Replace dataset with consolidated stellarator release

FORMAT.md

@@ -0,0 +1,151 @@
+---
+language:
+- en
+- zh
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: main/*.parquet
+- config_name: failed
+  data_files:
+  - split: train
+    path: failed/*.parquet
+---
+
+# SII Stellarator Configuration Dataset: Consolidated Format
+
+Schema version: `stellarator-1.1`. This card describes the new offline export,
+not a claim that the existing Hugging Face dataset has already been replaced.
+See `report.json` for this particular export's actual counts and `inventory.json`
+for its source file inventory. Upload and source-license review are separate steps.
+
+## One Row, One Configuration
+
+The primary data is `main/*.parquet`. It includes successful evaluations even when
+physical constraints are violated, plus explicitly labelled unknown-status partial
+evaluations. `failed/*.parquet` contains explicit solver failures with identifiable
+boundaries. These are not train/test splits. The HF `train` label above is only a
+loader convention; no evaluation holdout has been defined.
+
+Identical NFP and numerical Fourier coefficients produce one record across both
+groups. Prefer successful over unknown over failed status, then more finite
+metrics, then deterministic source ordering. Repeated evaluations are not kept as
+separate samples. A selected row is kept whole; missing fields are never filled
+from a different evaluation of the same boundary.
+
+## Boundary
+
+| Column | Type | Meaning |
+| --- | --- | --- |
+| `configuration_id` | string | SHA-256 of canonical NFP, mode grid and float64 coefficients |
+| `schema_version` | string | `stellarator-1.1` |
+| `boundary.n_field_periods` | int64 | Positive integer NFP; different NFP means different configuration |
+| `boundary.r_cos` | list[list[float64]] | Radial cosine coefficients; usually 5 by 9 |
+| `boundary.z_sin` | list[list[float64]] | Vertical sine coefficients; same shape as r_cos |
+
+For shape `(M, 2N+1)`, rows are poloidal modes `m=0..M-1` and columns
+are toroidal indices `n=-N..N`. Standard records use 5 by 9; DP also uses 8 by 15.
+`r_cos[0, n<0]` and `z_sin[0, n<=0]` must be zero under the source convention.
+Only stellarator-symmetric boundaries are represented; nonzero `r_sin` or `z_cos`
+are rejected, not discarded silently. Coefficients are never rescaled by major
+radius or standardized as model features. Recognized smaller centered odd-width
+arrays are padded to at least 5 by 9; larger arrays retain every coefficient,
+with a `higher_fourier_modes_preserved` flag. Flat length-45 arrays are reshaped row-major. Negative zero becomes
+zero. All other float64 values are retained exactly, with no tolerance/rounding.
+
+For the standard 5-by-9 grid, the ID hashes little-endian signed int64 NFP followed by float64 values in row-major
+`r_cos`, then row-major `z_sin` order (`struct.pack('<q90d', ...)`). This also makes
+JSON formatting and equivalent zero padding up to the standard grid irrelevant to deduplication.
+For a larger grid, the prefix is `b'higher-modes\x00'`, followed by NFP/int64,
+row count/int32 and column count/int32, then all coefficients in the same order.
+No higher coefficient is rounded or discarded to match the standard grid.
+
+## Nine Physical Metrics
+
+Every field is nullable float64. Null means unavailable or nonfinite, not zero.
+
+| Column | Meaning / legacy pool name |
+| --- | --- |
+| `metrics.aspect_ratio` | Aspect ratio / `aspect_ratio` |
+| `metrics.edge_rotational_transform_over_n_field_periods` | Edge iota divided by NFP / `iota_nfp` |
+| `metrics.average_triangularity` | Mean triangularity / `triangularity` |
+| `metrics.max_elongation` | Maximum elongation / `elongation` |
+| `metrics.edge_magnetic_mirror_ratio` | Edge magnetic mirror ratio / `mirror` |
+| `metrics.vacuum_well` | Vacuum well / `vac_well` |
+| `metrics.minimum_normalized_magnetic_gradient_scale_length` | L_delta_b / `l_delta_b` |
+| `metrics.qi` | QI diagnostic, not log10(QI) |
+| `metrics.flux_compression_in_regions_of_bad_curvature` | Bad-curvature flux compression / `flux_comp` |
+
+Source metric conventions are retained. Signed iota is not turned into an absolute
+value. Historical BO data that already lost its sign remains positive/absolute-only
+and is labelled accordingly. Its saved `qi_log10` is converted using `10**value`;
+this cannot undo historical clipping and is marked `qi_reconstructed_from_log10`.
+No predicted metric, current constraint score or newly computed physics value is
+substituted for a missing original diagnostic.
+
+## Evaluation, Source and Quality
+
+| Column | Type / interpretation |
+| --- | --- |
+| `evaluation.success` | Nullable bool; null means original status cannot be established |
+| `evaluation.error` | Nullable string; original error or explicit original forward-model failure flag |
+| `evaluation.fidelity` | Nullable string; recorded fidelity only |
+| `evaluation.solver_version` | Nullable string; recorded execution version, otherwise recorded configuration version |
+| `evaluation.feasibility` | Nullable float64; source score, not a unified feasibility definition |
+| `source.name` | Source family selected by the importer |
+| `source.path` | Relative source file path, without local machine root |
+| `source.record` | Parquet zero-based row; JSONL one-based line; nested objects add path components |
+| `quality.iota_representation` | `signed`, `absolute_only` or `unknown` |
+| `quality.metric_count` | Number of finite values among the nine metrics |
+| `quality.flags` | List of transformation, invalid-value or status-inference labels |
+
+Missing solver versions, fidelity and timestamps are not inferred from today's
+code. Pool success may be inferred from its documented source filtering; the flag
+`success_inferred_from_source` distinguishes this from an explicit result status.
+Complete unlabelled historical measurements can have
+`success_inferred_from_complete_metrics`. A partial record with no reliable status
+uses `evaluation_status_unknown`. Invalid scalar metrics carry
+`nonfinite_or_invalid:<metric>`. Unknown iota convention must not be treated as proof
+that its sign was preserved. Metric completeness does not establish fidelity or
+physical feasibility.
+
+`higher_fourier_modes_preserved` requires downstream models to support variable
+mode grids. `legacy_boundary_text_precision` identifies ExLLM arrays stored as
+rounded printed numbers; missing digits cannot be reconstructed. DP raw constraint
+values are retained, but its scaled objective is not assumed to equal raw L_delta_b
+(`scaled_target_not_used_as_raw_ldb`).
+
+## Loading and Filtering
+
+```python
+import pyarrow.dataset as ds
+
+data = ds.dataset('main', format='parquet')
+complete = data.to_table(filter=(ds.field('quality.metric_count') == 9)
+                         & (ds.field('evaluation.success') == True))
+# Request just the original 12 columns for compatible downstream consumers.
+core = complete.select([name for name in complete.column_names
+                        if name.startswith(('boundary.', 'metrics.'))])
+```
+
+For signed-iota training, additionally require `quality.iota_representation ==
+'signed'`. For raw-QI-only studies, exclude records carrying the reconstruction
+flag. `failed` is for failure analysis, not a source of zero-valued physics labels.
+No common `feasibility <= threshold` filter is guaranteed valid across all sources.
+
+## Coverage and Review
+
+The builder recognizes cleaned/raw original Parquet, historical and current
+optimization pools/CMA/sBO records, legacy BO, traditional AL final results, DP
+SQLite/NPZ evaluations, ExLLM stored measurements and
+FusionLeague QI/QI-frontier evaluation records. It excludes leaderboard references,
+quadratic tests, surrogate predictions, latent-only points and opaque binary model
+or optimizer state. Unsupported formats are not interpreted as physical labels.
+
+Input files are individually bounded snapshots, not one global freeze. Malformed
+or unusable records are indexed in `quarantine.jsonl` without fabricating a row;
+all source files should be retained for investigation. `report.json` lists duplicate
+counts, missing data, selected source distribution, read statistics and shard
+checksums. Check those reports before publication. Scientific comparability and
+licenses across sources require review; this export does not grant new permissions.

FORMAT.zh.md

@@ -0,0 +1,84 @@
+# SII 仿星器构型数据集：合并后的统一格式
+
+格式版本：`stellarator-1.1`。本文描述新生成的离线导出，不代表 Hugging Face 上的数据集已经替换。具体数量查看本次导出的 `report.json`，来源清单查看 `inventory.json`。上传和来源许可证审查是另外的步骤。
+
+## 一行一个构型
+
+主数据在 `main/*.parquet`，包含成功评估的可行与不可行点，以及明确标注状态未知的部分评估。`failed/*.parquet` 存储边界可识别的明确求解失败记录。两者不是训练/测试划分；英文卡片中的 HF `train` 只是加载命名，没有定义留出测试集。
+
+NFP 和傅里叶系数数值相同的构型，在两部分中合计只保留一条。优先成功，其次状态未知，最后失败；然后优先有限指标更多的记录，最后按固定来源顺序选择。不把重复评估当作不同样本，也不拿另一条评估的指标补齐当前记录。
+
+## 边界格式
+
+| 列名 | 类型 | 含义 |
+| --- | --- | --- |
+| `configuration_id` | string | 规范化 NFP、模态网格和 float64 系数的 SHA-256 |
+| `schema_version` | string | `stellarator-1.1` |
+| `boundary.n_field_periods` | int64 | 正整数 NFP；NFP 不同视为不同构型 |
+| `boundary.r_cos` | list[list[float64]] | 径向余弦系数，通常为 5 行 9 列 |
+| `boundary.z_sin` | list[list[float64]] | 竖直方向正弦系数，与 r_cos 形状相同 |
+
+对于 `(M, 2N+1)` 矩阵，行对应极向模态 `m=0..M-1`，列对应环向编号 `n=-N..N`。通常为 5 行 9 列，DP 还使用 8 行 15 列。按来源约定，`r_cos[0, n<0]` 和 `z_sin[0, n<=0]` 必须为零。只表示仿星器对称边界；非零 `r_sin`、`z_cos` 会拒收，不会静默丢弃。
+
+不按大半径缩放系数，也不做模型特征标准化。已识别的较小、居中的奇数列矩阵补零到至少 5 行 9 列；更大的矩阵保留所有系数，并标记 `higher_fourier_modes_preserved`。长度 45 的向量按行重排。负零统一为零，其他 float64 数值精确保留，不用容差或四舍五入去重。
+
+标准 5×9 网格的 ID 计算顺序为：小端 int64 NFP，随后按行展开的 `r_cos`，再随后按行展开的 `z_sin`，即 `struct.pack('<q90d', ...)`。因此 JSON 排版不同、等价的补零到标准网格不会制造重复构型。较大网格使用前缀 `b'higher-modes\x00'`，再接 NFP/int64、行数/int32、列数/int32，以及相同顺序的所有系数。不会为了匹配标准网格而四舍五入或丢弃高阶系数。
+
+## 九项物理指标
+
+均为可空 float64。null 表示未提供或非有限数值，不表示零。
+
+| 列名 | 含义 / 旧池字段 |
+| --- | --- |
+| `metrics.aspect_ratio` | 展弦比 / `aspect_ratio` |
+| `metrics.edge_rotational_transform_over_n_field_periods` | 边缘旋转变换除以 NFP / `iota_nfp` |
+| `metrics.average_triangularity` | 平均三角度 / `triangularity` |
+| `metrics.max_elongation` | 最大伸长度 / `elongation` |
+| `metrics.edge_magnetic_mirror_ratio` | 边缘磁镜比 / `mirror` |
+| `metrics.vacuum_well` | 真空磁阱 / `vac_well` |
+| `metrics.minimum_normalized_magnetic_gradient_scale_length` | L_delta_b / `l_delta_b` |
+| `metrics.qi` | QI 诊断值，不是 log10(QI) |
+| `metrics.flux_compression_in_regions_of_bad_curvature` | 不利曲率区磁通压缩 / `flux_comp` |
+
+保留来源的指标约定。原本带符号的 iota 不取绝对值；旧 BO 已丢失符号的数据保留绝对值并标注。其 `qi_log10` 用 `10**value` 反变换，标记 `qi_reconstructed_from_log10`，无法恢复历史截断之前的值。不用预测值、当前约束分数或新算的物理值冒充缺失诊断。
+
+## 评估、来源和质量
+
+| 列名 | 类型 / 解释 |
+| --- | --- |
+| `evaluation.success` | 可空 bool；null 表示无法确认原始状态 |
+| `evaluation.error` | 可空 string；原始报错或明确的原始前向模型失败标记 |
+| `evaluation.fidelity` | 可空 string；仅使用记录中已有精度 |
+| `evaluation.solver_version` | 可空 string；优先已有执行版本，否则使用已有配置版本 |
+| `evaluation.feasibility` | 可空 float64；来源分数，不代表统一可行性定义 |
+| `source.name` | 导入器选定的来源类别 |
+| `source.path` | 相对来源文件路径，不包含本机根目录 |
+| `source.record` | Parquet 从 0 开始的行号；JSONL 从 1 开始的行号；嵌套记录附加位置路径 |
+| `quality.iota_representation` | `signed`、`absolute_only` 或 `unknown` |
+| `quality.metric_count` | 九项指标中有限值的数量 |
+| `quality.flags` | 指标变换、无效值、状态推断等标记列表 |
+
+缺失的求解器版本、精度和时间不从当前代码倒推填写。根据池的已有过滤规则推断成功时标记 `success_inferred_from_source`；完整但没有状态字段的历史指标可以标记 `success_inferred_from_complete_metrics`。状态不明的残缺记录标记 `evaluation_status_unknown`。无效标量指标标记 `nonfinite_or_invalid:<metric>`。iota 约定为 unknown 不代表符号一定被保留。指标齐全也不等于求解精度一致或物理约束可行。
+
+## 读取与筛选
+
+带 `higher_fourier_modes_preserved` 的数据需要下游模型支持变长模态网格。`legacy_boundary_text_precision` 表示 ExLLM 将边界存为打印后的近似数字，无法恢复丢失的小数位。DP 保留约束原值，但不假定其缩放目标就是原始 L_delta_b，并标记 `scaled_target_not_used_as_raw_ldb`。
+
+```python
+import pyarrow.dataset as ds
+
+data = ds.dataset('main', format='parquet')
+complete = data.to_table(filter=(ds.field('quality.metric_count') == 9)
+                         & (ds.field('evaluation.success') == True))
+# 仅选原来的 12 列，兼容已有下游程序。
+core = complete.select([name for name in complete.column_names
+                        if name.startswith(('boundary.', 'metrics.'))])
+```
+
+训练带符号 iota 时，还应要求 `quality.iota_representation == 'signed'`。只研究原始 QI 时，应排除带重建标记的记录。`failed` 用于失败分析，不能把缺失物理指标当作零标签。不能保证统一的 `feasibility <= 某阈值` 在所有来源中都成立。
+
+## 覆盖与发布前检查
+
+当前适配原始及清洗 Parquet、历代及当前优化池/CMA/sBO、早期 BO、传统 AL 最终结果、DP SQLite/NPZ 评估、ExLLM 已保存的测量，以及 FusionLeague QI/QI-frontier 评估。排除榜单参考、quadratic 测试、代理预测、仅潜变量的点，以及不透明的二进制模型或优化器状态。未支持的格式不冒充物理标签。
+
+各源文件分别读取有边界的快照，不是全局同时冻结。损坏或不可用的记录在 `quarantine.jsonl` 定位，不虚构数据；保留源文件供追查。`report.json` 包含重复数、缺失指标、最终保留来源分布、读取统计和分片校验和。发布前需检查这些报告，并审查跨来源科学可比性与许可证；本导出不授予新的数据使用权限。

README.md

@@ -1,137 +1,54 @@
 ---
 license: cc-by-4.0
-language:
-- en
+language: [en, zh]
 pretty_name: SII Stellarator Configuration Dataset
-tags:
-- fusion
-- stellarator
-- quasi-isodynamic
-- scientific-machine-learning
-- tabular
+tags: [fusion, stellarator, quasi-isodynamic, scientific-machine-learning, tabular]
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: main/*.parquet
+- config_name: failed
+  data_files:
+  - split: train
+    path: failed/*.parquet
 ---
 
 # SII Stellarator Configuration Dataset
 
-**Wenyang Li**
-
-[**Shanghai Innovation Institute**](http://www.sii.edu.cn) | PhD student
-
+**Wenyang Li** | PhD student, [Shanghai Innovation Institute](http://www.sii.edu.cn)
 [AI-Driven Controlled Fusion Simulation, Control & Design Lab](https://github.com/SII-AI4Fusion)
+Email: [lwydsg@mail.nankai.edu.cn](mailto:lwydsg@mail.nankai.edu.cn)
+Phone: [+86-156-2003-5216](tel:+86-156-2003-5216)
 
-- **Phone**: [+86-156-2003-5216](tel:+86-156-2003-5216)
-- **Email**: [lwydsg@mail.nankai.edu.cn](mailto:lwydsg@mail.nankai.edu.cn)
-
----
+[中文说明](README.zh.md)
 
 ## Overview
-
-This directory hosts a **stellarator configuration dataset grouped by number
-of field periods (nfp)**. Each sample keeps only three things:
-**Fourier boundary coefficients, nfp, and 9 VMEC evaluation metrics** —
-every field is guaranteed to be non-`None` / non-`NaN`.
-
-Design goals:
-
-1. **Simplified content** — keep only the Fourier coefficients and the 9
-   VMEC metrics that configuration-design optimization actually needs; all
-   other construction parameters / diagnostics / multi-beta-branch
-   evaluations are stripped out.
-2. **Ready for future growth** — more QI (Quasi-Isodynamic) configurations
-   will be folded in later, so a unified, clean, schema-stable baseline
-   makes that painless.
-
-## Source
-
-Derived from a local copy of
-[proxima-fusion/constellaration](https://huggingface.co/datasets/proxima-fusion/constellaration),
-covering `n_field_periods` (nfp) = 1–5. The upstream library has broad
-coverage but not every row has complete VMEC evaluation results; this
-dataset keeps **only rows where every field is fully valid**.
-
-## File layout
-
-One file per nfp. The filename embeds the number of retained samples
-`n{count}`:
-
-| File | Samples | Note |
-| --- | --- | --- |
-| `nfp1_n14607.parquet` | 14,607 | nfp = 1 |
-| `nfp2_n19945.parquet` | 19,945 | nfp = 2 |
-| `nfp3_n68191.parquet` | 68,191 | nfp = 3 |
-| `nfp4_n27798.parquet` | 27,798 | nfp = 4 |
-| `nfp5_n28144.parquet` | 28,144 | nfp = 5 |
-| **Total** | **158,685** | |
-
-Parquet files use snappy compression with a flat column layout — they can
-be loaded column-wise directly.
-
-## Fields
-
-Each row (one sample) has three groups of fields:
-
-### 1. Field-period count `nfp`
-
-- Column: `boundary.n_field_periods`
-- Type: integer
-- Meaning: number of toroidal field periods
-- Values present in this dataset: 1, 2, 3, 4, 5
-
-### 2. Fourier boundary coefficients `boundary.r_cos` / `boundary.z_sin`
-
-- Type: 5×9 float matrix (`list<list<double>>`)
-- Meaning: Fourier expansion coefficients of the stellarator boundary
-  surface along the `(θ, φ)` periodic directions. `r_cos` is the cosine
-  component of R; `z_sin` is the sine component of Z (the configuration is
-  stellarator-symmetric, so only these two terms are retained).
-- Shape convention: row index corresponds to the poloidal mode `m`
-  (0…4, 5 rows); column index corresponds to the toroidal mode `n`
-  (−4…+4, 9 columns).
-
-### 3. VMEC evaluation metrics `metrics.*`
-
-| Column | Meaning |
-| --- | --- |
-| `metrics.aspect_ratio` | Aspect ratio |
-| `metrics.edge_rotational_transform_over_n_field_periods` | Edge iota / nfp (**no abs applied** — original sign preserved) |
-| `metrics.average_triangularity` | Average triangularity |
-| `metrics.max_elongation` | Maximum elongation |
-| `metrics.edge_magnetic_mirror_ratio` | Edge magnetic mirror ratio |
-| `metrics.vacuum_well` | Vacuum magnetic well |
-| `metrics.minimum_normalized_magnetic_gradient_scale_length` | Minimum normalized magnetic gradient scale length |
-| `metrics.qi` | Quasi-isodynamic residual (**no log10 applied** — raw value) |
-| `metrics.flux_compression_in_regions_of_bad_curvature` | Flux compression in regions of bad curvature |
-
-> For generality, `iota` is not abs-ed and `qi` is not log10-transformed —
-> raw values are stored; downstream code applies whatever transform it
-> needs.
-
-## Full schema
-
-Flat column names for straightforward column-wise access:
-
-| Column | Type |
-| --- | --- |
-| `boundary.n_field_periods` | `int64` |
-| `boundary.r_cos` | `list<list<double>>` (5×9) |
-| `boundary.z_sin` | `list<list<double>>` (5×9) |
-| `metrics.aspect_ratio` | `double` |
-| `metrics.edge_rotational_transform_over_n_field_periods` | `double` |
-| `metrics.average_triangularity` | `double` |
-| `metrics.max_elongation` | `double` |
-| `metrics.edge_magnetic_mirror_ratio` | `double` |
-| `metrics.vacuum_well` | `double` |
-| `metrics.minimum_normalized_magnetic_gradient_scale_length` | `double` |
-| `metrics.qi` | `double` |
-| `metrics.flux_compression_in_regions_of_bad_curvature` | `double` |
-
-## Data-quality guarantee
-
-Every sample satisfies:
-
-1. `nfp` is neither `None` nor `NaN`;
-2. The entire 5×9 matrices of `boundary.r_cos` and `boundary.z_sin` contain
-   no `None` / `NaN`;
-3. All **9 VMEC metrics** above are non-`None` and non-`NaN`.
-
-Any sample failing any of these conditions was removed before inclusion.
+This release consolidates stellarator boundary configurations and recorded physics evaluations, covering NFP 1–5. It replaces the previous 158,685-row release.
+The collection snapshot is dated 2026-09-23; schema version: `stellarator-1.1`.
+
+## Sources
+The original release was derived from [proxima-fusion/constellaration](https://huggingface.co/datasets/proxima-fusion/constellaration).
+This update also incorporates recorded sBO/CMA, traditional AL, earlier BO, DP, ExLLM and FusionLeague evaluations.
+Surrogate predictions and external leaderboard reference configurations are excluded.
+
+## Contents
+| Group | Configurations | Parquet files |
+| --- | ---: | ---: |
+| main | 2,761,123 | 56 |
+| failed | 103,482 | 3 |
+
+The main group includes successful but constraint-violating evaluations; 321,768 rows have all nine metrics.
+The failed group contains identifiable configurations with failed evaluations. These groups are not training/test partitions; `train` is only the loader split name.
+
+## Format and Quality
+Original `boundary.*` and `metrics.*` column names are retained, with additional status, provenance and quality fields.
+Exact NFP and Fourier coefficients identify a configuration. One whole evaluation is retained; metrics are never combined across evaluations.
+Missing metrics are null. Most boundaries are 5×9; 28 configurations across both groups retain 8×15 matrices.
+Legacy absolute-only iota, reconstructed QI and limited boundary precision are explicitly flagged.
+Unlike the previous release, this release does not guarantee nine complete metrics or a fixed matrix shape for every row.
+See [the full format specification](FORMAT.md) for fields, conventions and reading examples.
+
+## Limitations
+Historical solver settings and metric availability vary. Successful evaluation does not imply physical feasibility.
+The records were consolidated without rerunning physics solvers. Account for missingness, quality flags and correlated optimization trajectories when constructing benchmarks.

README.zh.md

@@ -0,0 +1,22 @@
+# SII 仿星器构型数据集
+
+[English](README.md)
+
+**Wenyang Li**，上海创智学院博士生。
+[AI-Driven Controlled Fusion Simulation, Control & Design Lab](https://github.com/SII-AI4Fusion)
+联系邮箱：lwydsg@mail.nankai.edu.cn；电话：+86-156-2003-5216。
+许可证：CC BY 4.0。
+
+## 概述与来源
+本版替换此前的 158,685 条数据，收集快照日期为 2026-09-23，格式版本为 `stellarator-1.1`，覆盖 NFP 1–5。
+原版来源于 [proxima-fusion/constellaration](https://huggingface.co/datasets/proxima-fusion/constellaration)；新版合并已有 sBO/CMA、传统 AL、早期 BO、DP、ExLLM 和 FusionLeague 真实评估，不包含代理预测和外部榜单参考构型。
+
+## 数据与格式
+主表 `main/` 共 2,761,123 条、56 个分片，其中 321,768 条九项指标完整；失败表 `failed/` 共 103,482 条、3 个分片。两者不是训练集与测试集，加载时的 `train` 只是分区名称。
+沿用 `boundary.*`、`metrics.*`，增加状态、来源及质量字段。按 NFP 和傅里叶系数精确去重，每个构型保留一整条评估，不跨评估拼接指标。
+缺失指标为 null；大多数边界为 5×9，两表合计有 28 个构型保留 8×15 高阶矩阵。旧 iota 绝对值、重建 QI 和有限边界打印精度均有标记。
+新版不再保证每行九项指标完整或矩阵形状固定。字段、约定和读取示例见[格式说明](FORMAT.zh.md)。
+
+## 使用限制
+主表保留求解成功但违反约束的点，失败构型另存。求解成功不代表物理可行。
+历史求解设置与指标完整程度不同，本次没有重新运行物理求解器。构建评测集时应考虑缺失值、质量标记和优化轨迹内的样本相关性。

nfp2_n19945.parquet → failed/part-00000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09529f07408e67c342cad6f37d1bcc944f955d450a878ad2a7b6791c7c8d9040
-size 13038020
+oid sha256:ffe6de3fdb7b65f88dede778164359384e173679c1bfe3c00a634b82ae993f11
+size 36849003

nfp3_n68191.parquet → failed/part-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:19334b2cfb94bd1c9935883e9c866bdeeccbd19cfe01918ac1f16ff304346439
-size 32037486
+oid sha256:f226c549a71953cf666d10a6ea97b9434af5e3d5b503d09b7c819fb02439e7a8
+size 36731851

nfp1_n14607.parquet → failed/part-00002.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:23f2fb02361c702d2403f33f845b42de6e9f19aac0de3ab72f9652ac317adb9f
-size 8983226
+oid sha256:5564b2de0b4a113c853525b71382f0d9c8f0f1fe0d4f650b3bcc766935856b11
+size 2862091

nfp4_n27798.parquet → main/part-00000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9cebb5a0558767fffa08e11898a0d262414bb71c2624828ab38b7cd78f9350d7
-size 19820402
+oid sha256:221fe3f596f6046dfb80d21c303010721b4fccf4aba499c6c62c44b3162ebb24
+size 40414296

main/part-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39d1b94de74f6553df1eb01a49eedc9c5e99119b1248a539f4c9edaa14a01552
+size 40368824

main/part-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98e252edf6132cb7cb30bd7d15cacb4ad3e4c5bd47d4e3916966448a7c4a6e1e
+size 40382598

main/part-00003.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5f5cb2e508f9e05bce8947434016f2b66783d5c13f78322a897caea25c811d3
+size 40393303

main/part-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec91f072e9ebda28ce2691de1d48a386539853675611719b7e60eda36a030a12
+size 40385720

main/part-00005.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3cdcff136046143e5487ebd6731281c0345887d7fa4e2a23267c35d29162cdb5
+size 40402782

main/part-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62522a57a42e7f7bfeca30782093c16d72502cfa64b62daebf76b6e1265b138f
+size 40361860

main/part-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:016a51c15e18d15356a18afc80af0e730def22c5467a3d7fcd4a5ff11d22e9bf
+size 40388758

main/part-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a49dda83ee36f2601b01681bf9e2cea1a8187aa0fd68c714069bda30d01ed0ef
+size 40378735

main/part-00009.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45614c39eb321dd01e00a1f326e818220400b7e2ad7b550cf5cacb4a301ba1b1
+size 40397444

main/part-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e660fa91aa77bbefbb0cb54fe5acbdcc1410298136a6fade2f5919ec415d984
+size 40387975

main/part-00011.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:707e0a856c02c3a9e2eb3997dde16dfffabafb5d20a329136228f9be0e1140cb
+size 40408186

main/part-00012.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5f6065bdbb64b3d8f8e6af53da743a0be18d82a386d71b0e25b29e26a376596
+size 40368255

main/part-00013.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c150c82ebde465c21c5541a144e52793aa71212749b0f5a43d90296b1e0b0ebd
+size 40382712

main/part-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61c65d7703e1a57b438e91fad6a9a3ef61ad49d6f142e90e6afacd1a3fcdc8fc
+size 40390650

main/part-00015.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e298f64b4cfb613cb1c5d29a5633ee79ea8f7ac491f5d8b18f218727925b85f
+size 40388296

main/part-00016.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e57b3ccaf354d339ab109d85dccb04a2f2edbdc8073e3af725538ad64d4735b6
+size 40397770

main/part-00017.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2036facf26f8c51c4d2653ced3e30f74b32e7fa0b163f244083df92a571acd02
+size 40414131

main/part-00018.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ca89d1bed1feeb11f34047cb434a89e6e5dc2f160466dc0a663d2a7b6cc8f84
+size 40369400

main/part-00019.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b4087229e84d7df804432b73e1efb026efb8802422ae596b9d3191e02012346
+size 40385726

main/part-00020.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e428dd9151649dee40da693590f05e0b82a2465d8d5f7dac03a9f7147dd0603
+size 40359118

main/part-00021.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f272bd98392eb22832499cd0a5ec0c052682d65975468e101c88d9b068199e3
+size 40390954

main/part-00022.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:325cd6eddd28f6efab3a3f3bff14ccc531da03bfc255ba96e73bd2757a75553c
+size 40383714

main/part-00023.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3acfd652875717d2289393766624bc16395170386be334032e7ecc22355147e6
+size 40400585

main/part-00024.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:356cb957b12f06db55bd1d91c9b7bb6523f87affafcf1b79fa49b93d97ffa65f
+size 40380731

main/part-00025.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f08e403e3432b5b57aebf4b9231189d7e2cf2c8f36c537f90b47f8331fe60331
+size 40415846

main/part-00026.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ac078321cea545a40e946eaa4bc1e573caed3f9587ec36046332356d43b6b2b
+size 40396409

main/part-00027.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2660f395979274d7f74d073ff489d5ef4bac8039495ef219ffeb4925f6f8840
+size 40386878

main/part-00028.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4cc9f05f86853c74297bb793fc980e4fe2d1972bf39fa88f34aceb23206bc759
+size 40399272

main/part-00029.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18c6e3c7d04560458f67bc2c3a0f5c531b990e0df2192238e3a3acb094b703e0
+size 40363050

main/part-00030.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:523ee089dfa135b391384d96d4bb76de17e77fb2db0a88d33ce67a03f6c4ea43
+size 40431820

main/part-00031.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53321945cf6fba5e48eb1a5b52d8c25df1dfe844ed34ca438d158bd8e9ffb2b1
+size 40386768

main/part-00032.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ec7e570447b796785b17c62052efc5c4d0318ee9feb56500c8abeda600761f2
+size 40425487

main/part-00033.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5dc57867f8e813dff6b7daaa6f82dea34c9708f734f470a0a1d0c76b026275f
+size 40423823

main/part-00034.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8fc859c5728dd43a28a60e882eb6a69cc2d179f9ed20b5629c2e63c0cfc7598c
+size 40378018

main/part-00035.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16d50732efeb8d25bace2a655b32597f4fdceb3f51d86b368120850184cd3cf5
+size 40410903

main/part-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c21ecd7f8a687502a3047134ab68f72a1c91ff30c76f2db1890d6bd68c95bf4
+size 40383499

main/part-00037.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:899ac1f16fd4601992777bfb6760250a9abab7c0738a219d9c8e89dc1b3eff3a
+size 40392442

main/part-00038.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70740127107cd3b7d57489212fed903a90ab62ade723b8aa945be46d67d1e2b9
+size 40422666

main/part-00039.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84305eeae04c2ee03c56c023b98777e8b91f5c1b7b70284de0da52d3386fcbb5
+size 40376546

main/part-00040.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1444c4a5cf33b8c7453cf1f0edb6468d473ea3acf77594b89afe923426b5fab2
+size 40377176

main/part-00041.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9465a77ea84618187fde8a0c9cf71ef8144798c3441ae1b29b505239fdd26a3c
+size 40382670

main/part-00042.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45e0f80ae87d7b74cd8a097abac284e5daba11fd377306e1be14fc2fcb1010ed
+size 40394247