423302404583,
        "calibration_BCE": 0.05386284759983872,
        "elapsed_seconds": 120.20352435112,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.045665014990605414,
        "calibration_BCE": 0.051180420129949396,
        "elapsed_seconds": 131.11775827407837,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04367117064539343,
        "calibration_BCE": 0.049185807569099196,
        "elapsed_seconds": 142.0300669670105,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.042581075829453766,
        "calibration_BCE": 0.049934090331106476,
        "elapsed_seconds": 152.9432728290558,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04205286799091846,
        "calibration_BCE": 0.04785734310728131,
        "elapsed_seconds": 163.85707902908325,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.0404174732696265,
        "calibration_BCE": 0.04720630844578599,
        "elapsed_seconds": 174.77386951446533,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.03906682036165148,
        "calibration_BCE": 0.045828508388634885,
        "elapsed_seconds": 185.68467831611633,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.038906998109072444,
        "calibration_BCE": 0.04565668568466649,
        "elapsed_seconds": 196.59982347488403,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.03858320505078882,
        "calibration_BCE": 0.04616705185861299,
        "elapsed_seconds": 207.50846314430237,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.037976592184975747,
        "calibration_BCE": 0.04567384208910393,
        "elapsed_seconds": 218.42388892173767,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.037798582433722916,
        "calibration_BCE": 0.04487678886182381,
        "elapsed_seconds": 229.33551836013794,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.03766638361942023,
        "calibration_BCE": 0.045138475267814865,
        "elapsed_seconds": 240.24070620536804,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03727016547136009,
        "calibration_BCE": 0.04471145350138346,
        "elapsed_seconds": 251.16275262832642,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.0367215295471251,
        "calibration_BCE": 0.04456953786214193,
        "elapsed_seconds": 262.07246828079224,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950040937749608,
    "calibration_contamination_rejection": 0.9536107176960663,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_3500",
    "seed": 20261101,
    "sha256": "26f1426ba1c157800301dc0d1f96801b759021f62c6f47e16465a93c397a18f9",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 3500,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 36.57142857142857,
    "elapsed_seconds": 263.418714761734,
    "best_step": 23000,
    "best_calibration_BCE": 0.04342549977158055,
    "cutoff": 0.3656831681728363,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.643902976,
    "common_static_gpu_data_GB": 4.648118272,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14418254008889198,
        "calibration_BCE": 0.11264303302186908,
        "elapsed_seconds": 10.947663307189941,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.09065904084779322,
        "calibration_BCE": 0.08996378798051313,
        "elapsed_seconds": 21.920541286468506,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07703273613005876,
        "calibration_BCE": 0.074566833449855,
        "elapsed_seconds": 32.862162351608276,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.06937865603528917,
        "calibration_BCE": 0.0733766840616862,
        "elapsed_seconds": 43.817079067230225,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.062305858984589575,
        "calibration_BCE": 0.07476121118718927,
        "elapsed_seconds": 54.76796531677246,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.05937261844146997,
        "calibration_BCE": 0.06244690787575462,
        "elapsed_seconds": 65.72104954719543,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05453647997789085,
        "calibration_BCE": 0.05847512151544745,
        "elapsed_seconds": 76.65904569625854,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.052051538810133934,
        "calibration_BCE": 0.05628012526541045,
        "elapsed_seconds": 87.61608576774597,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.04980698277615011,
        "calibration_BCE": 0.05809323661110618,
        "elapsed_seconds": 98.56778812408447,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.047521264538168906,
        "calibration_BCE": 0.0564483305497603,
        "elapsed_seconds": 109.51222705841064,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04619276504777372,
        "calibration_BCE": 0.05314512650460908,
        "elapsed_seconds": 120.46756935119629,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.045031492460519074,
        "calibration_BCE": 0.05748424518469608,
        "elapsed_seconds": 131.4157998561859,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04418585967924446,
        "calibration_BCE": 0.04697453994750977,
        "elapsed_seconds": 142.36005926132202,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04197509613260627,
        "calibration_BCE": 0.050484948984781904,
        "elapsed_seconds": 153.30906534194946,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.04139090071897954,
        "calibration_BCE": 0.04622607051964962,
        "elapsed_seconds": 164.26449489593506,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.040781143442727624,
        "calibration_BCE": 0.045642696658047766,
        "elapsed_seconds": 175.25359416007996,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.039589201137423514,
        "calibration_BCE": 0.04471400248209635,
        "elapsed_seconds": 186.20905017852783,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.03818069368624128,
        "calibration_BCE": 0.044043969310413704,
        "elapsed_seconds": 197.16233015060425,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.03818504007998854,
        "calibration_BCE": 0.04422131814667673,
        "elapsed_seconds": 208.11805176734924,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.0375990441814065,
        "calibration_BCE": 0.044342158230868255,
        "elapsed_seconds": 219.0702142715454,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03827742714621127,
        "calibration_BCE": 0.04349696488813921,
        "elapsed_seconds": 230.0232846736908,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.036727972080931065,
        "calibration_BCE": 0.043515850460168086,
        "elapsed_seconds": 240.97497296333313,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03691717238537967,
        "calibration_BCE": 0.04342549977158055,
        "elapsed_seconds": 251.9314727783203,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03675861784350127,
        "calibration_BCE": 0.04342598403005889,
        "elapsed_seconds": 262.8939070701599,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950018704077044,
    "calibration_contamination_rejection": 0.9564314143796072,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  },
  {
    "method": "graph_3500",
    "seed": 20261102,
    "sha256": "4f8247c184e086dadb826c0a52384f2e579093e138950ee11031fd94db1d0d41",
    "settings": {
      "kind": "graph",
      "width": 64,
      "buildings": 3500,
      "steps": 24000,
      "normal_weight": 0.0,
      "residual_weight": 0.0,
      "positive_weight": 1.0,
      "density_aug": false,
      "strong_aug": false,
      "consistency": false
    },
    "parameters": 59141,
    "steps": 24000,
    "sample_exposures": 1536000,
    "equivalent_training_epochs": 36.57142857142857,
    "elapsed_seconds": 263.3616271018982,
    "best_step": 23000,
    "best_calibration_BCE": 0.044404150853012546,
    "cutoff": 0.3665897250175476,
    "extra_EMA_teacher_forwards": 0,
    "peak_allocated_gpu_GB": 6.643902976,
    "common_static_gpu_data_GB": 4.648118272,
    "peak_above_common_data_GB": 1.995784704,
    "history": [
      {
        "step": 1000,
        "training_loss": 0.14858174969255925,
        "calibration_BCE": 0.10527253140536222,
        "elapsed_seconds": 10.959201097488403,
        "student_sample_exposures": 64000
      },
      {
        "step": 2000,
        "training_loss": 0.0919202987961471,
        "calibration_BCE": 0.0866590507738518,
        "elapsed_seconds": 21.91260814666748,
        "student_sample_exposures": 128000
      },
      {
        "step": 3000,
        "training_loss": 0.07898506482504308,
        "calibration_BCE": 0.07927950955015241,
        "elapsed_seconds": 32.87973141670227,
        "student_sample_exposures": 192000
      },
      {
        "step": 4000,
        "training_loss": 0.07092903428710998,
        "calibration_BCE": 0.07761004606304747,
        "elapsed_seconds": 43.830846309661865,
        "student_sample_exposures": 256000
      },
      {
        "step": 5000,
        "training_loss": 0.06452612600289286,
        "calibration_BCE": 0.06629949948859937,
        "elapsed_seconds": 54.77626967430115,
        "student_sample_exposures": 320000
      },
      {
        "step": 6000,
        "training_loss": 0.060196304071694615,
        "calibration_BCE": 0.061168978303851505,
        "elapsed_seconds": 65.71967720985413,
        "student_sample_exposures": 384000
      },
      {
        "step": 7000,
        "training_loss": 0.05646337953582406,
        "calibration_BCE": 0.05971338110259085,
        "elapsed_seconds": 76.66339612007141,
        "student_sample_exposures": 448000
      },
      {
        "step": 8000,
        "training_loss": 0.05454106623958796,
        "calibration_BCE": 0.05573190961895567,
        "elapsed_seconds": 87.6065456867218,
        "student_sample_exposures": 512000
      },
      {
        "step": 9000,
        "training_loss": 0.0503259354531765,
        "calibration_BCE": 0.05648680480032256,
        "elapsed_seconds": 98.55854392051697,
        "student_sample_exposures": 576000
      },
      {
        "step": 10000,
        "training_loss": 0.048635640585795045,
        "calibration_BCE": 0.05435037310051195,
        "elapsed_seconds": 109.51820755004883,
        "student_sample_exposures": 640000
      },
      {
        "step": 11000,
        "training_loss": 0.04702708005812019,
        "calibration_BCE": 0.05360040560635653,
        "elapsed_seconds": 120.46884608268738,
        "student_sample_exposures": 704000
      },
      {
        "step": 12000,
        "training_loss": 0.045462767945602535,
        "calibration_BCE": 0.05182128749038234,
        "elapsed_seconds": 131.42399549484253,
        "student_sample_exposures": 768000
      },
      {
        "step": 13000,
        "training_loss": 0.04371734295040369,
        "calibration_BCE": 0.048648173615426726,
        "elapsed_seconds": 142.37506318092346,
        "student_sample_exposures": 832000
      },
      {
        "step": 14000,
        "training_loss": 0.04250950664747506,
        "calibration_BCE": 0.0485898744987719,
        "elapsed_seconds": 153.3304169178009,
        "student_sample_exposures": 896000
      },
      {
        "step": 15000,
        "training_loss": 0.041340775318443775,
        "calibration_BCE": 0.04761577737981623,
        "elapsed_seconds": 164.28310418128967,
        "student_sample_exposures": 960000
      },
      {
        "step": 16000,
        "training_loss": 0.04086997496243566,
        "calibration_BCE": 0.04630733425255978,
        "elapsed_seconds": 175.23157262802124,
        "student_sample_exposures": 1024000
      },
      {
        "step": 17000,
        "training_loss": 0.0389394085323438,
        "calibration_BCE": 0.0457481171579072,
        "elapsed_seconds": 186.17783641815186,
        "student_sample_exposures": 1088000
      },
      {
        "step": 18000,
        "training_loss": 0.03971716374251991,
        "calibration_BCE": 0.045108114115397134,
        "elapsed_seconds": 197.13593244552612,
        "student_sample_exposures": 1152000
      },
      {
        "step": 19000,
        "training_loss": 0.0379885745998472,
        "calibration_BCE": 0.045479250705603394,
        "elapsed_seconds": 208.0899636745453,
        "student_sample_exposures": 1216000
      },
      {
        "step": 20000,
        "training_loss": 0.037792615806683895,
        "calibration_BCE": 0.04645055072668827,
        "elapsed_seconds": 219.03092098236084,
        "student_sample_exposures": 1280000
      },
      {
        "step": 21000,
        "training_loss": 0.03695423005428165,
        "calibration_BCE": 0.044884476679021665,
        "elapsed_seconds": 229.9879434108734,
        "student_sample_exposures": 1344000
      },
      {
        "step": 22000,
        "training_loss": 0.037350229694973675,
        "calibration_BCE": 0.04465610756151604,
        "elapsed_seconds": 240.93369007110596,
        "student_sample_exposures": 1408000
      },
      {
        "step": 23000,
        "training_loss": 0.03731818636739626,
        "calibration_BCE": 0.044404150853012546,
        "elapsed_seconds": 251.89327383041382,
        "student_sample_exposures": 1472000
      },
      {
        "step": 24000,
        "training_loss": 0.03734801404271275,
        "calibration_BCE": 0.044469967905680335,
        "elapsed_seconds": 262.83945536613464,
        "student_sample_exposures": 1536000
      }
    ],
    "calibration_roof_retention": 0.9950018704077044,
    "calibration_contamination_rejection": 0.9560770435483699,
    "forward_arithmetic_accounting": {
      "points_per_cloud": 512,
      "conv_linear_attention_matrix_forward_MACs": 159547392,
      "attention_extra_MACs": 0,
      "multiply_add_equals_two_FLOPs_for_counted_ops": true,
      "excluded": [
        "kNN/cdist/topk",
        "group/layer normalization",
        "activations/pooling",
        "PCA/eigh",
        "softmax",
        "biases",
        "losses",
        "backward/optimizer",
        "data augmentations"
      ],
      "interpretation": "Analytic subset of forward arithmetic; compare with measured wall time and peak GPU memory, not a complete FLOP or efficiency benchmark."
    }
  }
]

7. 真实点云质量预览（附加定性检查）

打开20栋真实AHN5点云＋2023正射RGB交互预览：旋转、缩放、原始class着色、同源teacher屋面对照及坐标复制；完成定性推理时还可切换DL分数与保留/删除点。

这20栋来自source训练城市的教师建筑ID，未参与本轮真实点云训练或独立精度验证；AHN官方class6是生产分类，class1不是树木GT。同源teacher叠加不能证明独立厘米级精度。新增RGB单独用于查看，主模型没有光学输入，影像具体拍摄日未证实。

附加定性探针：完成原始候选点98227个，保留88518个、删除9709个。仅使用源验证冻结的部署模型及原校准阈值，三次固定随机分组平均；补齐点不计投票。它与主实验单组512点推理不同，实际class2地面标量也与模拟训练的教师地面定义不同。分数不解释为已校准的真实屋顶概率；保留或删除数量不能证明筛点正确，更不能证明三维屋面重建精度。

定性探针来源与固定模型记录

{
  "result_sha256": "534bc9aa9b0c65c0f5bea52dac04b9f75ff4b4ff85e8f58adc5b102ff6057ccf",
  "utc": "2026-10-21T08:35:35.700400+00:00",
  "source_selection_path": "source:scaling/reports/source_family_selection.json",
  "source_selection_sha256": "645f41e5b04f418d7155e816705699a883135b0ce6011069b086f8ed847ce99b",
  "primary_method": "graph_20000_long_s20261031",
  "recipe_sha256": "632b5f639390237397796dd1243f8c3c5bea6a1552d2549c3e27be99d3e896e9",
  "scope": "20source-train-city qualitative probe only; no independent GT/validation accuracy; RGB display only",
  "no_local_threshold_or_checkpoint_selection": true,
  "checkpoint_changed_by_target_results": false,
  "status": "complete",
  "fixed_building_denominator": 20,
  "checkpoint_path": "source:scaling/models/graph_20000_long_s20261031.pt",
  "checkpoint_sha256": "0899d4f194dbad868ebf63f267dc6a54e7723491ab96c91347dfe323088d56b1",
  "checkpoint_state_strict_loaded_and_verified": true,
  "unchanged_calibration_cutoff": 0.43206343054771423,
  "GPU_execution_performed": true,
  "candidate_records": 98227,
  "kept": 88518,
  "removed": 9709,
  "score_is_calibrated_real_probability": false,
  "independent_accuracy_evaluated": false,
  "records": [
    {
      "id": "NL.IMBAG.Pand.0546100000003629",
      "status": "complete",
      "source_input_sha256": "fcfe819c8c204bb8129a531d4411c9abe327de321155b75238105b561716cecb",
      "n_original": 6936,
      "kept": 6740,
      "removed": 196,
      "retained_fraction": 0.9717416378316033,
      "score_quantiles": [
        0.04000580310821533,
        0.5152913928031921,
        0.972647488117218,
        0.9998016357421875,
        0.9999942779541016
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000003629.npz",
      "prediction_sha256": "f8d4e7b1b539a5566ace3f1ca5cb3a30d825ec70dbbcb16d13ff72692c1564c8",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000005073",
      "status": "complete",
      "source_input_sha256": "edcd120512149941c6eda51a292700efbcda2aa7e118a5fb0d88b0f2afce3cf4",
      "n_original": 2667,
      "kept": 2667,
      "removed": 0,
      "retained_fraction": 1.0,
      "score_quantiles": [
        0.8547865748405457,
        0.9939291477203369,
        0.9999780654907227,
        0.9999992251396179,
        1.0
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000005073.npz",
      "prediction_sha256": "f9b7dd756fcaba95dea784fc27219322316182f7ef911ef9fa5ab7efb65f54bc",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000004149",
      "status": "complete",
      "source_input_sha256": "20a8f003c5f0c9a6580158e3df3fd20abe10e204015d755ffeae737ea12bec40",
      "n_original": 2935,
      "kept": 1811,
      "removed": 1124,
      "retained_fraction": 0.6170357751277683,
      "score_quantiles": [
        0.022653160616755486,
        0.11955755949020389,
        0.6057978272438049,
        0.9999955296516418,
        0.9999998211860657
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000004149.npz",
      "prediction_sha256": "fb946ad6ca464d3d1928ad5db1ba35c3d9e4ec480c87794c6978ca06ba5b46bc",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000003619",
      "status": "complete",
      "source_input_sha256": "f0f0dfb6b1cee9340cba6c2e5e6e01af44b2fa516dc3a38fdbabdeec08022150",
      "n_original": 6144,
      "kept": 5949,
      "removed": 195,
      "retained_fraction": 0.96826171875,
      "score_quantiles": [
        0.05672616884112358,
        0.5136048257350921,
        0.9967122673988342,
        0.9999529719352722,
        0.9999977946281433
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000003619.npz",
      "prediction_sha256": "d3c71a7bf5f91627e6e8d2c026aef710eed95278cfa50938dd8b0ea744405277",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000000956",
      "status": "complete",
      "source_input_sha256": "9af2513066496a30d98bcde713e6a966744607f8a1ea9ef252949dbb4f29a83c",
      "n_original": 18394,
      "kept": 16106,
      "removed": 2288,
      "retained_fraction": 0.8756116124823312,
      "score_quantiles": [
        0.04978397116065025,
        0.32650002688169477,
        0.7000448405742645,
        0.933272337913513,
        0.9948928952217102
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000000956.npz",
      "prediction_sha256": "8c663f0844219957b52a96686094b2ca59aaf1cc8cdd6077c4895417e98aa568",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000004156",
      "status": "complete",
      "source_input_sha256": "fffb198e6794c25059fd436909184300038247337f95c0255221da19b80a02c8",
      "n_original": 3056,
      "kept": 1548,
      "removed": 1508,
      "retained_fraction": 0.506544502617801,
      "score_quantiles": [
        0.004594551865011454,
        0.05030013807117939,
        0.4394802451133728,
        0.9999998211860657,
        1.0
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000004156.npz",
      "prediction_sha256": "0fb605bb5acddc86f7093ee87c422adecd8b7e80f16671e78d088711ac7a19a1",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000002345",
      "status": "complete",
      "source_input_sha256": "f1aa54d1dcacbc8ee6b9bccae424087c0a42245159dab4327ab172cfd4b92586",
      "n_original": 5013,
      "kept": 4991,
      "removed": 22,
      "retained_fraction": 0.9956114103331338,
      "score_quantiles": [
        0.16743619740009308,
        0.6851834416389465,
        0.9924939274787903,
        0.9999997019767761,
        1.0
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000002345.npz",
      "prediction_sha256": "2c05ac6fbfb1b17df7dca6884df4fe86c8ac6f83dd4e393f8b8bfd27dab45bb2",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000003600",
      "status": "complete",
      "source_input_sha256": "13c62621894fc068227d5a175db6fa82b18befb4462403c06107d91f031d1e68",
      "n_original": 7020,
      "kept": 7020,
      "removed": 0,
      "retained_fraction": 1.0,
      "score_quantiles": [
        0.4509317874908447,
        0.9313598304986954,
        0.9994461536407471,
        0.9999935060739518,
        0.9999997019767761
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000003600.npz",
      "prediction_sha256": "2c718d20d2981e3702b1c606870bf636ea540b98d526adfea2a9cf6bcc458f3e",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000003601",
      "status": "complete",
      "source_input_sha256": "54cb67ee5a97e10b3473a57f3369a5fc90fdbb657e2723588261065649102934",
      "n_original": 6420,
      "kept": 6360,
      "removed": 60,
      "retained_fraction": 0.9906542056074766,
      "score_quantiles": [
        0.1532876193523407,
        0.6277570694684982,
        0.9810330271720886,
        0.999793142080307,
        0.9999991059303284
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000003601.npz",
      "prediction_sha256": "0eeb2298f10dc84940e2ec568511b6a324ac8d13858927f799e5e624ac504350",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000005062",
      "status": "complete",
      "source_input_sha256": "125a1480ef792b84efac16f85adddd68f8fd37846d53c2adbbd191c2aeaf900c",
      "n_original": 2844,
      "kept": 2394,
      "removed": 450,
      "retained_fraction": 0.8417721518987342,
      "score_quantiles": [
        0.11737260967493057,
        0.32350028157234195,
        0.693574845790863,
        0.9921497255563736,
        0.9999845623970032
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000005062.npz",
      "prediction_sha256": "3abcd5b54dec995f9e60f22f82d1f57c1d0932afa775c1376a6711a3e3cf45b0",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000003648",
      "status": "complete",
      "source_input_sha256": "76f200c0e400da62bc38c5761dd43b9e7beed517373b35faf2a42713ba3ccf2c",
      "n_original": 3066,
      "kept": 2731,
      "removed": 335,
      "retained_fraction": 0.8907371167645141,
      "score_quantiles": [
        0.007856331765651703,
        0.2546047568321228,
        0.8780481219291687,
        0.9943911582231522,
        0.9998158812522888
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000003648.npz",
      "prediction_sha256": "2a48fdd6d2a2d8ed0c2cee3b5f9e56c5c10057df52f1a8b978142e7c13394cdf",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000005058",
      "status": "complete",
      "source_input_sha256": "b05d0e76e677980c536c10b18d3110b2751467a67f0398d46cde8ce6879dce5f",
      "n_original": 4504,
      "kept": 2265,
      "removed": 2239,
      "retained_fraction": 0.502886323268206,
      "score_quantiles": [
        8.86033012648113e-06,
        0.00018284444595337846,
        0.4493345320224762,
        0.999988853931427,
        0.9999985098838806
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000005058.npz",
      "prediction_sha256": "c59c6c712c311e97bfeef9e371e9047e8c1ad7bbf5518fb9fb35f584dbcf3fcc",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000005074",
      "status": "complete",
      "source_input_sha256": "78d243bdb22d767833687d5b844d40474379379b3e055720164dd2eeceda56d8",
      "n_original": 2208,
      "kept": 2194,
      "removed": 14,
      "retained_fraction": 0.9936594202898551,
      "score_quantiles": [
        0.1712895780801773,
        0.6925442934036254,
        0.9880383014678955,
        0.9999382883310318,
        0.9999995231628418
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000005074.npz",
      "prediction_sha256": "2f47b9ea68390405023639f860e55bb8fd36eb72d83d48e69d73d5a425487484",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000003746",
      "status": "complete",
      "source_input_sha256": "68238db52047fc3a4ecbfb6b9db8857087c20ccfc1c5e7ef7263a717176a0c3d",
      "n_original": 5134,
      "kept": 5120,
      "removed": 14,
      "retained_fraction": 0.9972730814179976,
      "score_quantiles": [
        0.19499091804027557,
        0.6930528968572617,
        0.9663399159908295,
        0.9999859154224395,
        0.9999997019767761
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000003746.npz",
      "prediction_sha256": "86753b1fed6851dfcae99ba46d67b145fb1e5bdec5d2e9f45fdcc642bd0598a0",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000003976",
      "status": "complete",
      "source_input_sha256": "781990fce929df5c5ef4a6c9fa6c8c05e4660befedef08e372b3d09deb1596ee",
      "n_original": 7012,
      "kept": 6965,
      "removed": 47,
      "retained_fraction": 0.9932972047917855,
      "score_quantiles": [
        0.13005344569683075,
        0.7369957447052002,
        0.9987643659114838,
        0.9999838143587112,
        0.9999998211860657
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000003976.npz",
      "prediction_sha256": "949b4da5b344b3fbfe3ee48867d5791594a1ec993b5c3d3f6e4cbe94d4a31b7e",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000003739",
      "status": "complete",
      "source_input_sha256": "0c72411585c96e0f2ed9c322f5feca0b770fab5b88202002e527a3c5c7135574",
      "n_original": 3541,
      "kept": 3537,
      "removed": 4,
      "retained_fraction": 0.9988703756001129,
      "score_quantiles": [
        0.3568998873233795,
        0.8519836068153381,
        0.9985955357551575,
        0.9999866485595703,
        0.9999998211860657
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000003739.npz",
      "prediction_sha256": "50e79814fdc9a0c442b32ffe6cd47a56079d738f489ed0b0ce247c544217b71c",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000003952",
      "status": "complete",
      "source_input_sha256": "46466be37f59ea4f0d207285cee679e1c5e2843c44c545f66ae6251970ae51c8",
      "n_original": 3743,
      "kept": 3649,
      "removed": 94,
      "retained_fraction": 0.9748864547154689,
      "score_quantiles": [
        0.09620410203933716,
        0.5494783759117127,
        0.9807628989219666,
        0.9998220086097718,
        0.9999976754188538
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000003952.npz",
      "prediction_sha256": "3aa86e64f5618de93ad0697ef3f58b69cb9ed72c61e749ef5d693302161bb310",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000005049",
      "status": "complete",
      "source_input_sha256": "f57c8c26f62a11f6e1ba09eac181d8a5a506939a5f8fcb37c466632fd7d82c13",
      "n_original": 2202,
      "kept": 2198,
      "removed": 4,
      "retained_fraction": 0.9981834695731153,
      "score_quantiles": [
        0.3634984791278839,
        0.6891676843166351,
        0.990991860628128,
        0.9998356491327286,
        0.9999988675117493
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000005049.npz",
      "prediction_sha256": "df4beb6354cc82d5d8bf11b1b476360ff775bda9cf8e95222f24c90d2e96020c",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000002942",
      "status": "complete",
      "source_input_sha256": "08fe4b8c85b9f1d102181c0c7b3333687742318264e983afa3b398336b1461dd",
      "n_original": 2043,
      "kept": 2017,
      "removed": 26,
      "retained_fraction": 0.9872736172295644,
      "score_quantiles": [
        0.14417850971221924,
        0.6536734819412232,
        0.9852738380432129,
        0.9997481286525727,
        0.9999611973762512
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000002942.npz",
      "prediction_sha256": "210341c34f653c4c92a58f7094055d82d8b8fbad29138fa4931fa6488bf3d8d1",
      "XYZ_and_source_indices_original": true
    },
    {
      "id": "NL.IMBAG.Pand.0546100000004178",
      "status": "complete",
      "source_input_sha256": "592ceed4895da2893481e52358c7e0da832de2df151ada9919f5b7172190e47e",
      "n_original": 3345,
      "kept": 2256,
      "removed": 1089,
      "retained_fraction": 0.6744394618834081,
      "score_quantiles": [
        0.022098831832408905,
        0.12789686620235444,
        0.644974410533905,
        0.9999984860420227,
        1.0
      ],
      "all_points_exactly1vote_per_pass": true,
      "prediction_path": "paired_real_feasibility/qualitative_dl/predictions/NL_IMBAG_Pand_0546100000004178.npz",
      "prediction_sha256": "8bd6e2be70e5a350ca39d6ee86844ed1928b0c58ce712e8ce1fe26fed0364094",
      "XYZ_and_source_indices_original": true
    }
  ]
}