zhuk26 commited on
Commit
d8864a5
·
verified ·
1 Parent(s): 9e22bd9

Simplify dataset card and remove validation files

Browse files
README.md CHANGED
@@ -41,223 +41,7 @@ configs:
41
 
42
  AutoMetaBench is an evaluation benchmark for **literature search, literature screening, data extraction, and meta-analysis reproduction**. It connects review questions to included-study references, fixed screening pools, extracted quantitative results, and published pooled estimates. The release includes the numerical reference data together with a source-audit ledger and a read-only drift checker.
43
 
44
- - **Dataset:** [zhuk26/AutoMetaBench](https://huggingface.co/datasets/zhuk26/AutoMetaBench)
45
  - **Release version:** `2026-09-29`
46
- - **Project and code:** [zhuke26/AutoMeta](https://github.com/zhuke26/AutoMeta)
47
- - **Source revision:** [`8dfb61c0a69c7d791731b0de55c24ef8c5ef1ba6`](https://github.com/zhuke26/AutoMeta/tree/8dfb61c0a69c7d791731b0de55c24ef8c5ef1ba6/autometabench)
48
-
49
- ## Scope
50
-
51
- | Component | Release contents |
52
- | --- | ---: |
53
- | Review inventory and PICO/PECO questions | 27 reviews |
54
- | Included-study records | 421 review–study records |
55
- | Distinct nonempty included-study PMIDs | 391 |
56
- | Included-study records with no PMID | 23 |
57
- | Included-study records with no abstract | 45 |
58
- | Screening candidates | 81,000 rows: 27 fixed pools of 3,000 |
59
- | Screening reference targets | 421 occurrences at 417 distinct review–candidate positions |
60
- | Reviews with extraction data | 25 |
61
- | Extraction tables and rows | 98 CSV tables; 747 rows |
62
- | Audited scalar numeric extraction cells | 4,850 |
63
- | Fields eligible for extraction accuracy scoring | 4,164 |
64
- | Meta-analysis reference units | 95 units from 24 reviews |
65
- | Units eligible for effect-direction scoring | 94 |
66
-
67
- These counts describe different evaluation scopes. The 27-review inventory does not imply that every review has an extraction table or a meta-analysis reference. PMIDs `41531667` and `41721779` are included in search and screening only. PMID `41939286` contributes three extraction tables containing 16 pollutant-level pooled results, with no meta-analysis unit; those 16 summary rows are excluded from study-level extraction scoring.
68
-
69
- The included-study count is measured from the distributed records. Repeated inclusion of a study in different reviews produces multiple rows. A missing PMID is not a distinct study identifier; excluding missing identifiers gives **391**, correcting the source README's count of 392.
70
-
71
- ## Benchmark tasks
72
-
73
- | Task | Input | Reference and interpretation |
74
- | --- | --- | --- |
75
- | Literature search | A review question and its PICO/PECO elements | Retrieve relevant primary studies and compare them with `included_studies.jsonl`. References are the studies included by the source review, rather than an exhaustive contemporary search of the topic. |
76
- | Literature screening | A review question and its fixed 3,000-record title/abstract pool | Rank candidates against the 1-based positions in the corresponding screening truth file. The fixed pool separates ranking performance from live retrieval performance. |
77
- | Data extraction | The relevant source article and outcome | Compare extracted values with the source-preserving result tables, applying the field and summary-row exclusions below. Source PDFs must be obtained separately. |
78
- | Meta-analysis reproduction | The selected outcome's extracted data and source analysis specification | Compare reproduced estimates and intervals with the selected published pooled result, using the source model and reporting precision. Score effect direction on the 94 applicable units. |
79
-
80
- The release supplies evaluation inputs and references. It defines **no train/development/test partition**. The directory named `validation/` contains provenance, audit evidence, and consistency checks; it is **not a machine-learning validation split**. Use review-level grouping if constructing your own partitions, and account for primary studies shared across reviews. Keep reference answers outside the inputs exposed to systems under evaluation.
81
-
82
- ## Files and schema
83
-
84
- Paths below are relative to the root of this Hugging Face dataset repository. The original extraction and meta-analysis CSVs preserve the numerical files from the linked GitHub source revision.
85
-
86
- | Path | Contents and important fields |
87
- | --- | --- |
88
- | `reviews.jsonl` | 27 review questions. Fields: `review_pmid`, `title`, `abstract`, `population`, `intervention_or_exposure`, `comparator`, `outcome`. All are nonempty strings in this release. |
89
- | `included_studies.jsonl` | 421 rows. Fields: `review_pmid`, integer `record_index`, nullable string `pmid`, `title`, nullable string `abstract`, integer `year`. The index is 1-based within each review. |
90
- | `screening/candidates/<review_pmid>.jsonl` | One fixed 3,000-row pool per review. Fields: `candidate_index`, `review_pmid`, `title`, `abstract`. Candidate records omit PMID. |
91
- | `screening/truth/<review_pmid>.json` | Fields: `review_pmid`, `pool_size`, `truth_input_count`, `truth_unique_count`, `truth_positions`. Positions are 1-based candidate indices. |
92
- | `data_extraction/results/*.csv` | 98 UTF-8 tables, with 30 distinct column schemas retained from the source reviews. Filenames identify the review and, when applicable, the outcome. |
93
- | `meta_analysis/units.csv` | 95 pooled-reference rows with `unit_id`, `review_pmid`, `result_file`, `outcome`, `analysis_type`, `effect_measure`, `pooled_effect`, `ci_lower`, `ci_upper`, `direction`, `i2`, `tau2`. |
94
- | `validation/` | Per-cell evidence, selected meta-analysis references, corrections, source and file hashes, audit documentation, and `validate.py`. |
95
-
96
- `review_pmid` is the common review-level join key. `result_file` joins each meta-analysis unit to its extraction CSV. Screening truth is tied to the original `(review_pmid, candidate_index)` pair: retain those identifiers when ranking or loading a pool. The 417 distinct target positions are counted within their respective reviews; candidate indices repeat across pools. Four included-study occurrences collapse onto already occupied positions, so `truth_input_count` and `truth_unique_count` need not match.
97
-
98
- Screening measures described by the project include Recall@10, Recall@30, Recall@50, the depth required for 95% recall (`k95`), and work saved over sampling at 95% recall (`WSS@95`). Report the metric convention and reference version used with your results.
99
-
100
- ### Extraction scoring
101
-
102
- The extraction CSVs contain 747 rows and 5,748 header-defined cells, all populated. Their 4,850 scalar numeric cells include identifiers and statistics outside the extraction accuracy calculation. The **4,164 scoring fields** comprise 3,891 original result fields and 273 scalar numeric study-characteristic fields merged into those tables.
103
-
104
- To reproduce the documented scoring mask, normalize column names to lowercase and replace runs of non-alphanumeric characters with `_`, then exclude:
105
-
106
- - Identifier columns: `study`, `study_name`, `first_author`, `author`, `year`, `title`, `pollutant`.
107
- - Context columns: `location`, `geographical_setting`, `outcome`, `exposure`, `source_figure`, `outcome_unit`, `model`.
108
- - Columns whose normalized name contains `weight` or `heterogeneity`.
109
- - Computed statistics: `i2`, `tau2`, `tau_2`, `h2`, `h_2`, `q`, `df`, `z`, `z_value`, `p`, `p_value`, and names ending in `_p` or `_p_value`.
110
- - Summary rows whose study label contains `pooled`, `overall`, `subtotal`, `prediction`, `random effects model`, `common effect model`, `fixed effect model`, or `mean effect`; also exclude any row with a nonempty `Pollutant` field.
111
-
112
- Rows may represent studies, study arms, comparisons, or explicitly identified summaries. Do not assume a universal row schema. In particular, a characteristic-table `Sample Size` need not equal the forest plot's outcome-specific arm totals. Per-cell provenance in `validation/extraction_numeric_cells.csv` identifies the source of each numeric value.
113
-
114
- ### Meta-analysis interpretation
115
-
116
- `effect_measure` includes SMD, MD, WMD, beta, OR, RR, and Proportion. The additional file `validation/meta_reference.json` specifies the selected pooled row, source model, field provenance, missingness, and separately reported overall-effect and heterogeneity p-values. The CSV itself has no `model` or `p_value` column.
117
-
118
- - MA-045 and MA-046 use the **common-effect** pooled row. Their reported τ² values describe heterogeneity and are not parameters used for common-effect weighting.
119
- - MA-092–MA-095 use the source's **Bayesian analysis**. A generic frequentist random-effects calculation is not an equivalent reproduction.
120
- - MA-014 is a single-group prevalence estimate with `direction=not_applicable`. Retain its estimate and interval in evaluation, and exclude it from the direction denominator.
121
- - Direction is relative to 1 for OR/RR and 0 for difference measures. It describes the point estimate's direction, not statistical significance or whether an outcome is clinically beneficial.
122
- - Empty I² or τ² cells indicate unpopulated reference fields. Do not convert them to zero. A reported τ is not τ²; the release does not manufacture missing τ² values by squaring a reported τ.
123
- - Compare estimates at the reported precision. Rounded source inputs and differences in estimation methods can prevent exact numerical reconstruction. A printed p-value such as `0.00` represents source rounding.
124
-
125
- ## Construction and source audit
126
-
127
- The project documents a PubMed search using review-level terms such as “systematic review” and “meta-analysis,” a 2026 publication-date filter, and manual screening of the first 200 retrieved records. Eligibility focused on public health, health-related research, and clinical medicine, with a clear question and an identifiable included-study set. Review questions, bibliographic references, and quantitative results were curated from the selected reviews.
128
-
129
- This describes the documented selection procedure. The distributed review schema has no publication-date field, so this card does not independently certify that every review was published in 2026. The selection procedure also does not establish that the material was absent from any model's training data.
130
-
131
- The **2026-09-29 numerical source audit** covers all existing extraction tables and meta-analysis units. Its ledger assigns sources to all 4,850 numeric extraction cells: 4,576 forest-plot values, 272 characteristic-table values, one intervention sample size explicitly calculated as `12 + 12` from two reported table arms, and one publication year from a reference list. All 95 pooled estimate/interval triplets were checked against their selected forest-plot rows.
132
-
133
- The correction log records three extraction changes and 99 meta-analysis field changes across 59 units. The latter comprise 23 numeric discrepancies, 50 previously missing reported values, four τ/τ² labeling corrections, 19 rounding or display alignments, and three direction labels. Re-score results produced with an earlier reference version.
134
-
135
- | Audit file | Purpose |
136
- | --- | --- |
137
- | `extraction_numeric_cells.csv` | Numeric cell, source value, source type, and PDF page/figure/table location. Data-row numbering begins at 1 after the header. |
138
- | `meta_reference.json` | Expected released row plus model details, selected source evidence, and field-level status for each meta-analysis unit. |
139
- | `corrections.csv` | All 102 before/after correction entries and their evidence locations. |
140
- | `outcome_coverage.csv` | Coverage and caveats for all 98 extraction tables. |
141
- | `source_manifest.csv` | All 27 review PMIDs, public source links, source-PDF SHA-256 hashes, and stage coverage. |
142
- | `verified_file_hashes.csv` | SHA-256 hashes of the 98 extraction CSVs and the meta-analysis CSV. |
143
- | `summary.json` and `audit_report_zh.md` | Machine-readable audit summary and detailed Chinese report. |
144
- | `validate.py` | Read-only comparison with the released numerical ledger, row references, and hashes. |
145
-
146
- PDF page numbers refer to physical file pages starting at 1. The publication contains **no source PDFs or reproduced forest-plot images**. Obtain articles through their public source links and the access arrangements applicable to each source.
147
-
148
- ### Known source and metadata limitations
149
-
150
- An accurate transcription can preserve an error or disagreement in a source article. For pooled estimates and intervals, this release selects the documented forest-plot row when it conflicts with a table, abstract, or narrative, and retains the disagreement in the audit files. Examples include:
151
-
152
- - PMID `41652340`: CCL2 is −3.29 in the forest versus −3.639 in Table 2; IL-8 is −0.25 versus −0.284. These differences exceed rounding. The evidence does not establish whether they reflect different posterior summaries.
153
- - PMID `42003421`, SBP: the selected forest interval has an upper bound of 0.69, whereas the article narrative reports 0.96.
154
- - PMID `41639651`: the Baccarelli SBP value is −0.01 in the forest versus −1.01 in the table; the article also has inconsistent study-year labels.
155
- - PMID `42032055`: the Tedeschi characteristic-table sample size is 186, while the forest arm totals are 86 + 103. Both are retained with their respective provenance.
156
-
157
- Some retained heterogeneity values are reported in article text rather than in the selected forest plot. Consult the per-field evidence in `validation/meta_reference.json` for the source and status of each value. An unpopulated field is distinct from a verified numerical zero and from supplementary evidence recorded only in the sidecar.
158
-
159
- The numerical audit does not establish primary-study validity, verify every bibliographic title verbatim, independently reproduce every published calculation, or fill omitted outcomes. Missing PMIDs and abstracts limit identifier-based and abstract-only evaluation. The small, curated review inventory and fixed screening pools do not represent all medical questions or a live PubMed search. The dataset contains bibliographic text and aggregate published results; it contains no individual participant records and is intended for methodological evaluation rather than patient-level decisions.
160
-
161
- ## Dataset Viewer and Python loading
162
-
163
- The six configurations below use the split name `benchmark`. They are derived
164
- Parquet views for convenient browsing and `datasets` loading. Original CSV/JSON
165
- files remain available in their source-preserving directories.
166
-
167
- | Configuration | Rows | Meaning |
168
- | --- | ---: | --- |
169
- | `reviews` (default) | 27 | Review questions and PICO/PECO |
170
- | `included_studies` | 421 | Included-study reference records, preserving missing values |
171
- | `screening_candidates` | 81,000 | Fixed pools; identify each row by review PMID and candidate index |
172
- | `screening_truth` | 27 | One row per review, with the original truth-position list intact |
173
- | `data_extraction` | 5,748 | One original CSV cell per row; preserves field name and exact value string |
174
- | `meta_analysis` | 95 | Pooled references; `null` direction stays a string and blank statistics stay missing |
175
-
176
- ```python
177
- from datasets import load_dataset
178
-
179
- reviews = load_dataset("zhuk26/AutoMetaBench", "reviews", split="benchmark")
180
- meta = load_dataset("zhuk26/AutoMetaBench", "meta_analysis", split="benchmark")
181
- fields = load_dataset("zhuk26/AutoMetaBench", "data_extraction", split="benchmark")
182
- scored_fields = fields.filter(lambda row: row["is_evaluable"])
183
- assert len(scored_fields) == 4164
184
- ```
185
-
186
- The extraction view uses `(result_file, data_row, field)` as its cell key;
187
- `data_row` starts at 1 after the CSV header. `value` is always the exact CSV
188
- string, preserving heterogeneous labels, decimal precision and signed zero.
189
- `has_numeric_audit` and the source columns expose the numerical audit where it
190
- exists. Non-numeric identifiers/context are not represented as numeric audit
191
- passes. The six configurations are different benchmark components, not separate
192
- training or testing partitions. Do not concatenate them into one dataset.
193
-
194
- ### Release metadata corrections
195
-
196
- Hugging Face preparation corrected two audit-annotation issues while preserving
197
- all original numerical data files: MA-091 tau2=0.030 comes from the abstract
198
- rather than Figure 2, and MA-083's original Figure 3e specifies IV, Random.
199
- The corrected meta field totals are 412 forest-verified values, five retained
200
- text/abstract-sourced values, and 58 blank fields. Detailed old/new annotations
201
- are recorded in `validation/release_metadata_corrections.json`.
202
-
203
- ## Download and verify
204
-
205
- Download this repository as a **dataset** with `huggingface_hub`. The following example downloads the files and runs the bundled read-only checker:
206
-
207
- ```python
208
- from pathlib import Path
209
- import subprocess
210
- import sys
211
-
212
- from huggingface_hub import snapshot_download
213
-
214
- benchmark = Path(snapshot_download(
215
- repo_id="zhuk26/AutoMetaBench",
216
- repo_type="dataset",
217
- ))
218
-
219
- subprocess.run(
220
- [sys.executable, str(benchmark / "validation" / "validate.py")],
221
- check=True,
222
- )
223
- ```
224
-
225
- For a reproducible evaluation, pin the Hugging Face commit used by supplying its commit hash as `revision=` and record it alongside the release version. The GitHub source revision above identifies the source release; it is not a Hugging Face repository revision.
226
-
227
- The expected numerical check reports `status: pass`, 98 extraction files, 747 rows, 4,850 numeric cells, 95 meta-analysis units, and no errors. Without local source PDFs, `source_pdfs_hash_checked` is 0. With separately obtained matching PDFs named `<review_pmid>.pdf`, run:
228
-
229
- ```python
230
- subprocess.run(
231
- [
232
- sys.executable,
233
- str(benchmark / "validation" / "validate.py"),
234
- "--source-pdf-dir",
235
- "/path/to/source_pdfs",
236
- ],
237
- check=True,
238
- )
239
- ```
240
-
241
- The optional check compares all 27 local PDF hashes with the manifest. A pass establishes consistency with the recorded reference and source versions; it does not perform a fresh visual audit or demonstrate absence of errors in the original articles. Dataset-viewer previews, when provided, are conveniences for inspection; use the released files and audit references for reproducible scoring.
242
-
243
- ## License and source rights
244
-
245
- The AutoMeta project releases its code and associated dataset materials under the [MIT license](https://github.com/zhuke26/AutoMeta/blob/8dfb61c0a69c7d791731b0de55c24ef8c5ef1ba6/LICENSE). That license applies to the project contributors' rights in their contributions. Bibliographic metadata and abstracts originate from PubMed and the cited publications, and numerical observations are transcribed from published review articles. Third-party text and source publications retain their applicable rights; the project's MIT license does not relicense those works. Source PDFs are not redistributed.
246
-
247
- ## Reference
248
-
249
- To identify this dataset release, cite its name, version, URL, and the Hugging Face revision actually used:
250
-
251
- > AutoMetaBench. Version 2026-09-29. Hugging Face Datasets. https://huggingface.co/datasets/zhuk26/AutoMetaBench. Source revision: 8dfb61c0a69c7d791731b0de55c24ef8c5ef1ba6.
252
-
253
- ```bibtex
254
- @misc{autometabench_20260929,
255
- title = {AutoMetaBench},
256
- year = {2026},
257
- howpublished = {Hugging Face Datasets},
258
- url = {https://huggingface.co/datasets/zhuk26/AutoMetaBench},
259
- note = {Version 2026-09-29; source revision 8dfb61c0a69c7d791731b0de55c24ef8c5ef1ba6}
260
- }
261
- ```
262
-
263
- Please also acknowledge the source reviews relevant to your evaluation. Their PMID links are recorded in `validation/source_manifest.csv`.
 
41
 
42
  AutoMetaBench is an evaluation benchmark for **literature search, literature screening, data extraction, and meta-analysis reproduction**. It connects review questions to included-study references, fixed screening pools, extracted quantitative results, and published pooled estimates. The release includes the numerical reference data together with a source-audit ledger and a read-only drift checker.
43
 
44
+ - **Dataset:** [**zhuk26/AutoMetaBench**](https://huggingface.co/datasets/zhuk26/AutoMetaBench)
45
  - **Release version:** `2026-09-29`
46
+ - **Project and code:** [**zhuke26/AutoMeta**](https://github.com/zhuke26/AutoMeta)
47
+ - **Source revision:** [**`8dfb61c0a69c7d791731b0de55c24ef8c5ef1ba6`**](https://github.com/zhuke26/AutoMeta/tree/8dfb61c0a69c7d791731b0de55c24ef8c5ef1ba6/autometabench)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/README.md DELETED
@@ -1,64 +0,0 @@
1
- # AutoMetaBench source verification
2
-
3
- Audit date: 2026-09-29. See [the full Chinese report](audit_report_zh.md).
4
-
5
- All 98 existing extraction CSVs were reviewed against original PDF forest
6
- plots and explicitly identified characteristic tables. The numeric-cell ledger
7
- covers 4850 numeric values (including numeric identifiers); this is distinct
8
- from the extraction scoring mask of 4164 fields. All 95 pooled effect/interval
9
- triplets were reviewed against their selected forest plot rows.
10
-
11
- ## Files
12
-
13
- - `extraction_numeric_cells.csv`: source values for all numeric extraction cells;
14
- data-row numbers start at 1 after the header. `derived_from_reported_table`
15
- explicitly marks the one combined intervention N, 12 + 12.
16
- - `meta_reference.json`: selected pooled row, per-field source and absence status,
17
- original model/p-value details, and source conflicts. `expected_unit` is the
18
- adjudicated release row; `numeric_fields` preserves source evidence separately.
19
- - `corrections.csv`: applied old/new values with exact source locations.
20
- - `outcome_coverage.csv`: all 98 result files and source caveats.
21
- - `source_manifest.csv`: all 27 source reviews, public PMID URLs, PDF SHA-256
22
- hashes, and stage-specific file counts. PDFs are not redistributed here.
23
- - `verified_file_hashes.csv`: hashes of the 99 reviewed numerical data files.
24
- - `summary.json`: audit coverage and correction counts.
25
- - `validate.py`: read-only comparison against the reviewed source ledger.
26
-
27
- ## Interpretation
28
-
29
- A verified transcription can preserve an inconsistency already present in a
30
- paper. This audit does not establish primary-data validity or independently
31
- reproduce every published calculation. Characteristic sample sizes and forest
32
- outcome-arm sample sizes have separate provenance; do not substitute them.
33
- Five retained heterogeneity values come from results text or the abstract, not forest plots.
34
- The sidecar explicitly identifies them. Missing forest statistics remain blank;
35
- blank is never an inferred zero. No recalculated statistics are substituted
36
- for published reference values.
37
-
38
- The source plot has priority over conflicting abstract/table values for the
39
- pooled effect and interval. Compatible rounding changes are labeled separately
40
- from numeric discrepancies. Model and p-value details are in the JSON sidecar,
41
- not in nonexistent `model`/`p_value` columns of `units.csv`.
42
-
43
- MA-014 is single-group prevalence: direction is `not_applicable`, leaving 94
44
- units applicable to treatment/exposure-effect direction scoring. MA-045/046 use
45
- the common-effect row; their tau2 is a separately reported heterogeneity
46
- estimate, not a common-effect weighting parameter. Re-score evaluations that
47
- used the earlier reference version.
48
-
49
- ## Run
50
-
51
- From the repository root:
52
-
53
- ```bash
54
- python3 validation/validate.py
55
- python3 validation/validate.py --source-pdf-dir /path/to/paper
56
- ```
57
-
58
- The second form also checks all 27 local PDFs named `<review_pmid>.pdf`.
59
- PDF page numbers are one-based physical file pages. A pass detects no drift
60
- from the visually reviewed values; the script does not itself perform a new
61
- visual audit. Source revisions require review and a new recorded audit, not
62
- just refreshing hashes to silence a mismatch.
63
-
64
- HF release preparation corrected MA-091 tau2 provenance and MA-083 model metadata; see `release_metadata_corrections.json`. The original 99 numerical data files remain byte-identical to the GitHub source revision.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/audit_report_zh.md DELETED
@@ -1,225 +0,0 @@
1
- # AutoMetaBench 数值与原始森林图核验报告
2
-
3
- 审计日期:2026-09-29。对象:`data_extraction/results/` 与 `meta_analysis/units.csv` 的当前开源版本。数值依据为27篇本地原始论文 PDF,来源版本见 `source_manifest.csv`。
4
-
5
- ## 结论
6
-
7
- **全量核对完成,已将有明确原文依据的更正写回 benchmark。** 98张表、747行的4,850个数值单元格均已核对,数值未核对项为0。研究/分析行的效应量及置信区间与选定森林图一致;一处附加对照组样本量原来录错,现已修正。95个合并结果的效应量与区间已直接逐图核对。
8
-
9
- “全部数值都来自森林图”这一描述需要修正:273个数值是研究特征表字段(其中1个为两组人数相加),另有1个年份来自参考文献;合并结果中5个原有异质性值仅见于正文或摘要。这些数值的来源已逐项标注,不能把它们标成森林图直接转录。
10
-
11
- 这是**忠实转录与来源一致性核验**。论文内部的图、表、正文冲突已保存,未用重新计算的结果冒充论文数值;本次不能证明原论文中的每个统计计算或原始研究数据本身正确。
12
-
13
- ## 实际覆盖范围
14
-
15
- | 项目 | 核验结果 |
16
- |---|---:|
17
- | 综述文献清单 / 原始PDF | 27 / 27 |
18
- | 有提取表的综述 | 25 |
19
- | 提取表 / 数据行 | 98 / 747 |
20
- | 核验的数值单元格 | 4,850 |
21
- | 直接来源于森林图 | 4,576 |
22
- | 直接来源于研究特征表 | 272 |
23
- | 由特征表明确组人数相加 | 1 |
24
- | 来自参考文献的年份 | 1 |
25
- | 未核对数值 | 0 |
26
- | 合并结果 / 涉及综述 | 95 / 24 |
27
- | 可用于治疗/暴露效应方向评分的单元 | 94 |
28
-
29
- PMID 41531667、41721779在综述清单中,但当前两个目标目录都没有其数据。PMID 41939286有3张污染物层面的合并结果表、没有`units.csv`单元;其16行是汇总结果,不能当成16项原始研究。未擅自补建新结局或扩大任务数据范围。
30
-
31
- 4,850是本次数值审计范围,包含年份和不参与提取准确率评分的统计字段;它不是4,164个评测字段的替代分母。复核原有计数规则后,4,164仍成立:需排除`z_value`和非空`Pollutant`行,README原先遗漏的这两条说明已经补上。
32
-
33
- ## 已写回的更正
34
-
35
- 提取数据更正3项:
36
-
37
- - `39952828.csv`,第11条数据Tsutsumi的`CG (N)`:**10 → 12**。依据PDF第4页Table 1。该行`IG (N)=24`是两个12人干预组之和,已标明推导来源。
38
- - `41065407_PainIntensity.csv`,第3条数据的研究臂名称:`Bruce-Low 2012 (1)` → `Bruce-Low 2012 (2 x week)`。图中该行对应−1.59 [−2.24, −0.95],效应数值未改。
39
- - `41639651_SBP.csv`,第3条数据区间上限:`0.00` → `-0.00`,仅对齐森林图显示,数值相等。
40
-
41
- 合并结果共更改**59/95个单元、99个字段**,分类如下。显示精度调整与真实数值差异分开统计。
42
-
43
- | 类型 | 字段数 |
44
- |---|---:|
45
- | 数值与森林图存在实际差异 | 23 |
46
- | 补齐森林图明确报告而原来留空的I²/τ² | 50 |
47
- | 删除误填进τ²列的τ | 4 |
48
- | 对齐森林图的四舍五入精度 | 16 |
49
- | 仅显示格式调整,数值不变 | 3 |
50
- | 方向标签更正 | 3 |
51
-
52
- 代表性差异如下;全部102条提取及合并结果更正见文末明细和`corrections.csv`。
53
-
54
- | PMID / 结局 | 原值 | 森林图参考值 | 原始PDF位置 |
55
- |---|---|---|---|
56
- | 41065407 / Disability | −2.76 [−3.90, −1.62] | **−2.29 [−3.23, −1.35]** | p10,Figure 3 |
57
- | 42003421 / SBP | CI上限0.96 | **0.69** | p6,Figure 2 |
58
- | 41601519 / DepressiveSymptoms | I²=81% | **90.13%** | p10,Figure 5 |
59
- | 41617837 / Suicide | CI下限1.00 | **1.01** | p10,Figure 4 |
60
- | 41743679 / ArmCircumference | −0.06 | **+0.06** | p11,Figure 4f |
61
- | 41743679 / ThighCircumference | −0.04 | **+0.04** | p11,Figure 4g |
62
- | 41743679 / WaistToHipRatio | I²=100% | **0%** | p11,Figure 4e |
63
- | 41938012 / HighTemperature | I²=89.0% | **98.5%** | p19,Figure 6 |
64
- | 41652340 / CCL2 | −3.639 | **−3.29** | p7,Figure 3 |
65
- | 41652340 / IL-8 | −0.284 | **−0.25** | p7,Figure 3 |
66
-
67
- PMID 41652340正文报告的是τ:3.207、0.142、0.507、0.219,不能直接放入τ²列。本次清空4个错误τ²目标值,原始τ保留在来源说明中;没有自行平方生成新的参考答案。
68
-
69
- MA-014是单组患病率/症状比例0.40 [0.24, 0.58],没有治疗组与对照组比较。原有`direction=negative`改为`not_applicable`。其数值和区间仍可评分,治疗/暴露效应方向评分分母应为94。
70
-
71
- ## 非森林图来源、缺失值和模型选择
72
-
73
- 95个单元共有475个效应值/区间/I²/τ²位置:412个直接核对森林图,5个既有值只在正文或摘要核对,58个为森林图未报告而保留空白。空白不等于0,也不等于未开展核验。
74
-
75
- | 单元 | 字段和值 | 实际来源 | 处理 |
76
- |---|---|---|---|
77
- | MA-027 / 40325863 | I²=91.002,τ²=1.731 | p10,Results 3.5正文 | 保留正确原值,明确`verified_nonforest` |
78
- | MA-011 / 41566983 InfectionLongTerm | I²=99.27 | p4,Results正文 | 保留正确原值,明确`verified_nonforest` |
79
- | MA-012 / 41566983 InfectionShortTerm | I²=76.49 | p4,Results正文 | 保留正确原值,明确`verified_nonforest` |
80
- | MA-091 / 41639864 VEGF | τ²=0.030 | p1,摘要Results | 保留正确原值,来源修正为`verified_nonforest` |
81
- | MA-013 / 41566983 MortalityShortTerm | 正文I²=89.72,图中无I² | p4正文;Figure 3在p8 | 原目标继续留空;89.72仅在补充来源信息保存 |
82
-
83
- MA-045/046(41901064)森林图同时给出common-effect与random-effects行。benchmark对应common-effect行,已明确记录;图示τ²为异质性估计,不能说它参与了common-effect权重。其他固定模型图上即使打印τ²,也只按其实际来源解释。
84
-
85
- MA-092–095使用论文的Bayesian方法;不能仅按普通随机效应模型重算并期待同样答案。`meta_reference.json`保存每个单元的模型描述,区分总体效应p与异质性p。图示`p=0.00/0.000`只是原文舍入显示,不是数学上的p恰好为0。
86
-
87
- 研究特征表中的`Sample Size`与森林图的特定结局分析人数不是同一字段。典型情况包括42032055的Carmignani(表60、图20+20)和Tedeschi(表186、图86+103),以及41835126的部分研究。两个值各自忠实于来源;本次不把全研究人数替换进图中的分析臂人数。存在无法凭综述消除的原文差异,记录在`outcome_coverage.csv`和来源说明中。
88
-
89
- ## 核验方式与可复查证据
90
-
91
- 1. 在修改前冻结98张提取表、95个合并结果和README,保存原始文件哈希。
92
- 2. 四个审阅分组打开并渲染原始PDF,按PMID、结局、图号、子组逐行核对全部现有数值;特征表另行打开核对。
93
- 3. 对关键数值冲突、τ/τ²含义、正负号和样本量修改进行第二位审阅者独立复核。源图与正文不一致时,效应及区间优先采用指定森林图,保留正文差异说明。
94
- 4. 将每个数值的来源页、图表、原值和核验值保存为逐单元格台账;模型与缺失值状态另存。
95
- 5. 写回更正后,校验所有4,850个数值、95个单元、99个数据文件哈希及27个原始PDF哈希。校验器还检查结局关联、区间、方向和缺失状态。校验通过表示与这次人工逐图建立的参考台账一致,不是程序自动重做视觉审阅。
96
-
97
- 旧版审计元数据虽写`complete`,但只标记273个单元格已确认,5,475个单元格标为`not_visible`;本报告不以该状态作为正确性证据。
98
-
99
- ## 交付文件
100
-
101
- | 文件 | 用途 |
102
- |---|---|
103
- | `extraction_numeric_cells.csv` | 4,850个数值单元格的原值、来源值及页/图/表定位 |
104
- | `meta_reference.json` | 95个单元的选定结果、模型、字段来源、缺失状态与说明 |
105
- | `corrections.csv` | 102条写回更正,含原值与新值 |
106
- | `outcome_coverage.csv` | 98张表的核验覆盖、非森林图字段和原文差异 |
107
- | `source_manifest.csv` | 27篇来源PMID、公开链接、PDF哈希与阶段覆盖 |
108
- | `verified_file_hashes.csv` | 99个最终数据文件的哈希 |
109
- | `summary.json` | 数量与状态汇总 |
110
- | `validate.py` | 只读核验器,检测相对已审阅版本的数据漂移 |
111
-
112
- PDF页码指文件从1开始的物理页码,可能不同于期刊印刷页码。公开台账不包含原文PDF或森林图复制件。原始研究真实性、所有文献标题的逐字一致性及遗漏结局补建不属于本次数值审计范围。
113
-
114
- 更正后的gold/reference已发生变化,之前针对旧版本计算的元分析准确率和方向一致率应重新评分;本次没有修改论文中的实验结果。
115
-
116
- ## 全部更正明细
117
-
118
- | 文件或单元 | 字段 | 原值 | 新值 | 类型 | PDF页 / 图表 |
119
- |---|---|---|---|---|---|
120
- | 39952828.csv:数据行11 | CG (N) | 10 | 12 | 数值差异 | p4 / Table 1 |
121
- | MA-055 | pooled_effect | -2.76 | -2.29 | 数值差异 | p10 / Figure 3 |
122
- | MA-055 | ci_lower | -3.90 | -3.23 | 数值差异 | p10 / Figure 3 |
123
- | MA-055 | ci_upper | -1.62 | -1.35 | 数值差异 | p10 / Figure 3 |
124
- | MA-055 | i2 | 空白 | 94 | 补齐已报告值 | p10 / Figure 3 |
125
- | MA-055 | tau2 | 空白 | 2.62 | 补齐已报告值 | p10 / Figure 3 |
126
- | MA-056 | tau2 | 空白 | 0.52 | 补齐已报告值 | p8 / Figure 1 |
127
- | 41065407_PainIntensity.csv:数据行3 | Study name | Bruce-Low 2012 (1) | Bruce-Low 2012 (2 x week) | 标识/方向 | p8 / Figure 1 |
128
- | MA-057 | i2 | 空白 | 74 | 补齐已报告值 | p9 / Figure 2 |
129
- | MA-057 | tau2 | 空白 | 0.45 | 补齐已报告值 | p9 / Figure 2 |
130
- | MA-040 | ci_lower | -3.84 | -3.83 | 数值差异 | p8 / Figure 3A |
131
- | MA-001 | pooled_effect | 0.856 | 0.86 | 对齐图示精度 | p23 / Figure 3 |
132
- | MA-001 | ci_lower | 0.769 | 0.77 | 对齐图示精度 | p23 / Figure 3 |
133
- | MA-001 | ci_upper | 0.943 | 0.94 | 对齐图示精度 | p23 / Figure 3 |
134
- | MA-002 | pooled_effect | 0.724 | 0.72 | 对齐图示精度 | p23 / Figure 4 |
135
- | MA-002 | ci_lower | 0.549 | 0.55 | 对齐图示精度 | p23 / Figure 4 |
136
- | MA-002 | ci_upper | 0.900 | 0.90 | 显示调整 | p23 / Figure 4 |
137
- | MA-002 | i2 | 99.5 | 98.2 | 数值差异 | p23 / Figure 4 |
138
- | MA-085 | tau2 | 空白 | 0.12 | 补齐已报告值 | p11 / Figure 8 |
139
- | MA-087 | tau2 | 空白 | 0.00 | 补齐已报告值 | p10 / Figure 6 |
140
- | MA-088 | tau2 | 空白 | 0.08 | 补齐已报告值 | p12 / Figure 9 |
141
- | MA-089 | tau2 | 空白 | 0.30 | 补齐已报告值 | p11 / Figure 7 |
142
- | MA-090 | tau2 | 空白 | 3.30 | 补齐已报告值 | p8 / Figure 4 |
143
- | MA-078 | tau2 | 空白 | 0.84 | 补齐已报告值 | p9 / Figure 3 |
144
- | MA-079 | i2 | 67 | 67.00 | 显示调整 | p9 / Figure 4 |
145
- | MA-079 | tau2 | 空白 | 0.05 | 补齐已报告值 | p9 / Figure 4 |
146
- | MA-077 | i2 | 81 | 90.13 | 数值差异 | p10 / Figure 5 |
147
- | MA-077 | tau2 | 空白 | 0.10 | 补齐已报告值 | p10 / Figure 5 |
148
- | MA-008 | ci_lower | 1.00 | 1.01 | 数值差异 | p10 / Figure 4: Suicide |
149
- | 41639651_SBP.csv:数据行3 | 95% CI Upper | 0.00 | -0.00 | 显示调整 | p6 / Figure 2A |
150
- | MA-091 | pooled_effect | -0.075 | -0.08 | 对齐图示精度 | p6 / Figure 2 |
151
- | MA-091 | ci_lower | -0.246 | -0.25 | 对齐图示精度 | p6 / Figure 2 |
152
- | MA-091 | ci_upper | 0.096 | 0.10 | 对齐图示精度 | p6 / Figure 2 |
153
- | MA-092 | pooled_effect | -3.639 | -3.29 | 数值差异 | p7 / Figure 3: CCL2 |
154
- | MA-092 | ci_upper | 15.594 | 15.59 | 对齐图示精度 | p7 / Figure 3: CCL2 |
155
- | MA-092 | tau2 | 3.207 | 空白 | τ/τ²含义纠正 | p5 / Results text below Table 1 |
156
- | MA-093 | pooled_effect | -0.036 | -0.04 | 对齐图示精度 | p7 / Figure 3: IL-6 |
157
- | MA-093 | ci_lower | -0.968 | -0.97 | 对齐图示精度 | p7 / Figure 3: IL-6 |
158
- | MA-093 | ci_upper | 0.908 | 0.91 | 对齐图示精度 | p7 / Figure 3: IL-6 |
159
- | MA-093 | tau2 | 0.142 | 空白 | τ/τ²含义纠正 | p5 / Results text below Table 1 |
160
- | MA-094 | pooled_effect | -0.284 | -0.25 | 数值差异 | p7 / Figure 3: IL-8 |
161
- | MA-094 | ci_lower | -3.543 | -3.54 | 对齐图示精度 | p7 / Figure 3: IL-8 |
162
- | MA-094 | ci_upper | 2.877 | 2.88 | 对齐图示精度 | p7 / Figure 3: IL-8 |
163
- | MA-094 | tau2 | 0.507 | 空白 | τ/τ²含义纠正 | p5 / Results text below Table 1 |
164
- | MA-095 | ci_lower | -1.437 | -1.44 | 对齐图示精度 | p7 / Figure 3: TNF-alpha |
165
- | MA-095 | ci_upper | 1.377 | 1.38 | 对齐图示精度 | p7 / Figure 3: TNF-alpha |
166
- | MA-095 | tau2 | 0.219 | 空白 | τ/τ²含义纠正 | p5 / Results text below Table 1 |
167
- | MA-081 | tau2 | 空白 | 0.47 | 补齐已报告值 | p11 / Figure 3b |
168
- | MA-082 | tau2 | 空白 | 0.06 | 补齐已报告值 | p11 / Figure 3c |
169
- | MA-084 | tau2 | 空白 | 0.60 | 补齐已报告值 | p11 / Figure 3d |
170
- | MA-083 | tau2 | 空白 | 0.04 | 补齐已报告值 | p11 / Figure 3e |
171
- | MA-060 | tau2 | 空白 | 3.77 | 补齐已报告值 | p9 / Figure 2c |
172
- | MA-074 | tau2 | 空白 | 0.04 | 补齐已报告值 | p10 / Figure 3d |
173
- | MA-070 | tau2 | 空白 | 10.10 | 补齐已报告值 | p10 / Figure 3e |
174
- | MA-071 | tau2 | 空白 | 0.12 | 补齐已报告值 | p10 / Figure 3g |
175
- | MA-076 | i2 | 100 | 0 | 数值差异 | p11 / Figure 4e |
176
- | MA-059 | pooled_effect | -0.06 | 0.06 | 数值差异 | p11 / Figure 4f |
177
- | MA-059 | direction | negative | positive | 标识/方向 | p11 / Figure 4f |
178
- | MA-072 | pooled_effect | -0.04 | 0.04 | 数值差异 | p11 / Figure 4g |
179
- | MA-072 | direction | negative | positive | 标识/方向 | p11 / Figure 4g |
180
- | MA-029 | ci_lower | -3.15 | -3.16 | 数值差异 | p7 / Figure 3B |
181
- | MA-044 | ci_upper | 9.32 | 9.33 | 数值差异 | p35 / Figure 2 |
182
- | MA-042 | ci_lower | -6.31 | -6.32 | 数值差异 | p36 / Figure 5 |
183
- | MA-043 | pooled_effect | -6.71 | -6.72 | 数值差异 | p37 / Figure 3 |
184
- | MA-043 | ci_lower | -10.38 | -10.39 | 数值差异 | p37 / Figure 3 |
185
- | MA-043 | ci_upper | -3.04 | -3.05 | 数值差异 | p37 / Figure 3 |
186
- | MA-041 | pooled_effect | -1.16 | -1.17 | 数值差异 | p38 / Figure 4 |
187
- | MA-041 | ci_lower | -1.92 | -1.93 | 数值差异 | p38 / Figure 4 |
188
- | MA-041 | ci_upper | -0.40 | -0.41 | 数值差异 | p38 / Figure 4 |
189
- | MA-034 | i2 | 89.0 | 98.5 | 数值差异 | p19 / Figure 6 |
190
- | MA-028 | i2 | 空白 | 0.0 | 补齐已报告值 | p7 / Figure 3E |
191
- | MA-029 | i2 | 空白 | 76.4 | 补齐已报告值 | p7 / Figure 3B |
192
- | MA-030 | i2 | 空白 | 25.1 | 补齐已报告值 | p7 / Figure 3A |
193
- | MA-031 | i2 | 空白 | 0.0 | 补齐已报告值 | p7 / Figure 3D |
194
- | MA-032 | i2 | 空白 | 97.6 | 补齐已报告值 | p7 / Figure 3C |
195
- | MA-022 | i2 | 空白 | 5.9 | 补齐已报告值 | p7 / Figure 2A |
196
- | MA-022 | tau2 | 空白 | 0 | 补齐已报告值 | p7 / Figure 2A |
197
- | MA-026 | i2 | 空白 | 41.0 | 补齐已报告值 | p7 / Figure 2B |
198
- | MA-026 | tau2 | 空白 | 0.0082 | 补齐已报告值 | p7 / Figure 2B |
199
- | MA-025 | i2 | 空白 | 0.0 | 补齐已报告值 | p8 / Figure 2C |
200
- | MA-025 | tau2 | 空白 | 0 | 补齐已报告值 | p8 / Figure 2C |
201
- | MA-023 | i2 | 空白 | 17.0 | 补齐已报告值 | p8 / Figure 2D |
202
- | MA-023 | tau2 | 空白 | 0.0053 | 补齐已报告值 | p8 / Figure 2D |
203
- | MA-024 | i2 | 空白 | 28.9 | 补齐已报告值 | p9 / Figure 2E |
204
- | MA-024 | tau2 | 空白 | 0.0030 | 补齐已报告值 | p9 / Figure 2E |
205
- | MA-014 | tau2 | 空白 | 1.2602 | 补齐已报告值 | p7 / Figure 2 |
206
- | MA-044 | i2 | 空白 | 68.0 | 补齐已报告值 | p35 / Figure 2 |
207
- | MA-041 | i2 | 空白 | 0.0 | 补齐已报告值 | p38 / Figure 4 |
208
- | MA-045 | tau2 | 空白 | 0 | 补齐已报告值 | p8 / Figure 3B |
209
- | MA-046 | tau2 | 空白 | 2.5933 | 补齐已报告值 | p8 / Figure 3A |
210
- | MA-033 | i2 | 空白 | 92.1 | 补齐已报告值 | p17 / Figure 4A |
211
- | MA-035 | i2 | 空白 | 80.3 | 补齐已报告值 | p17 / Figure 4B |
212
- | MA-020 | tau2 | 空白 | 0.10 | 补齐已报告值 | p9 / Figure 2 |
213
- | MA-017 | tau2 | 空白 | 5.81 | 补齐已报告值 | p9 / Figure 3, subgroup 1.7.1 |
214
- | MA-019 | tau2 | 空白 | 148.61 | 补齐已报告值 | p9 / Figure 3, subgroup 1.7.2 |
215
- | MA-018 | tau2 | 空白 | 124.69 | 补齐已报告值 | p9 / Figure 3, subgroup 1.7.3 |
216
- | MA-016 | tau2 | 空白 | 1.17 | 补齐已报告值 | p9 / Figure 3, subgroup 1.7.4 |
217
- | MA-048 | ci_upper | 0.96 | 0.69 | 数值差异 | p6 / Figure 2 |
218
- | MA-048 | i2 | 54 | 54.0 | 显示调整 | p6 / Figure 2 |
219
- | MA-052 | i2 | 空白 | 0.0 | 补齐已报告值 | p8 / Figure 2 |
220
- | MA-053 | i2 | 空白 | 0.0 | 补齐已报告值 | p9 / Figure 3 |
221
- | MA-014 | direction | negative | not_applicable | 标识/方向 | p7 / Figure 2 |
222
-
223
- ## Hugging Face 发布准备中的说明纠正
224
-
225
- 再次检查来源标注时发现两处元数据错误:MA-091的τ²=0.030仅在摘要报告,原汇总台账误标森林图来源;MA-083的模型应为IV, Random,原辅助说明误写IV, Fixed。已对照原始PDF纠正来源页、模型和重复说明。原98张提取CSV、95个合并结果及其数值未改变。详情见`release_metadata_corrections.json`。
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/corrections.csv DELETED
@@ -1,103 +0,0 @@
1
- file,row_or_unit,review_pmid,column,old,new,change_class,pdf_page,figure_or_table,reason
2
- data_extraction/results/39952828.csv,11,39952828,CG (N),10,12,numeric_value_change,4,Table 1,Original Table 1 reports control N=12 for Tsutsumi/reference 50; target incorrectly contains 10. Forest plot does not report N.
3
- meta_analysis/units.csv,MA-055,41065407,pooled_effect,-2.76,-2.29,numeric_value_change,10,Figure 3,Target differs from the overall forest-plot diamond and corresponding reported result.
4
- meta_analysis/units.csv,MA-055,41065407,ci_lower,-3.90,-3.23,numeric_value_change,10,Figure 3,Target differs from the overall forest-plot diamond and corresponding reported result.
5
- meta_analysis/units.csv,MA-055,41065407,ci_upper,-1.62,-1.35,numeric_value_change,10,Figure 3,Target differs from the overall forest-plot diamond and corresponding reported result.
6
- meta_analysis/units.csv,MA-055,41065407,i2,,94,fill_reported_blank,10,Figure 3,Published value is present in original source but target is blank.
7
- meta_analysis/units.csv,MA-055,41065407,tau2,,2.62,fill_reported_blank,10,Figure 3,Published value is present in original source but target is blank.
8
- meta_analysis/units.csv,MA-056,41065407,tau2,,0.52,fill_reported_blank,8,Figure 1,Published value is present in original source but target is blank.
9
- data_extraction/results/41065407_PainIntensity.csv,3,41065407,Study name,Bruce-Low 2012 (1),Bruce-Low 2012 (2 x week),identifier_or_direction,8,Figure 1,"Forest row carrying -1.59 [-2.24,-0.95] is labeled 2 x week; target (1) is ambiguous and does not reproduce the arm identifier."
10
- meta_analysis/units.csv,MA-057,41065407,i2,,74,fill_reported_blank,9,Figure 2,Published value is present in original source but target is blank.
11
- meta_analysis/units.csv,MA-057,41065407,tau2,,0.45,fill_reported_blank,9,Figure 2,Published value is present in original source but target is blank.
12
- meta_analysis/units.csv,MA-040,41388744,ci_lower,-3.84,-3.83,numeric_value_change,8,Figure 3A,Forest plot and narrative disagree; select forest value under the requested forest-source contract and preserve discrepancy in notes.
13
- meta_analysis/units.csv,MA-001,41507873,pooled_effect,0.856,0.86,align_forest_rounding,23,Figure 3,Align to exact displayed forest precision; existing prose value is compatible after rounding.
14
- meta_analysis/units.csv,MA-001,41507873,ci_lower,0.769,0.77,align_forest_rounding,23,Figure 3,Align to exact displayed forest precision; existing prose value is compatible after rounding.
15
- meta_analysis/units.csv,MA-001,41507873,ci_upper,0.943,0.94,align_forest_rounding,23,Figure 3,Align to exact displayed forest precision; existing prose value is compatible after rounding.
16
- meta_analysis/units.csv,MA-002,41507873,pooled_effect,0.724,0.72,align_forest_rounding,23,Figure 4,Align to exact displayed forest precision; existing prose value is compatible after rounding.
17
- meta_analysis/units.csv,MA-002,41507873,ci_lower,0.549,0.55,align_forest_rounding,23,Figure 4,Align to exact displayed forest precision; existing prose value is compatible after rounding.
18
- meta_analysis/units.csv,MA-002,41507873,ci_upper,0.900,0.90,display_only,23,Figure 4,Align to exact displayed forest precision; existing prose value is compatible after rounding.
19
- meta_analysis/units.csv,MA-002,41507873,i2,99.5,98.2,numeric_value_change,23,Figure 4,Forest plot and narrative disagree; select forest value under the requested forest-source contract and preserve discrepancy in notes.
20
- meta_analysis/units.csv,MA-085,41561801,tau2,,0.12,fill_reported_blank,11,Figure 8,Published value is present in original source but target is blank.
21
- meta_analysis/units.csv,MA-087,41561801,tau2,,0.00,fill_reported_blank,10,Figure 6,Published value is present in original source but target is blank.
22
- meta_analysis/units.csv,MA-088,41561801,tau2,,0.08,fill_reported_blank,12,Figure 9,Published value is present in original source but target is blank.
23
- meta_analysis/units.csv,MA-089,41561801,tau2,,0.30,fill_reported_blank,11,Figure 7,Published value is present in original source but target is blank.
24
- meta_analysis/units.csv,MA-090,41561801,tau2,,3.30,fill_reported_blank,8,Figure 4,Published value is present in original source but target is blank.
25
- meta_analysis/units.csv,MA-078,41601519,tau2,,0.84,fill_reported_blank,9,Figure 3,Original Figure 3 explicitly displays tau2 = 0.84.
26
- meta_analysis/units.csv,MA-079,41601519,i2,67,67.00,display_only,9,Figure 4,Original Figure 4 explicitly displays i2 = 67.00. Numerical value is unchanged; retain source display precision.
27
- meta_analysis/units.csv,MA-079,41601519,tau2,,0.05,fill_reported_blank,9,Figure 4,Original Figure 4 explicitly displays tau2 = 0.05.
28
- meta_analysis/units.csv,MA-077,41601519,i2,81,90.13,numeric_value_change,10,Figure 5,Original Figure 5 explicitly displays i2 = 90.13.
29
- meta_analysis/units.csv,MA-077,41601519,tau2,,0.10,fill_reported_blank,10,Figure 5,Original Figure 5 explicitly displays tau2 = 0.10.
30
- meta_analysis/units.csv,MA-008,41617837,ci_lower,1.00,1.01,numeric_value_change,10,Figure 4: Suicide,Original Figure 4: Suicide explicitly displays ci_lower = 1.01.
31
- data_extraction/results/41639651_SBP.csv,3,41639651,95% CI Upper,0.00,-0.00,display_only,6,Figure 2A,Forest row Ren et al 2019 displays -0.00; current 0.00 is numerically equivalent and also occurs in Table 1. Strict forest transcription retains the minus sign.
32
- meta_analysis/units.csv,MA-091,41639864,pooled_effect,-0.075,-0.08,align_forest_rounding,6,Figure 2,"Original Figure 2 explicitly displays pooled_effect = -0.08. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
33
- meta_analysis/units.csv,MA-091,41639864,ci_lower,-0.246,-0.25,align_forest_rounding,6,Figure 2,"Original Figure 2 explicitly displays ci_lower = -0.25. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
34
- meta_analysis/units.csv,MA-091,41639864,ci_upper,0.096,0.10,align_forest_rounding,6,Figure 2,"Original Figure 2 explicitly displays ci_upper = 0.10. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
35
- meta_analysis/units.csv,MA-092,41652340,pooled_effect,-3.639,-3.29,numeric_value_change,7,Figure 3: CCL2,Figure 3 mean-effect row displays -3.29; Table 2 and p5 Results display -3.639. Difference is not rounding. Posterior summary distinction is not explicitly resolved in the main article. Correction implements forest-source convention and must retain this conflict note.
36
- meta_analysis/units.csv,MA-092,41652340,ci_upper,15.594,15.59,align_forest_rounding,7,Figure 3: CCL2,"Original Figure 3: CCL2 explicitly displays ci_upper = 15.59. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
37
- meta_analysis/units.csv,MA-092,41652340,tau2,3.207,,remove_mislabeled_tau,5,Results text below Table 1,"Published 3.207 is labeled tau (heterogeneity standard deviation), not tau-squared. The target tau2 is semantically misfiled; no published tau2 was found."
38
- meta_analysis/units.csv,MA-093,41652340,pooled_effect,-0.036,-0.04,align_forest_rounding,7,Figure 3: IL-6,"Original Figure 3: IL-6 explicitly displays pooled_effect = -0.04. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
39
- meta_analysis/units.csv,MA-093,41652340,ci_lower,-0.968,-0.97,align_forest_rounding,7,Figure 3: IL-6,"Original Figure 3: IL-6 explicitly displays ci_lower = -0.97. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
40
- meta_analysis/units.csv,MA-093,41652340,ci_upper,0.908,0.91,align_forest_rounding,7,Figure 3: IL-6,"Original Figure 3: IL-6 explicitly displays ci_upper = 0.91. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
41
- meta_analysis/units.csv,MA-093,41652340,tau2,0.142,,remove_mislabeled_tau,5,Results text below Table 1,"Published 0.142 is labeled tau (heterogeneity standard deviation), not tau-squared. The target tau2 is semantically misfiled; no published tau2 was found."
42
- meta_analysis/units.csv,MA-094,41652340,pooled_effect,-0.284,-0.25,numeric_value_change,7,Figure 3: IL-8,Figure 3 mean-effect row displays -0.25; Table 2 and p5 Results display -0.284. Difference is not rounding. Posterior summary distinction is not explicitly resolved in the main article. Correction implements forest-source convention and must retain this conflict note.
43
- meta_analysis/units.csv,MA-094,41652340,ci_lower,-3.543,-3.54,align_forest_rounding,7,Figure 3: IL-8,"Original Figure 3: IL-8 explicitly displays ci_lower = -3.54. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
44
- meta_analysis/units.csv,MA-094,41652340,ci_upper,2.877,2.88,align_forest_rounding,7,Figure 3: IL-8,"Original Figure 3: IL-8 explicitly displays ci_upper = 2.88. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
45
- meta_analysis/units.csv,MA-094,41652340,tau2,0.507,,remove_mislabeled_tau,5,Results text below Table 1,"Published 0.507 is labeled tau (heterogeneity standard deviation), not tau-squared. The target tau2 is semantically misfiled; no published tau2 was found."
46
- meta_analysis/units.csv,MA-095,41652340,ci_lower,-1.437,-1.44,align_forest_rounding,7,Figure 3: TNF-alpha,"Original Figure 3: TNF-alpha explicitly displays ci_lower = -1.44. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
47
- meta_analysis/units.csv,MA-095,41652340,ci_upper,1.377,1.38,align_forest_rounding,7,Figure 3: TNF-alpha,"Original Figure 3: TNF-alpha explicitly displays ci_upper = 1.38. Existing value is the more precise Table/abstract value; this change follows the requested forest display precision, not a substantive discrepancy."
48
- meta_analysis/units.csv,MA-095,41652340,tau2,0.219,,remove_mislabeled_tau,5,Results text below Table 1,"Published 0.219 is labeled tau (heterogeneity standard deviation), not tau-squared. The target tau2 is semantically misfiled; no published tau2 was found."
49
- meta_analysis/units.csv,MA-081,41656984,tau2,,0.47,fill_reported_blank,11,Figure 3b,Original Figure 3b explicitly displays tau2 = 0.47.
50
- meta_analysis/units.csv,MA-082,41656984,tau2,,0.06,fill_reported_blank,11,Figure 3c,Original Figure 3c explicitly displays tau2 = 0.06.
51
- meta_analysis/units.csv,MA-084,41656984,tau2,,0.60,fill_reported_blank,11,Figure 3d,Original Figure 3d explicitly displays tau2 = 0.60.
52
- meta_analysis/units.csv,MA-083,41656984,tau2,,0.04,fill_reported_blank,11,Figure 3e,Original Figure 3e explicitly displays tau2 = 0.04.
53
- meta_analysis/units.csv,MA-060,41743679,tau2,,3.77,fill_reported_blank,9,Figure 2c,Original Figure 2c explicitly displays tau2 = 3.77.
54
- meta_analysis/units.csv,MA-074,41743679,tau2,,0.04,fill_reported_blank,10,Figure 3d,Original Figure 3d explicitly displays tau2 = 0.04.
55
- meta_analysis/units.csv,MA-070,41743679,tau2,,10.10,fill_reported_blank,10,Figure 3e,Original Figure 3e explicitly displays tau2 = 10.10.
56
- meta_analysis/units.csv,MA-071,41743679,tau2,,0.12,fill_reported_blank,10,Figure 3g,Original Figure 3g explicitly displays tau2 = 0.12.
57
- meta_analysis/units.csv,MA-076,41743679,i2,100,0,numeric_value_change,11,Figure 4e,Original Figure 4e explicitly displays i2 = 0.
58
- meta_analysis/units.csv,MA-059,41743679,pooled_effect,-0.06,0.06,numeric_value_change,11,Figure 4f,Original Figure 4f explicitly displays pooled_effect = 0.06.
59
- meta_analysis/units.csv,MA-059,41743679,direction,negative,positive,identifier_or_direction,11,Figure 4f,Direction follows the sign printed for the pooled effect.
60
- meta_analysis/units.csv,MA-072,41743679,pooled_effect,-0.04,0.04,numeric_value_change,11,Figure 4g,Original Figure 4g explicitly displays pooled_effect = 0.04.
61
- meta_analysis/units.csv,MA-072,41743679,direction,negative,positive,identifier_or_direction,11,Figure 4g,Direction follows the sign printed for the pooled effect.
62
- meta_analysis/units.csv,MA-029,41761990,ci_lower,-3.15,-3.16,numeric_value_change,7,Figure 3B,Original forest plot numeric value.
63
- meta_analysis/units.csv,MA-044,41888952,ci_upper,9.32,9.33,numeric_value_change,35,Figure 2,Original forest plot numeric value.
64
- meta_analysis/units.csv,MA-042,41888952,ci_lower,-6.31,-6.32,numeric_value_change,36,Figure 5,Original forest plot numeric value.
65
- meta_analysis/units.csv,MA-043,41888952,pooled_effect,-6.71,-6.72,numeric_value_change,37,Figure 3,Original forest plot numeric value.
66
- meta_analysis/units.csv,MA-043,41888952,ci_lower,-10.38,-10.39,numeric_value_change,37,Figure 3,Original forest plot numeric value.
67
- meta_analysis/units.csv,MA-043,41888952,ci_upper,-3.04,-3.05,numeric_value_change,37,Figure 3,Original forest plot numeric value.
68
- meta_analysis/units.csv,MA-041,41888952,pooled_effect,-1.16,-1.17,numeric_value_change,38,Figure 4,Original forest plot numeric value.
69
- meta_analysis/units.csv,MA-041,41888952,ci_lower,-1.92,-1.93,numeric_value_change,38,Figure 4,Original forest plot numeric value.
70
- meta_analysis/units.csv,MA-041,41888952,ci_upper,-0.40,-0.41,numeric_value_change,38,Figure 4,Original forest plot numeric value.
71
- meta_analysis/units.csv,MA-034,41938012,i2,89.0,98.5,numeric_value_change,19,Figure 6,Original forest plot numeric value.
72
- meta_analysis/units.csv,MA-028,41761990,i2,,0.0,fill_reported_blank,7,Figure 3E,Published in original forest plot; current field is empty.
73
- meta_analysis/units.csv,MA-029,41761990,i2,,76.4,fill_reported_blank,7,Figure 3B,Published in original forest plot; current field is empty.
74
- meta_analysis/units.csv,MA-030,41761990,i2,,25.1,fill_reported_blank,7,Figure 3A,Published in original forest plot; current field is empty.
75
- meta_analysis/units.csv,MA-031,41761990,i2,,0.0,fill_reported_blank,7,Figure 3D,Published in original forest plot; current field is empty.
76
- meta_analysis/units.csv,MA-032,41761990,i2,,97.6,fill_reported_blank,7,Figure 3C,Published in original forest plot; current field is empty.
77
- meta_analysis/units.csv,MA-022,41830012,i2,,5.9,fill_reported_blank,7,Figure 2A,Published in original forest plot; current field is empty.
78
- meta_analysis/units.csv,MA-022,41830012,tau2,,0,fill_reported_blank,7,Figure 2A,Published in original forest plot; current field is empty.
79
- meta_analysis/units.csv,MA-026,41830012,i2,,41.0,fill_reported_blank,7,Figure 2B,Published in original forest plot; current field is empty.
80
- meta_analysis/units.csv,MA-026,41830012,tau2,,0.0082,fill_reported_blank,7,Figure 2B,Published in original forest plot; current field is empty.
81
- meta_analysis/units.csv,MA-025,41830012,i2,,0.0,fill_reported_blank,8,Figure 2C,Published in original forest plot; current field is empty.
82
- meta_analysis/units.csv,MA-025,41830012,tau2,,0,fill_reported_blank,8,Figure 2C,Published in original forest plot; current field is empty.
83
- meta_analysis/units.csv,MA-023,41830012,i2,,17.0,fill_reported_blank,8,Figure 2D,Published in original forest plot; current field is empty.
84
- meta_analysis/units.csv,MA-023,41830012,tau2,,0.0053,fill_reported_blank,8,Figure 2D,Published in original forest plot; current field is empty.
85
- meta_analysis/units.csv,MA-024,41830012,i2,,28.9,fill_reported_blank,9,Figure 2E,Published in original forest plot; current field is empty.
86
- meta_analysis/units.csv,MA-024,41830012,tau2,,0.0030,fill_reported_blank,9,Figure 2E,Published in original forest plot; current field is empty.
87
- meta_analysis/units.csv,MA-014,41835126,tau2,,1.2602,fill_reported_blank,7,Figure 2,Published in original forest plot; current field is empty.
88
- meta_analysis/units.csv,MA-044,41888952,i2,,68.0,fill_reported_blank,35,Figure 2,Published in original forest plot; current field is empty.
89
- meta_analysis/units.csv,MA-041,41888952,i2,,0.0,fill_reported_blank,38,Figure 4,Published in original forest plot; current field is empty.
90
- meta_analysis/units.csv,MA-045,41901064,tau2,,0,fill_reported_blank,8,Figure 3B,Published in original forest plot; current field is empty.
91
- meta_analysis/units.csv,MA-046,41901064,tau2,,2.5933,fill_reported_blank,8,Figure 3A,Published in original forest plot; current field is empty.
92
- meta_analysis/units.csv,MA-033,41938012,i2,,92.1,fill_reported_blank,17,Figure 4A,Published in original forest plot; current field is empty.
93
- meta_analysis/units.csv,MA-035,41938012,i2,,80.3,fill_reported_blank,17,Figure 4B,Published in original forest plot; current field is empty.
94
- meta_analysis/units.csv,MA-020,41970007,tau2,,0.10,fill_reported_blank,9,Figure 2,Value displayed in selected original forest plot. missing_reported_value
95
- meta_analysis/units.csv,MA-017,41970007,tau2,,5.81,fill_reported_blank,9,"Figure 3, subgroup 1.7.1",Value displayed in selected original forest plot. missing_reported_value
96
- meta_analysis/units.csv,MA-019,41970007,tau2,,148.61,fill_reported_blank,9,"Figure 3, subgroup 1.7.2",Value displayed in selected original forest plot. missing_reported_value
97
- meta_analysis/units.csv,MA-018,41970007,tau2,,124.69,fill_reported_blank,9,"Figure 3, subgroup 1.7.3",Value displayed in selected original forest plot. missing_reported_value
98
- meta_analysis/units.csv,MA-016,41970007,tau2,,1.17,fill_reported_blank,9,"Figure 3, subgroup 1.7.4",Value displayed in selected original forest plot. missing_reported_value
99
- meta_analysis/units.csv,MA-048,42003421,ci_upper,0.96,0.69,numeric_value_change,6,Figure 2,Value displayed in selected original forest plot. source_conflict
100
- meta_analysis/units.csv,MA-048,42003421,i2,54,54.0,display_only,6,Figure 2,Value displayed in selected original forest plot. precision_alignment
101
- meta_analysis/units.csv,MA-052,42015424,i2,,0.0,fill_reported_blank,8,Figure 2,Value displayed in selected original forest plot. missing_reported_value
102
- meta_analysis/units.csv,MA-053,42015424,i2,,0.0,fill_reported_blank,9,Figure 3,Value displayed in selected original forest plot. missing_reported_value
103
- meta_analysis/units.csv,MA-014,41835126,direction,negative,not_applicable,identifier_or_direction,7,Figure 2,Single-group prevalence Events/Total has no treatment/control comparator; treatment-effect direction is not applicable. Independently confirmed in final semantic validation.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/extraction_numeric_cells.csv DELETED
The diff for this file is too large to render. See raw diff
 
validation/hf_release_verification.json DELETED
@@ -1,47 +0,0 @@
1
- {
2
- "stage": "pre_upload",
3
- "source_github_revision": "8dfb61c0a69c7d791731b0de55c24ef8c5ef1ba6",
4
- "raw_data_files_byte_identical": 155,
5
- "raw_numeric_audit": {
6
- "status": "pass",
7
- "extraction_files": 98,
8
- "extraction_rows": 747,
9
- "numeric_cells_checked": 4850,
10
- "meta_units": 95,
11
- "source_pdfs_hash_checked": 27,
12
- "errors": []
13
- },
14
- "datasets_load_checks": {
15
- "reviews": {
16
- "rows": 27,
17
- "status": "pass"
18
- },
19
- "included_studies": {
20
- "rows": 421,
21
- "status": "pass"
22
- },
23
- "screening_candidates": {
24
- "rows": 81000,
25
- "status": "pass"
26
- },
27
- "screening_truth": {
28
- "rows": 27,
29
- "status": "pass"
30
- },
31
- "data_extraction": {
32
- "rows": 5748,
33
- "status": "pass"
34
- },
35
- "meta_analysis": {
36
- "rows": 95,
37
- "status": "pass"
38
- }
39
- },
40
- "card_metadata_validation": "pass",
41
- "metadata_findings_corrected": 2,
42
- "library_versions": {
43
- "datasets": "5.0.1",
44
- "huggingface_hub": "1.33.0",
45
- "pyarrow": "25.0.1"
46
- }
47
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/meta_reference.json DELETED
The diff for this file is too large to render. See raw diff
 
validation/outcome_coverage.csv DELETED
@@ -1,99 +0,0 @@
1
- review_pmid,result_file,rows,numeric_cells,forest_numeric_cells,table_numeric_cells,derived_numeric_cells,other_source_numeric_cells,unchecked_numeric_cells,pdf_pages,figure,notes
2
- 39952828,39952828.csv,13,78,51,25,1,1,0,4;7;9;10,Figure 3a,"All 13 study SMD/CI triplets match Figure 3a; figure omits Mortazavi year, confirmed as 2012 in reference 36 on PDF p9. Table 1 p4 is the source for all CG/IG sample sizes. Tsutsumi (reference 50) control group is 12, not CSV 10. IG 24 is the combined size of two explicitly reported 12-person exercise arms; it is not printed as a single 24 cell. Titles are bibliographic labels outside the numeric audit; not certified as verbatim. Study identities/reference mapping checked using bibliography pp9-10. Cunha is 2022 in Figure 3a and reference 14; Figure 3b labels the same study 2021. Current 2022 agrees with the selected overall analysis."
3
- 40325863,40325863.csv,8,48,40,8,0,0,0,5;6;7;8;10,Figure 2,"For all eight studies, the five plotted numeric fields (effect, lower CI, upper CI, Z, p) match exactly after comma decimal separator normalization. Source p=0.00 is rounded published text, not a mathematical zero. Sample Size (N) verified in Table 1 pp5-8; these are numbers enrolled in each study, not necessarily the analyzed two-arm N for the plotted effect. Figure labels Ozdogar 2022 and Siengsukon 2021; Table 1 labels these studies 2023 and 2020b. Retain forest labels under forest-source precedence; record source inconsistency. I2=91.002%, tau2=1.731, Q=77.796, heterogeneity p<0.001, and random-effects model reported in same-page results text, not inside Figure 2. Figure pooled p=0.00; prose says p<0.001."
4
- 41065407,41065407_Disability.csv,13,52,52,0,0,0,0,10,Figure 3,"All 13 study triplets match Figure 3. The overall diamond is -2.29 [-3.23,-1.35], consistently repeated in p9 prose; current MA-055 values -2.76 [-3.90,-1.62] are incorrect."
5
- 41065407,41065407_PainIntensity.csv,10,40,40,0,0,0,0,8,Figure 1,All ten effect/CI triplets and years match Figure 1. CSV row 3 study arm is mislabeled (1); source says Bruce-Low 2012 (2 x week). Correct the label to distinguish it from row 4 (1 x week). Plotted effect/CI columns are rounded to two decimals; more precise SMD and SE columns also exist in the figure but are not substituted.
6
- 41065407,41065407_QoL.csv,6,24,24,0,0,0,0,9,Figure 2,All six study triplets match Figure 2; I2=74% and tau2=0.45 are printed under the overall diamond and were missing from target.
7
- 41388744,41388744_DBP.csv,16,64,64,0,0,0,0,6;8,Figure 3B,"All 16 WMD/CI triplets and years match Figure 3B. Figure caption explicitly calls this weighted mean difference (WMD), while extraction column uses generic Mean Difference (mmHg). Figure heterogeneity p is printed as p<0.000; preserve verbatim in metadata and flag it as a source formatting anomaly. Same-page model note says random effects. Overall-effect p=0.15 is from p6 text, not the forest heterogeneity test."
8
- 41388744,41388744_SBP.csv,16,144,64,80,0,0,0,5;6;8,Figure 3A,All 16 study WMD/CI triplets and years match Figure 3A; all 80 extra numeric characteristics match Table 1 pp5-6. Source conflict: Figure 3A pooled lower CI is -3.83; abstract and results text pp6-7 and Table 2 p9 say -3.84. Figure value selected because this is a forest-source benchmark. Overall-effect p<0.001 is from p6 prose; heterogeneity p=0.123 is from the plot.
9
- 41507873,41507873_Alzheimer.csv,3,12,12,0,0,0,0,23,Figure 3,"All three study OR/CI triplets and years match Figure 3. Plot pooled 0.86 [0.77,0.94] is compatible with prose/current target 0.856 [0.769,0.943]; changes are precision alignment to the plot, not substantive numerical errors."
10
- 41507873,41507873_Depression.csv,3,15,12,3,0,0,0,17;18;23,Figure 4,"All three study OR/CI triplets and years match Figure 4. N values 249405,249405,2424 match Table 2 pp17-18. Plot pooled 0.72 [0.55,0.90] is compatible with prose/current 0.724 [0.549,0.900]; precision alignment only. Substantive source conflict: Figure 4 overall I2 is 98.2%; same-page prose says 99.5%. Select 98.2 under forest-source precedence and retain both as a documented source discrepancy."
11
- 41561801,41561801_Adiponectin.csv,4,16,16,0,0,0,0,11,Figure 8,
12
- 41561801,41561801_BMI.csv,11,44,44,0,0,0,0,9,Figure 5,All 11 study MD/CI triplets and years match Figure 5. Model is fixed effect; tau2 is not reported and must not be imputed as zero.
13
- 41561801,41561801_CRP.csv,12,48,48,0,0,0,0,10,Figure 6,"All 12 study MD/CI triplets and years match Figure 6, including row 5 upper CI printed -0.00. Overall tau2 is explicitly printed 0.00 and should replace the missing target value."
14
- 41561801,41561801_IL-6.csv,9,36,36,0,0,0,0,12,Figure 9,All nine study MD/CI triplets and years match Figure 9. Forest pooled CI upper endpoint is -0.05; p11 prose omits that minus sign (0.05). The target already correctly follows the forest.
15
- 41561801,41561801_TNF-alpha.csv,11,44,44,0,0,0,0,11,Figure 7,
16
- 41561801,41561801_WC.csv,8,32,32,0,0,0,0,8,Figure 4,
17
- 41566983,41566983_InfectionLongTerm.csv,5,20,20,0,0,0,0,4;5;6;7,Figure 2B,"All five effect/CI triplets and years match Figure 2B. First three Jiang 2020 rows map in order to Wuhan, XiaoGan, Huanggang; these location disambiguators are not represented in the current CSV schema. Figure labels Wu 2020, while Table 1 p5 says Wu 2021. Retain figure year. Do not reinterpret unusually large or small estimates: these are the published forest values. Plot uses generic Estimate/Effect size; results p4 identifies the pooled measure as RR and provides I2=99.27%; tau2 not reported."
18
- 41566983,41566983_InfectionShortTerm.csv,16,64,64,0,0,0,0,4;5;6;7,Figure 2A,"All 16 effect/CI triplets and years match Figure 2A. All Location labels verified against plot and Table 1 pp5-6 (including numeric city/state counts). Figure labels Singh 2021, while Table 1 p5 says Singh 2022. Retain figure year. Plot uses generic Estimate/Effect size; results p4 identifies pooled RR and I2=76.49%; tau2 not reported."
19
- 41566983,41566983_MortalityShortTerm.csv,5,20,20,0,0,0,0,4;5;6;8,Figure 3,"All five effect/CI triplets and years match Figure 3. Locations verified against plot and Table 1 p5. Figure labels Singh 2021 while Table 1 p5 says 2022; retain forest year. I2=89.72% is explicitly reported in Results p4, but missing from target; it is not printed inside Figure 3. RR measure also confirmed in Results p4. Tau2 and numerical overall-effect p are not reported."
20
- 41601519,41601519_DepressiveSymptoms.csv,16,160,160,0,0,0,0,10,Figure 5,"All study rows, year labels, and existing numeric fields match the displayed forest plot."
21
- 41601519,41601519_InsomniaSeverity.csv,16,160,160,0,0,0,0,9,Figure 3,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Henry 2020 has positive g 0.78 in the original plot, despite the remaining studies mostly favoring dCBT-I; retained as published. Figures contain the same arm means/SDs as the CSV, so no sign reversal was inferred."
22
- 41601519,41601519_SleepEfficiency.csv,6,60,60,0,0,0,0,9,Figure 4,"All study rows, year labels, and existing numeric fields match the displayed forest plot."
23
- 41617837,41617837_Anxiety.csv,4,12,12,0,0,0,0,10,Figure 4: Anxiety,"All study RR/CI cells and geographical settings match; study years embedded in Study labels also match. Duplicate Niu 2023 and da Silva 2020 rows, when present, were mapped by displayed order and effect values, not deduplicated. Model comes from p9 Discussion; the forest plot does not print a model caption."
24
- 41617837,41617837_CompositeMentalIllnesses.csv,17,51,51,0,0,0,0,10,Figure 4: Composite mental health illnesses,"All study RR/CI cells and geographical settings match; study years embedded in Study labels also match. Duplicate Niu 2023 and da Silva 2020 rows, when present, were mapped by displayed order and effect values, not deduplicated. Model comes from p9 Discussion; the forest plot does not print a model caption."
25
- 41617837,41617837_Depression.csv,5,15,15,0,0,0,0,10,Figure 4: Depression,"All study RR/CI cells and geographical settings match; study years embedded in Study labels also match. Duplicate Niu 2023 and da Silva 2020 rows, when present, were mapped by displayed order and effect values, not deduplicated. Model comes from p9 Discussion; the forest plot does not print a model caption."
26
- 41617837,41617837_MentalHealthDisorders.csv,10,30,30,0,0,0,0,9,Figure 3,"All study RR/CI cells and geographical settings match; study years embedded in Study labels also match. Duplicate Niu 2023 and da Silva 2020 rows, when present, were mapped by displayed order and effect values, not deduplicated. Model comes from p9 Discussion; the forest plot does not print a model caption."
27
- 41617837,41617837_SchizophreniaPsychoticDisorders.csv,6,18,18,0,0,0,0,10,Figure 4: Schizophrenia or psychotic disorders,"All study RR/CI cells and geographical settings match; study years embedded in Study labels also match. Duplicate Niu 2023 and da Silva 2020 rows, when present, were mapped by displayed order and effect values, not deduplicated. Model comes from p9 Discussion; the forest plot does not print a model caption."
28
- 41617837,41617837_Suicide.csv,5,15,15,0,0,0,0,10,Figure 4: Suicide,"All study RR/CI cells and geographical settings match; study years embedded in Study labels also match. Duplicate Niu 2023 and da Silva 2020 rows, when present, were mapped by displayed order and effect values, not deduplicated. Model comes from p9 Discussion; the forest plot does not print a model caption."
29
- 41639651,41639651_DBP.csv,10,40,40,0,0,0,0,6,Figure 2B,All 10 study effect/CI rows and forest year labels match numerically. Tau-squared is not reported in the main figure or main-PDF text. Source table has year discrepancies (Wu 2013 vs forest Wu 2012; Guo 2023 vs forest Guo 2022) and Baccarelli SBP -1.01 vs forest -0.01; the extraction correctly follows the forest.
30
- 41639651,41639651_SBP.csv,10,50,40,10,0,0,0,6;5,Figure 2A,"All 10 study effect/CI rows and forest year labels match numerically. Tau-squared is not reported in the main figure or main-PDF text. Source table has year discrepancies (Wu 2013 vs forest Wu 2012; Guo 2023 vs forest Guo 2022) and Baccarelli SBP -1.01 vs forest -0.01; the extraction correctly follows the forest. All 10 Sample Size values match Table 1 (p5), with comma separators normalized; they are study characteristics, not printed forest data. Ren 2019 upper CI displays -0.00 in the forest and 0.00 in Table 1; signed-zero display correction is recorded separately."
31
- 41639864,41639864_VEGF.csv,14,56,56,0,0,0,0,6;1,Figure 2,"All 14 study rows (including both Duggan 2016 comparisons) match. Plot pooled SMD -0.08 [-0.25, 0.10] is a rounded display of Table 2 and abstract -0.075 [-0.246, 0.096]. Target contains the valid higher-precision table values; listed pooled/CI changes only enforce forest display precision. Tau2 = 0.030 and overall p = 0.390 were visually verified in p1 Abstract, not the plot."
32
- 41652340,41652340_CCL2.csv,2,8,8,0,0,0,0,7;3;5,Figure 3: CCL2,"Both study rows, study years and MD/CI values match original Figure 3. Figure explicitly labels pooled row ""mean effect (mu)"" and a separate prediction row. I2 and tau2 are not reported. Main text p5 reports tau (heterogeneity SD), not tau2. Current tau2 must be cleared and tau retained separately; do not square a posterior summary to invent tau2. Forest pooled point differs from Table 2 beyond rounding. The paper does not establish whether this is a different posterior summary (e.g. mean/median); no such assumption was made. Use forest value only when enforcing the requested forest-source convention, and retain source-conflict annotation."
33
- 41652340,41652340_IL-6.csv,2,8,8,0,0,0,0,7;3;5,Figure 3: IL-6,"Both study rows, study years and MD/CI values match original Figure 3. Figure explicitly labels pooled row ""mean effect (mu)"" and a separate prediction row. I2 and tau2 are not reported. Main text p5 reports tau (heterogeneity SD), not tau2. Current tau2 must be cleared and tau retained separately; do not square a posterior summary to invent tau2."
34
- 41652340,41652340_IL-8.csv,2,8,8,0,0,0,0,7;3;5,Figure 3: IL-8,"Both study rows, study years and MD/CI values match original Figure 3. Figure explicitly labels pooled row ""mean effect (mu)"" and a separate prediction row. I2 and tau2 are not reported. Main text p5 reports tau (heterogeneity SD), not tau2. Current tau2 must be cleared and tau retained separately; do not square a posterior summary to invent tau2. Forest pooled point differs from Table 2 beyond rounding. The paper does not establish whether this is a different posterior summary (e.g. mean/median); no such assumption was made. Use forest value only when enforcing the requested forest-source convention, and retain source-conflict annotation."
35
- 41652340,41652340_TNF-alpha.csv,2,8,8,0,0,0,0,7;3;5,Figure 3: TNF-alpha,"Both study rows, study years and MD/CI values match original Figure 3. Figure explicitly labels pooled row ""mean effect (mu)"" and a separate prediction row. I2 and tau2 are not reported. Main text p5 reports tau (heterogeneity SD), not tau2. Current tau2 must be cleared and tau retained separately; do not square a posterior summary to invent tau2."
36
- 41656984,41656984_Anxiety.csv,6,60,60,0,0,0,0,11,Figure 3a,"All study rows, year labels, and existing numeric fields match the displayed forest plot."
37
- 41656984,41656984_Depression.csv,13,143,130,13,0,0,0,11;5;6;7;8,Figure 3b,"All study rows, year labels, and existing numeric fields match the displayed forest plot. All 13 Sample Size values match original Table 2 pp5–8. These are whole-study enrollment figures and differ legitimately from experimental/control analysis totals; do not substitute their sums. All six arm-summary numeric fields, SMD/CI and Year were separately checked in Figure 3b."
38
- 41656984,41656984_QualityOfLife.csv,12,120,120,0,0,0,0,11,Figure 3c,"All study rows, year labels, and existing numeric fields match the displayed forest plot."
39
- 41656984,41656984_SelfEfficacy.csv,10,100,100,0,0,0,0,11,Figure 3e,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Figure 3e explicitly reports IV, Random; tau2=0.04 and I2=69% are reported in the same panel."
40
- 41656984,41656984_SelfManagementBehavior.csv,5,50,50,0,0,0,0,11,Figure 3d,"All study rows, year labels, and existing numeric fields match the displayed forest plot."
41
- 41743679,41743679_30CST.csv,4,40,40,0,0,0,0,9,Figure 2b,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
42
- 41743679,41743679_ArmCircumference.csv,4,40,40,0,0,0,0,11,Figure 4f,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero. Correct Figure 4 panel mapping is e=Waist-to-hip ratio, f=Arm circumference, g=Thigh circumference; previous evidence map interchanged e and g."
43
- 41743679,41743679_ArmCurlTest.csv,5,50,50,0,0,0,0,9,Figure 2c,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
44
- 41743679,41743679_BMI.csv,6,60,60,0,0,0,0,11,Figure 4b,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
45
- 41743679,41743679_BackScratch.csv,4,40,40,0,0,0,0,10,Figure 3f,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
46
- 41743679,41743679_BodyFatMass.csv,4,40,40,0,0,0,0,9,Figure 2d,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
47
- 41743679,41743679_BodyFatPercentage.csv,6,60,60,0,0,0,0,9,Figure 2e,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
48
- 41743679,41743679_BodyWeight.csv,12,120,120,0,0,0,0,11,Figure 4a,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
49
- 41743679,41743679_HDL-C.csv,4,40,40,0,0,0,0,10,Figure 3b,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
50
- 41743679,41743679_HandGripTest.csv,4,40,40,0,0,0,0,9,Figure 2a,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
51
- 41743679,41743679_HipCircumference.csv,3,30,30,0,0,0,0,11,Figure 4d,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
52
- 41743679,41743679_LDL-C.csv,3,30,30,0,0,0,0,10,Figure 3c,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
53
- 41743679,41743679_SitAndReach.csv,6,60,60,0,0,0,0,10,Figure 3e,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
54
- 41743679,41743679_TUG.csv,3,30,30,0,0,0,0,10,Figure 3g,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
55
- 41743679,41743679_ThighCircumference.csv,4,40,40,0,0,0,0,11,Figure 4g,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero. Correct Figure 4 panel mapping is e=Waist-to-hip ratio, f=Arm circumference, g=Thigh circumference; previous evidence map interchanged e and g."
56
- 41743679,41743679_TotalCholesterol.csv,3,30,30,0,0,0,0,10,Figure 3a,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
57
- 41743679,41743679_Triglycerides.csv,4,40,40,0,0,0,0,10,Figure 3d,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
58
- 41743679,41743679_WaistCircumference.csv,4,40,40,0,0,0,0,11,Figure 4c,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero."
59
- 41743679,41743679_WaistToHipRatio.csv,5,50,50,0,0,0,0,11,Figure 4e,"All study rows, year labels, and existing numeric fields match the displayed forest plot. Main original PDF contains this complete forest panel, including all arm means/SDs/N; no supplementary file or Table 2 substitution was required. For fixed-effect plots tau2 is not displayed and is not treated as zero. Correct Figure 4 panel mapping is e=Waist-to-hip ratio, f=Arm circumference, g=Thigh circumference; previous evidence map interchanged e and g."
60
- 41761990,41761990_Albumin.csv,2,10,8,2,0,0,0,7;5,Figure 3E,
61
- 41761990,41761990_BUN.csv,4,16,16,0,0,0,0,7,Figure 3B,Forest pooled lower CI is -3.16; Table 2 p8 and results prose report -3.15. Correction follows requested forest-plot truth; retain source-discrepancy flag.
62
- 41761990,41761990_Creatinine.csv,12,108,48,60,0,0,0,7;5,Figure 3A,
63
- 41761990,41761990_Urea.csv,3,12,12,0,0,0,0,7,Figure 3D,
64
- 41761990,41761990_UricAcid.csv,3,12,12,0,0,0,0,7,Figure 3C,The forest literally prints heterogeneity p < 0.000; this is preserved as a reported string rather than interpreted as an exact probability.
65
- 41830012,41830012_AllCauseMortality.csv,11,66,66,0,0,0,0,7,Figure 2A,Forest upper CI 1.06 agrees with benchmark. Results prose above figure says 1.07; forest is used.
66
- 41830012,41830012_CardiovascularDeath.csv,7,42,42,0,0,0,0,8,Figure 2D,
67
- 41830012,41830012_MACE.csv,7,42,42,0,0,0,0,9,Figure 2E,
68
- 41830012,41830012_MI.csv,10,60,60,0,0,0,0,8,Figure 2C,
69
- 41830012,41830012_Stroke.csv,11,66,66,0,0,0,0,7,Figure 2B,
70
- 41835126,41835126_HRISymptoms.csv,12,84,72,12,0,0,0,7;4,Figure 2,"All 12 study rows verified in Figure 2. Sample Size is from Table 2 and is intentionally distinct from forest Total for rows 1 (165 vs 569), 10 (90 vs 168), and 12 (61 vs 1165). Both columns match their respective published sources; do not overwrite one using the other. Forest also shows common-effect overall proportion 0.35 [0.33,0.36], total 4430. Benchmark correctly selects random-effects 0.40 [0.24,0.58]."
71
- 41888952,41888952_HDL-C.csv,13,52,52,0,0,0,0,38,Figure 4,"Forest pooled row is -1.17 [-1.93,-0.41]; article abstract/results prose reports -1.16 [-1.92,-0.40]. Follow forest."
72
- 41888952,41888952_LDL-C.csv,11,44,44,0,0,0,0,36,Figure 5,Forest lower CI is -6.32; article abstract/results prose reports -6.31. Follow forest.
73
- 41888952,41888952_TC.csv,13,52,52,0,0,0,0,37,Figure 3,"Forest pooled row is -6.72 [-10.39,-3.05]; article abstract/results prose reports -6.71 [-10.38,-3.04]. Follow forest. Preserve row 5 WMD -0.00 exactly as printed."
74
- 41888952,41888952_TG.csv,12,48,48,0,0,0,0,35,Figure 2,Forest upper CI is 9.33; article abstract/results prose reports 9.32. Follow forest.
75
- 41901064,41901064_DBP.csv,11,110,110,0,0,0,0,8,Figure 3B,Forest measure label is MD (Mean Difference); the article calls pooled differences WMD. They are synonymous here; no unit effect_measure correction recommended. Both common and random effects rows have identical published numeric values.
76
- 41901064,41901064_SBP.csv,11,110,110,0,0,0,0,8,Figure 3A,"Forest measure label is MD (Mean Difference); the article calls pooled differences WMD. They are synonymous here; no unit effect_measure correction recommended. Benchmark matches the common-effect pooled row. Random-effects pooled row is -1.23 [-3.68,1.23] and must not be substituted without changing the selected model."
77
- 41938012,41938012_CO.csv,5,20,20,0,0,0,0,17,Figure 4A,"Thimonier row 2 point estimate 1.98 and CI [1.49,2.54] both match the printed plot; retained exactly without deriving a different value from CI."
78
- 41938012,41938012_HighTemperature.csv,12,48,48,0,0,0,0,19,Figure 6,"Forest I2 is 98.5%, not the 89.0% in MA-034; 89.0% belongs to the particulate-matter forest."
79
- 41938012,41938012_NO2.csv,8,32,32,0,0,0,0,17,Figure 4B,
80
- 41938012,41938012_Ozone.csv,7,28,28,0,0,0,0,16,Figure 3B,
81
- 41938012,41938012_ParticulateMatter.csv,11,44,44,0,0,0,0,18,Figure 5,
82
- 41938012,41938012_SO2.csv,6,24,24,0,0,0,0,16,Figure 3A,
83
- 41939286,41939286_CVDAdmission.csv,6,24,24,0,0,0,0,9,Figure 2,"Each CSV row is already a pollutant-specific pooled RR, not an individual study. All four numeric columns match Figure 2. No meta_analysis/units.csv unit exists for this outcome. Forest gives RR per 10 micrograms/m3 increment. Tau2 and overall/heterogeneity p-values are not reported in this plot. Random-effects model is reported in Methods section 2.4, PDF p4."
84
- 41939286,41939286_CVDMortality.csv,5,20,20,0,0,0,0,9,Figure 2,"Each CSV row is already a pollutant-specific pooled RR, not an individual study. All four numeric columns match Figure 2. No meta_analysis/units.csv unit exists for this outcome. Forest gives RR per 10 micrograms/m3 increment. Tau2 and overall/heterogeneity p-values are not reported in this plot. Random-effects model is reported in Methods section 2.4, PDF p4."
85
- 41939286,41939286_CVDOverall.csv,5,20,20,0,0,0,0,9,Figure 2,"Each CSV row is already a pollutant-specific pooled RR, not an individual study. All four numeric columns match Figure 2. No meta_analysis/units.csv unit exists for this outcome. Forest gives RR per 10 micrograms/m3 increment. Tau2 and overall/heterogeneity p-values are not reported in this plot. Random-effects model is reported in Methods section 2.4, PDF p4."
86
- 41970007,41970007_DBP.csv,8,80,80,0,0,0,0,27,"Supplementary Figure S16, subgroup 2.2.2",
87
- 41970007,41970007_Ozone.csv,2,20,20,0,0,0,0,9,"Figure 3, subgroup 1.7.4",
88
- 41970007,41970007_PM1.csv,4,40,40,0,0,0,0,9,"Figure 3, subgroup 1.7.1",
89
- 41970007,41970007_PM10.csv,4,40,40,0,0,0,0,9,"Figure 3, subgroup 1.7.3",
90
- 41970007,41970007_PM25.csv,11,110,110,0,0,0,0,9,"Figure 3, subgroup 1.7.2",
91
- 41970007,41970007_RespiratorySymptoms.csv,10,100,100,0,0,0,0,9,Figure 2,
92
- 41970007,41970007_SBP.csv,8,80,80,0,0,0,0,27,"Supplementary Figure S16, subgroup 2.2.1",
93
- 42003421,42003421_DBP.csv,12,48,48,0,0,0,0,8,Figure 4,units.csv uses MD; the source forest says WMD. WMD is a weighted MD and the existing normalized measure is retained; source label preserved in reference metadata.
94
- 42003421,42003421_SBP.csv,12,84,48,36,0,0,0,6,Figure 2,"These characteristic fields match the review table, not the forest plot: Baseline BMI (kg/m2), Duration (week), Sample Size. units.csv uses MD; the source forest says WMD. WMD is a weighted MD and the existing normalized measure is retained; source label preserved in reference metadata. Forest upper CI is 0.69; article narrative and existing benchmark show 0.96. Forest value takes precedence. The p=0.013 printed on the overall row is heterogeneity p, not effect p."
95
- 42015424,42015424_ESelectin.csv,5,20,20,0,0,0,0,9,Figure 4,
96
- 42015424,42015424_ICAM.csv,8,48,32,16,0,0,0,8,Figure 2,"These characteristic fields match the review table, not the forest plot: Duration (week), Sample Size. Francois year is 2017 in the forest but 2018 in Table 1. Year retains forest label; Table 1 duration/N attach by study identity."
97
- 42015424,42015424_VCAM.csv,4,16,16,0,0,0,0,9,Figure 3,
98
- 42032055,42032055_Dyspareunia.csv,4,40,40,0,0,0,0,7,"Figure 2, subgroup 2.1.2","Unusual Mucci SD=0.0286, control SD=24.6586 and Tedeschi means=4.6702/6.3762 are faithfully printed in the original forest; preserved without inference."
99
- 42032055,42032055_VaginalDryness.csv,7,77,70,7,0,0,0,7,"Figure 2, subgroup 2.1.1","These characteristic fields match the review table, not the forest plot: Sample Size. Sample Size is the Table 1 study-level N, not the forest outcome arm sum. Table N vs forest arm sums: Carmignani 60 vs40; Manonai60 vs36; Mucci89 vs58; Ribeiro60 vs38; Tedeschi186 vs189. Keep both reported values with their distinct sources; do not replace arm Ns with Sample Size. Tedeschi table/figure discrepancy remains a source limitation."
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/release_metadata_corrections.json DELETED
@@ -1,104 +0,0 @@
1
- {
2
- "source_revision": "8dfb61c0a69c7d791731b0de55c24ef8c5ef1ba6",
3
- "finding_count": 2,
4
- "numeric_values_changed": 0,
5
- "changes": [
6
- {
7
- "id": "PROV-001",
8
- "unit_id": "MA-091",
9
- "review_pmid": "41639864",
10
- "kind": "numeric_field_provenance",
11
- "path": [
12
- "numeric_fields",
13
- "tau2"
14
- ],
15
- "before": {
16
- "value": "0.030",
17
- "status": "verified_forest",
18
- "source": {
19
- "page": 6,
20
- "figure": "Figure 2",
21
- "source_type": "forest_plot"
22
- }
23
- },
24
- "after": {
25
- "value": "0.030",
26
- "status": "verified_nonforest",
27
- "source": {
28
- "page": 1,
29
- "figure": "Abstract, Results",
30
- "source_type": "nonforest_text"
31
- }
32
- },
33
- "reason": "The original abstract explicitly prints τ² = 0.030. Figure 2 prints pooled SMD/CI and I² with heterogeneity p but no τ². The existing reported_details.tau2_source and tau2_status are already correct; the consolidated numeric_fields source ignored them.",
34
- "target_file": "validation/meta_reference.json",
35
- "evidence": [
36
- {
37
- "pdf_page": 1,
38
- "locator": "Abstract, Results",
39
- "quote": "p = 0.390; I² = 32.9%; τ² = 0.030; PI = -0.5 to 0.35",
40
- "verification": "original PDF text and rendered full page"
41
- },
42
- {
43
- "pdf_page": 6,
44
- "locator": "Figure 2",
45
- "quote": "Overall, DL (I² = 32.9%, p = 0.112); -0.08 (-0.25, 0.10)",
46
- "verification": "rendered full page; τ² not printed in figure"
47
- }
48
- ]
49
- },
50
- {
51
- "id": "META-001",
52
- "unit_id": "MA-083",
53
- "review_pmid": "41656984",
54
- "kind": "auxiliary_model_metadata",
55
- "path": [
56
- "reported_details",
57
- "model"
58
- ],
59
- "before": "IV, Fixed",
60
- "after": "IV, Random",
61
- "reason": "Both estimate and graphic headers in Figure 3e explicitly read IV, Random, 95% CI. The old audit B assertion that this panel says IV, Fixed is incorrect. This is an auxiliary model-label correction; no unit target numeric value changes.",
62
- "target_file": "validation/meta_reference.json",
63
- "evidence": [
64
- {
65
- "pdf_page": 11,
66
- "locator": "Figure 3e, self-efficacy",
67
- "quote": "Std. Mean Difference / IV, Random, 95% CI; Total (95% CI) 0.18 [0.02, 0.33]; Tau² = 0.04; I² = 69%; overall P = 0.03",
68
- "verification": "rendered full page and enlarged original-PDF panel"
69
- }
70
- ]
71
- },
72
- {
73
- "id": "META-001-NOTE",
74
- "unit_id": "MA-083",
75
- "review_pmid": "41656984",
76
- "kind": "auxiliary_model_note",
77
- "target_file": "validation/meta_reference.json",
78
- "path": [
79
- "notes"
80
- ],
81
- "before": "All study rows, year labels, and existing numeric fields match the displayed forest plot. Source header explicitly says IV, Fixed while the same panel reports tau2=0.04 and I2=69%; recorded exactly as printed, without relabeling as random.",
82
- "after": "All study rows, year labels, and existing numeric fields match the displayed forest plot. Figure 3e explicitly reports IV, Random; tau2=0.04 and I2=69% are reported in the same panel.",
83
- "reason": "Remove the duplicated false assertion that Figure 3e says IV, Fixed; evidence is META-001.",
84
- "evidence": []
85
- },
86
- {
87
- "id": "META-001-COVERAGE",
88
- "target_file": "validation/outcome_coverage.csv",
89
- "selector": {
90
- "result_file": "41656984_SelfEfficacy.csv"
91
- },
92
- "column": "notes",
93
- "before": "All study rows, year labels, and existing numeric fields match the displayed forest plot. Source header explicitly says IV, Fixed while the same panel reports tau2=0.04 and I2=69%; recorded exactly as printed, without relabeling as random.",
94
- "after": "All study rows, year labels, and existing numeric fields match the displayed forest plot. Figure 3e explicitly reports IV, Random; tau2=0.04 and I2=69% are reported in the same panel.",
95
- "reason": "Same erroneous fixed-model note as META-001-NOTE.",
96
- "evidence": []
97
- }
98
- ],
99
- "numeric_source_statuses": {
100
- "verified_forest": 412,
101
- "not_reported_in_forest": 58,
102
- "verified_nonforest": 5
103
- }
104
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/source_manifest.csv DELETED
@@ -1,28 +0,0 @@
1
- review_pmid,title,source_url,source_pdf_sha256,extraction_files,meta_units
2
- 39952828,A systematic review and meta-analysis on physical activity for the treatment of anxiety in older adults,https://pubmed.ncbi.nlm.nih.gov/39952828/,46582d19c193f7f1410cf4790164c9b7d9adaf4c392519e90ed2feb2aeca53fc,1,1
3
- 40325863,The Efficacy of Non-Pharmacological Interventions for Sleep Disorders in People With Multiple Sclerosis: A Systematic Review and Meta-Analysis,https://pubmed.ncbi.nlm.nih.gov/40325863/,c5de9c8045deb79b83fc0369c4d39335e3a5edcaf3b7f404e25bdef6e9b7ea86,1,1
4
- 41065407,"Does resistance training improve pain intensity, quality of life, and disability in people with chronic nonspecific low back pain? A systematic review and meta-analysis",https://pubmed.ncbi.nlm.nih.gov/41065407/,4807f65be2df65ceb5368c89d9aa86d187300c2b17b740c0057094691c1589e3,3,3
5
- 41388744,Antihypertensive Effects of Curcumin/Turmeric Supplementation in Prediabetes and Diabetes: A Systematic Review and Meta-Analysis of Randomised Controlled Trials,https://pubmed.ncbi.nlm.nih.gov/41388744/,0a33d5f8e63756ada90326748a63a7c4111df9bb18ec7b75456cb62c2aa03d00,2,2
6
- 41507873,Association between green space exposure and elderly health: a systematic review and meta-analysis,https://pubmed.ncbi.nlm.nih.gov/41507873/,cfd0fc4ecd2907e83b11e7b727e6a4fbd9591bc255ed2bd74272aad706146213,2,2
7
- 41531667,Ambient Temperature and Risk of Renal Colic: A Systematic Review and Meta-analysis,https://pubmed.ncbi.nlm.nih.gov/41531667/,85e49d2e2613cd35a78db1bb755b4b3ccd1cb5e68d9d153751340b352a6a9d69,0,0
8
- 41561801,Effect of exercise on abdominal obesity and inflammatory response in the older adults: a systematic review and meta-analysis of randomized controlled trials,https://pubmed.ncbi.nlm.nih.gov/41561801/,db547442c990055510e8137bc2c39543116c5e99b23ef60b2c5580410c705918,6,6
9
- 41566983,Association of Particulate Matter (PM2.5) With COVID-19 Infection and Mortality in Low-and Middle-income Asian Countries: A Systematic Review and Meta-analysis,https://pubmed.ncbi.nlm.nih.gov/41566983/,5709f1bc2de413694853b26df396e81f16f72a436f8d517a8e29939ae1cc667d,3,3
10
- 41601519,Effectiveness of unguided digital cognitive behavioral therapy for insomnia on depressive symptoms: a systematic review and meta-analysis of randomized controlled trials,https://pubmed.ncbi.nlm.nih.gov/41601519/,17f65ac24ec6a70279bf1d5443672fce917ca952d951f4724586beb07ac55bb6,3,3
11
- 41617837,Associations of heat exposure with mental health and suicide in children and adolescents: a systematic review and meta-analysis,https://pubmed.ncbi.nlm.nih.gov/41617837/,775467ed3f90485cc2b6423b2bf57d465c83934820399cefb0562c6de96ce78a,6,6
12
- 41639651,Association between PM2.5 exposure and blood pressure among young adults in Asia: systematic review and meta-analysis,https://pubmed.ncbi.nlm.nih.gov/41639651/,307667f5c4e6327a41a751aa0059e79166708cb4c510c7bedf1f6a218076036f,2,2
13
- 41639864,Association between aerobic exercise and vascular endothelial growth factor (VEGF) concentration in adults: a systematic review and meta-analysis of randomized controlled trials,https://pubmed.ncbi.nlm.nih.gov/41639864/,074604aff530331dc9a21d29f304e5b8873ebc2d294ca6c6916392be8b767cca,1,1
14
- 41652340,Evidence gaps in the effects of exercise on SASP-Related biomarkers in older adults: a systematic review and meta-analysis of randomized controlled trials,https://pubmed.ncbi.nlm.nih.gov/41652340/,f7bbfd4b0aed707edd6eeeca302479193667516ab91d0a7b2bebb86348461121,4,4
15
- 41656984,"Effectiveness of Health Coaching Interventions on Anxiety, Depression, Self-Management Behaviors, Self-Efficacy, and Quality of Life Among Older Adults: A Systematic Review and Meta-Analysis of Randomized Controlled Trials",https://pubmed.ncbi.nlm.nih.gov/41656984/,9552f7b248e754cf6b070980c5de8b0adbc7108b9a178a86e9c93581bb5f60f2,5,5
16
- 41721779,Effects of inspiratory muscle training on obstructive sleep apnea: a systematic review and meta-analysis,https://pubmed.ncbi.nlm.nih.gov/41721779/,dac291b5b6e388ea58380b866ac267e73b20076d8c9ba9c1f01456af92a8cee3,0,0
17
- 41743679,Effects of aquatic exercise on improving body composition and muscle strength in the older adults: a systematic review and meta-analysis of randomized controlled trials,https://pubmed.ncbi.nlm.nih.gov/41743679/,592bf2a49d771dc85a82de00e6bce3ff3cb04d7132713fa7ec99cd232165bf29,19,19
18
- 41761990,Effects of Curcumin/Turmeric Supplementation on Kidney Function in Individuals With Diabetes: A Systematic Review and Meta-Analysis of Randomised Controlled Trials,https://pubmed.ncbi.nlm.nih.gov/41761990/,c7ea7a0505be772a3b0e83182ffe19378fdcada7d13a65b9ce41dde25cf6f82d,5,5
19
- 41830012,Complex Effects of B-Vitamin Combinations on Cardiovascular Diseases: A Systematic Review and Meta-Analysis of Randomized Controlled Trials over Three Decades,https://pubmed.ncbi.nlm.nih.gov/41830012/,9e5695a117f87018f5d8b5fecc9d67a8fe42ef805264ac22c13ad4f335c39d9c,5,5
20
- 41835126,Heat Exposure and Health Outcomes in Construction Workers: A Systematic Review and Meta-Analysis,https://pubmed.ncbi.nlm.nih.gov/41835126/,e28134ed8059dca7bc3ce8046d80fc05426d9c90b1d3cd4d069179f6bb0c17fb,1,1
21
- 41888952,The effect of hydrogen-rich water interventions on lipid profiles in adults with overweight or obesity and associated metabolic disorders: a systematic review and meta-analysis of randomized controlled trials,https://pubmed.ncbi.nlm.nih.gov/41888952/,c6c7030d1df3314bfc2df92c7cb8510575eeb4b1ba8c2a0269027fa85096babd,4,4
22
- 41901064,Effects of Nicotinamide Mononucleotide Supplementation on Blood Pressure: A Systematic Review and Meta-Analysis of Randomized Controlled Trials,https://pubmed.ncbi.nlm.nih.gov/41901064/,8d3c137ec223734bc8c2d7f398a6c72476ec9e1b5b45713d8dd3d7616ef2a9fc,2,2
23
- 41938012,Impact of climate change on spontaneous abortion: a systematic review and meta-analysis,https://pubmed.ncbi.nlm.nih.gov/41938012/,1f82671d3f782209d8d4554cc19f30eb7156a5817d12805794cfd6aaf6ab4200,6,6
24
- 41939286,Impact of Ambient Air Pollution on Cardiovascular Diseases in Low- and Lower-Middle-Income Countries: A Systematic Review and Meta-Analysis,https://pubmed.ncbi.nlm.nih.gov/41939286/,02bb26844b2fcdcdfbc1d3bda26aa872d0654cc6da992cacaa73c9ebfd72fd6c,3,0
25
- 41970007,Effectiveness of Personal-Level Protective Measures Against Air Pollution on Respiratory Health: A Systematic Review and Meta-Analysis of Randomized Controlled Trials,https://pubmed.ncbi.nlm.nih.gov/41970007/,399f07aeff8d976cf3975f0a304829160d7bac0ad069109442a3a9b432391acb,7,7
26
- 42003421,The Effect of Cranberry Consumption on Blood Pressure: A Systematic Review and Meta-Analysis of Randomized Controlled Trials,https://pubmed.ncbi.nlm.nih.gov/42003421/,202eb3f28bd5460731da9091532d1a07cbf7b4af02a15e4c7685d9fead684329,2,2
27
- 42015424,The effect of low-carbohydrate diets on serum concentration of soluble adhesion molecules: a systematic review and meta-analysis of randomized controlled trials,https://pubmed.ncbi.nlm.nih.gov/42015424/,47bc65a65e3a1aebe57a04ac5efff86911eb5a0e387cbe0e4b5e9fc654318f11,3,3
28
- 42032055,Soy isoflavone supplementation and sexual function in postmenopausal women: a systematic review and meta-analysis of randomized controlled trials,https://pubmed.ncbi.nlm.nih.gov/42032055/,94fd0f4a21da73a2af7bcbd8a2b1ee0221c37ad340f2384f70ac82fbb8127cba,2,2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/summary.json DELETED
@@ -1,112 +0,0 @@
1
- {
2
- "reviews_in_inventory": 27,
3
- "reviews_with_extraction": 25,
4
- "reviews_with_meta": 24,
5
- "outcome_tables": 98,
6
- "rows": 747,
7
- "numeric_cells_checked": 4850,
8
- "numeric_source_counts": {
9
- "forest_plot": 4576,
10
- "nonforest_table": 272,
11
- "nonforest_reference": 1,
12
- "derived_from_reported_table": 1
13
- },
14
- "unchecked_numeric_cells": 0,
15
- "extraction_changes": {
16
- "numeric_value_change": 1,
17
- "identifier_or_direction": 1,
18
- "display_only": 1
19
- },
20
- "meta_changes": {
21
- "numeric_value_change": 23,
22
- "fill_reported_blank": 50,
23
- "align_forest_rounding": 16,
24
- "display_only": 3,
25
- "remove_mislabeled_tau": 4,
26
- "identifier_or_direction": 3
27
- },
28
- "meta_units_changed": 59,
29
- "total_change_entries": 102,
30
- "nonforest_existing_meta_fields": [
31
- {
32
- "unit_id": "MA-011",
33
- "field": "i2",
34
- "value": "99.27",
35
- "status": "verified_nonforest",
36
- "source": {
37
- "page": 4,
38
- "figure": "Results: corresponding PM2.5 association",
39
- "source_type": "nonforest_text"
40
- }
41
- },
42
- {
43
- "unit_id": "MA-012",
44
- "field": "i2",
45
- "value": "76.49",
46
- "status": "verified_nonforest",
47
- "source": {
48
- "page": 4,
49
- "figure": "Results: corresponding PM2.5 association",
50
- "source_type": "nonforest_text"
51
- }
52
- },
53
- {
54
- "unit_id": "MA-027",
55
- "field": "i2",
56
- "value": "91.002",
57
- "status": "verified_nonforest",
58
- "source": {
59
- "page": 10,
60
- "figure": "Results section 3.5",
61
- "source_type": "nonforest_text"
62
- }
63
- },
64
- {
65
- "unit_id": "MA-027",
66
- "field": "tau2",
67
- "value": "1.731",
68
- "status": "verified_nonforest",
69
- "source": {
70
- "page": 10,
71
- "figure": "Results section 3.5",
72
- "source_type": "nonforest_text"
73
- }
74
- },
75
- {
76
- "unit_id": "MA-091",
77
- "field": "tau2",
78
- "value": "0.030",
79
- "status": "verified_nonforest",
80
- "source": {
81
- "page": 1,
82
- "figure": "Abstract, Results",
83
- "source_type": "nonforest_text"
84
- }
85
- }
86
- ],
87
- "meta_numeric_field_statuses": {
88
- "verified_forest": 412,
89
- "not_reported_in_forest": 58,
90
- "verified_nonforest": 5
91
- },
92
- "skipped_changes": [
93
- {
94
- "file": "meta_analysis/units.csv",
95
- "row": "MA-013",
96
- "unit_id": "MA-013",
97
- "column": "i2",
98
- "old": "",
99
- "new": "89.72",
100
- "evidence_page": 4,
101
- "evidence_figure": "Results: corresponding PM2.5 association",
102
- "reason": "Published value is present in original source but target is blank.",
103
- "confidence": "high",
104
- "category": "missing_reported_value",
105
- "source_type": "nonforest_text",
106
- "reviewer": "a",
107
- "disposition": "not_added_to_units; reported only in narrative and previously blank"
108
- }
109
- ],
110
- "direction_evaluable_units": 94,
111
- "single_group_prevalence_units": 1
112
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/validate.py DELETED
@@ -1,129 +0,0 @@
1
- #!/usr/bin/env python3
2
- """Check benchmark drift against the 2026-09-29 visually reviewed source ledger.
3
-
4
- This does not perform a new visual/OCR audit of a PDF. PDF hashes identify the
5
- version used by reviewers; the source ledger records their numeric transcription.
6
- """
7
- from __future__ import annotations
8
- import argparse
9
- import csv
10
- import hashlib
11
- import json
12
- from decimal import Decimal, InvalidOperation
13
- from pathlib import Path
14
-
15
-
16
- def rows(path):
17
- with path.open(encoding='utf-8-sig', newline='') as handle:
18
- reader = csv.DictReader(handle)
19
- fields = reader.fieldnames or []
20
- if not fields or len(set(fields)) != len(fields):
21
- raise ValueError(f'Invalid or duplicate headers: {path}')
22
- result = list(reader)
23
- if any(None in row or any(v is None for v in row.values()) for row in result):
24
- raise ValueError(f'Malformed CSV row: {path}')
25
- return result
26
-
27
-
28
- def numeric(value):
29
- try:
30
- return Decimal(value).is_finite()
31
- except (InvalidOperation, ValueError, TypeError):
32
- return False
33
-
34
-
35
- def digest(path):
36
- return hashlib.sha256(path.read_bytes()).hexdigest()
37
-
38
-
39
- def validate(root, pdf_dir=None):
40
- validation = root / 'validation'
41
- errors = []
42
- ledger = rows(validation / 'extraction_numeric_cells.csv')
43
- reference = json.loads((validation / 'meta_reference.json').read_text())
44
- source_manifest = rows(validation / 'source_manifest.csv')
45
- files = sorted((root / 'data_extraction/results').glob('*.csv'))
46
- tables = {'data_extraction/results/' + f.name: rows(f) for f in files}
47
- actual_keys = {(f, i, c) for f, rs in tables.items()
48
- for i, row in enumerate(rs, 1) for c, v in row.items() if numeric(v)}
49
- expected_keys = set()
50
- for item in ledger:
51
- key = (item['file'], int(item['data_row']), item['column'])
52
- if key in expected_keys:
53
- errors.append(f'Duplicate source-ledger entry: {key}')
54
- expected_keys.add(key)
55
- try:
56
- actual = tables[key[0]][key[1] - 1][key[2]]
57
- if Decimal(actual) != Decimal(item['source_value']):
58
- errors.append(f'Extraction mismatch {key}: {actual} != {item["source_value"]}')
59
- except (KeyError, IndexError, InvalidOperation):
60
- errors.append(f'Missing or nonnumeric extraction cell: {key}')
61
- if not item['source_type'] or not item['pdf_page'] or not item['figure_or_table']:
62
- errors.append(f'Missing provenance: {key}')
63
- if actual_keys != expected_keys:
64
- errors.append(f'Numeric coverage mismatch: {len(actual_keys - expected_keys)} unreviewed; '
65
- f'{len(expected_keys - actual_keys)} missing')
66
- units = rows(root / 'meta_analysis/units.csv')
67
- by_id = {u['unit_id']: u for u in units}
68
- if len(by_id) != len(units):
69
- errors.append('Duplicate meta unit IDs')
70
- if set(by_id) != {r['unit_id'] for r in reference}:
71
- errors.append('Meta unit scope differs from reviewed reference')
72
- for ref in reference:
73
- uid = ref['unit_id']
74
- actual = by_id.get(uid, {})
75
- if actual != ref['expected_unit']:
76
- errors.append(f'Meta row differs from reviewed reference: {uid}')
77
- for key, field in ref['numeric_fields'].items():
78
- value = actual.get(key)
79
- if field['status'] == 'not_reported_in_forest':
80
- if value != '':
81
- errors.append(f'Unreported meta field populated: {uid}.{key}')
82
- elif not numeric(value) or Decimal(value) != Decimal(field['value']):
83
- errors.append(f'Meta numeric mismatch: {uid}.{key}')
84
- if actual:
85
- effect = Decimal(actual['pooled_effect'])
86
- if not Decimal(actual['ci_lower']) <= effect <= Decimal(actual['ci_upper']):
87
- errors.append(f'Invalid pooled CI: {uid}')
88
- threshold = Decimal(1) if actual['effect_measure'] in ('OR', 'RR') else Decimal(0)
89
- direction = 'positive' if effect > threshold else 'negative' if effect < threshold else 'null'
90
- if actual['analysis_type'] == 'proportion':
91
- direction = 'not_applicable'
92
- if actual['direction'] != direction:
93
- errors.append(f'Wrong pooled direction: {uid}')
94
- if actual['i2'] and not 0 <= Decimal(actual['i2']) <= 100:
95
- errors.append(f'I2 out of range: {uid}')
96
- if actual['tau2'] and Decimal(actual['tau2']) < 0:
97
- errors.append(f'Negative tau2: {uid}')
98
- if 'data_extraction/results/' + actual['result_file'] not in tables:
99
- errors.append(f'Missing linked extraction file: {uid}')
100
- for entry in rows(validation / 'verified_file_hashes.csv'):
101
- p = root / entry['file']
102
- if not p.is_file() or digest(p) != entry['sha256']:
103
- errors.append(f'File drift from reviewed version: {entry["file"]}')
104
- reviews = [json.loads(line) for line in (root / 'reviews.jsonl').read_text().splitlines()]
105
- if {r['review_pmid'] for r in reviews} != {r['review_pmid'] for r in source_manifest}:
106
- errors.append('Source manifest does not cover review inventory')
107
- checked_pdfs = 0
108
- if pdf_dir:
109
- for source in source_manifest:
110
- p = pdf_dir / (source['review_pmid'] + '.pdf')
111
- if not p.is_file() or digest(p) != source['source_pdf_sha256']:
112
- errors.append(f'Source PDF missing or changed: {source["review_pmid"]}')
113
- else:
114
- checked_pdfs += 1
115
- result = {'status': 'pass' if not errors else 'fail', 'extraction_files': len(files),
116
- 'extraction_rows': sum(map(len, tables.values())),
117
- 'numeric_cells_checked': len(expected_keys), 'meta_units': len(units),
118
- 'source_pdfs_hash_checked': checked_pdfs, 'errors': errors}
119
- return result
120
-
121
-
122
- if __name__ == '__main__':
123
- parser = argparse.ArgumentParser(description=__doc__)
124
- parser.add_argument('--benchmark-root', type=Path, default=Path(__file__).resolve().parents[1])
125
- parser.add_argument('--source-pdf-dir', type=Path)
126
- args = parser.parse_args()
127
- result = validate(args.benchmark_root, args.source_pdf_dir)
128
- print(json.dumps(result, indent=2, ensure_ascii=False))
129
- raise SystemExit(0 if result['status'] == 'pass' else 1)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/verification_result.json DELETED
@@ -1,15 +0,0 @@
1
- {
2
- "status": "pass",
3
- "extraction_files": 98,
4
- "extraction_rows": 747,
5
- "numeric_cells_checked": 4850,
6
- "meta_units": 95,
7
- "source_pdfs_hash_checked": 27,
8
- "errors": [],
9
- "changed_cells_reconciled_to_log": 102,
10
- "negative_controls": {
11
- "reintroduced_sample_size_error_detected": true,
12
- "invalid_prevalence_direction_detected": true
13
- },
14
- "audit_date": "2026-09-29"
15
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
validation/verified_file_hashes.csv DELETED
@@ -1,100 +0,0 @@
1
- file,sha256
2
- data_extraction/results/39952828.csv,ab69d7b4a02eb16a34d4c54ac6cd93f50d112ec77fcf0605a4137d32f9f0910c
3
- data_extraction/results/40325863.csv,b677b41437f46196da0d2acef92398ed71120ad6bf82d077068107bba4ad5e91
4
- data_extraction/results/41065407_Disability.csv,db469de9821266ab934a0bc830236751ce88e4f0890458d9caf8f30d2b02ce03
5
- data_extraction/results/41065407_PainIntensity.csv,bb4780065cd7484606292700c2975b26d08b0dd2bc1ce7511180dfcb3df21977
6
- data_extraction/results/41065407_QoL.csv,76eb847d8364598c07c3af4120efbbae7e88729abaecc2b48efa49b65a81b003
7
- data_extraction/results/41388744_DBP.csv,302e00fdc7b7330368607b78c9ea6884601b71edd136d496952ceb71170732a1
8
- data_extraction/results/41388744_SBP.csv,6f6876b562957fdeea8bbbff171131bcdb3e71f2b65632d15350e61ad3dd2163
9
- data_extraction/results/41507873_Alzheimer.csv,13d8834f7fc08c556f82195b6d514a03dd3225a00bfbfa52b5928235c09e99df
10
- data_extraction/results/41507873_Depression.csv,8f48ec7b8d4acc7ce2828009a0dd99dcc2d9282543cd3fa19e29a64bcca5f6a5
11
- data_extraction/results/41561801_Adiponectin.csv,4f256ec48532e8973c25e0ec1709b9b08eff4e4a57714324d06949f2bc308a9d
12
- data_extraction/results/41561801_BMI.csv,53a0a645977f65729f0fab670c749c52bdb9dbd5309183ccb2ae568c2c5fee60
13
- data_extraction/results/41561801_CRP.csv,3c8a85c0557a35debfb88036d69f9051d30cd3f8d9b56190b9d3a7fc0439c6a2
14
- data_extraction/results/41561801_IL-6.csv,b6372f18cb07fcc95f5ae4d424c9696b272322f3b644e4c736c6578a529b5304
15
- data_extraction/results/41561801_TNF-alpha.csv,dc1cec33b5a87de94ddc22122c5548fc0b9cd2997d10d2cffd886f694dc7a575
16
- data_extraction/results/41561801_WC.csv,80ab0563920087cbc56322a0b7f1dc6c314c596864854bf9e4af44f55c04e339
17
- data_extraction/results/41566983_InfectionLongTerm.csv,25ce7316119156b7a4bc3215631d7072e7d5183fe128349c0484455a83a9626a
18
- data_extraction/results/41566983_InfectionShortTerm.csv,34048b8cbf2ee5d2c4b518767ded399b41fd8d47629425f432ef96c738104fed
19
- data_extraction/results/41566983_MortalityShortTerm.csv,3e128ff172c2321b8666dfec1f0a9ef76552f3b68833d8776f26c55574afda74
20
- data_extraction/results/41601519_DepressiveSymptoms.csv,fc2c481f2fa8b312c742be83712e37daa65f8cb743f0b87c17c30ae73bede006
21
- data_extraction/results/41601519_InsomniaSeverity.csv,1b34b0951afc3bea892754a5fb02d91c8bbe956e25c17ea78b9beaca307300c8
22
- data_extraction/results/41601519_SleepEfficiency.csv,e65085f81c2cb7979f6291a186cfd51ccd14dc9aad51cb519fed6538d8d35cc4
23
- data_extraction/results/41617837_Anxiety.csv,5492345260178eea01b50f242ed84ae99fe91c383f2200232e3290f3e7d55671
24
- data_extraction/results/41617837_CompositeMentalIllnesses.csv,e5f6194542301012b35b01928ede3e6690aaa0df726fbdad5adf374e52e7dddd
25
- data_extraction/results/41617837_Depression.csv,6e3dd3baba4e0da3c8847df83a3a65898785099481cee6e229055238df03f014
26
- data_extraction/results/41617837_MentalHealthDisorders.csv,69d4f1b63023b05e2b0e7c0f339c5172ee24563d97104ac37b5aa7b12c819916
27
- data_extraction/results/41617837_SchizophreniaPsychoticDisorders.csv,0ff6c0efd97c3c3c4a4163a070dd320cddf264c39644a49d21bf7f6d0b752b99
28
- data_extraction/results/41617837_Suicide.csv,ad9f78f3cd1980ca180c20d431456092627482a1ff68a32932a42634d832b161
29
- data_extraction/results/41639651_DBP.csv,67afd03785b04329da2bb91ce7d5dd5a786b60d84a1b8c71a204b6763930c71e
30
- data_extraction/results/41639651_SBP.csv,f7e79fb4832726552cdfae20b79b6b8c14f2e1492a539a6caff92fee8c4a96b9
31
- data_extraction/results/41639864_VEGF.csv,99e24461dd37587f27079548919cc8fb1d50ffcdfbb94da4550b12b59945423f
32
- data_extraction/results/41652340_CCL2.csv,a2369741448a745b40cf100479a73f439eb3710f033981894738e9d0ccc26101
33
- data_extraction/results/41652340_IL-6.csv,554d64231d7c8639031887aa3727e5cc5c8a16d3066579a18d32106a3dc7e98c
34
- data_extraction/results/41652340_IL-8.csv,a951bdf9316a7798dd8ccf6b6b5755e6d9fc75ac9cdeeb9ec6422a71a97e2eb0
35
- data_extraction/results/41652340_TNF-alpha.csv,5a06fb47fb93f49707bdda5323bbb3686d0860b01656e44a9b0fdd320a82f969
36
- data_extraction/results/41656984_Anxiety.csv,26230eec426958d87513bc9c1d8752fe8a110b14ce201b0bfecbc9b7bd17fb65
37
- data_extraction/results/41656984_Depression.csv,b30d06b072734ce8f4e88c6ba0898aa43a8a0b8da79cad8949d34630ad1605cf
38
- data_extraction/results/41656984_QualityOfLife.csv,787ecac4f6dca5af57b18f3aa0b092b19291e4321ed82d931984c8f49c129d16
39
- data_extraction/results/41656984_SelfEfficacy.csv,295221dd77da967719afea223ff92b9bc9739321cbe6f2e7b3f719c259e31e27
40
- data_extraction/results/41656984_SelfManagementBehavior.csv,6ea30f8c43cc0c06d3d3ad941f759b17f8c7d1a93d95264c0566e750830fecfc
41
- data_extraction/results/41743679_30CST.csv,ce97f772f988483c4a7d3906fa976f983ed4e63bd5a9163c1004d468565e2756
42
- data_extraction/results/41743679_ArmCircumference.csv,24f0e75742a60b66429e38e9c6ec76cb721a9d7e3d9e6c83f56c0305c40417cc
43
- data_extraction/results/41743679_ArmCurlTest.csv,3106e7d60a12aa2c86693e434df5c72995a5ef3a02895130836b84406878e293
44
- data_extraction/results/41743679_BMI.csv,893e8e8dc27c8ca28d27be0f71395315989bfaa664de521317d952f498c663f8
45
- data_extraction/results/41743679_BackScratch.csv,356aa217cf5d8b4d28afd1c58437fe8c8f9849253259e3863215495fb6fd4f27
46
- data_extraction/results/41743679_BodyFatMass.csv,5b8bd3205ddee84ae786c1831b0c21c74f0d6dd8d57903b519bacd44018ba0e5
47
- data_extraction/results/41743679_BodyFatPercentage.csv,f0e86455cdce3b6eebfc7074661bf7683688a7f8bc7a411851c4d90605aa2ee4
48
- data_extraction/results/41743679_BodyWeight.csv,0b86efd0410ba0baed94be1a0069858029595a6e77675e7fe6e7d99e19f82fde
49
- data_extraction/results/41743679_HDL-C.csv,2137adae1179c9eab5f6d50953a4b49c8f8cb1eac32f9dd03a6e601955e3436b
50
- data_extraction/results/41743679_HandGripTest.csv,02d87a8855e30081755ed408a1e1ae4b794f4ef9fbbe25929bcb2c15a97b4f6b
51
- data_extraction/results/41743679_HipCircumference.csv,cbe9f46a55a34ded250b5f0a0281e188eb4818873077786ce950f86dc18d5588
52
- data_extraction/results/41743679_LDL-C.csv,36c8403cce27d1512707cff65d643d1a0ecf43923a77cad9b19ade5d0c09f9c8
53
- data_extraction/results/41743679_SitAndReach.csv,c9ac0cf3a28a663afac70d797a857fbcfed3d260594951d7acfcd96310fb30f9
54
- data_extraction/results/41743679_TUG.csv,2e9942f9945c43302a17d370114aa1369b3fc43ac38d24cdea62da397e87945f
55
- data_extraction/results/41743679_ThighCircumference.csv,f4b18e552b63c5a38d0c0f3048afb412d91fe6abf9efa1b5e64933e8a2097250
56
- data_extraction/results/41743679_TotalCholesterol.csv,fc97d5482172b5574d3b3091fe2cba2f78f4668696afd57c2f55ec024221ff38
57
- data_extraction/results/41743679_Triglycerides.csv,888009ab6214ad5ddb85dfa215b1abda2aa253888dc36215e97f33bfe0741657
58
- data_extraction/results/41743679_WaistCircumference.csv,bd2ccd4565f3c9b0cb3dffe226e52dd05e9ead3461923f05af430bccdab3679e
59
- data_extraction/results/41743679_WaistToHipRatio.csv,b9eb86926ef2e29c8e62f85196188b3b99f07b452404a76e652d409dd9c6d503
60
- data_extraction/results/41761990_Albumin.csv,cfd05c234a89e50a8cf6c6751f2c971f821029961395503197362a2d6d691c81
61
- data_extraction/results/41761990_BUN.csv,5d9fc62d40c24b28bd9a1fcf5917163b14a6fab0c91cc38e10f7be608be159f9
62
- data_extraction/results/41761990_Creatinine.csv,5b23b3d41a7484c49959853bbe8f54a043983d5de57136ffda0f33c8e357ce6d
63
- data_extraction/results/41761990_Urea.csv,6addb0ce839b6f544a061dee41fb165f0064c9be62f8bd3716c8b90e064f32b5
64
- data_extraction/results/41761990_UricAcid.csv,fa6020c8415206802ef6112aabe47720c5e2d11e351b8f8d86772c8539c63ac4
65
- data_extraction/results/41830012_AllCauseMortality.csv,5f05211d58b88d6b8af8a3f7ab14b34ba6203a2bf33f3964aa40dff0b7bc534b
66
- data_extraction/results/41830012_CardiovascularDeath.csv,b5024cac3f5c0dbd11e8f2b6b8d180ab1a08946e598d447bc768a039737e5890
67
- data_extraction/results/41830012_MACE.csv,95381883bf5023cdfd9343312c401c58bb99088c99ceb895a1175a08c42f3440
68
- data_extraction/results/41830012_MI.csv,61b316b1220e811c1a0007fddbfb1bbd08504162e3b164948698a8b9b879d85c
69
- data_extraction/results/41830012_Stroke.csv,6a9e9c6180309791a5c08e6d0fe33a7ab63f3da34b48b91c49ab6929caef996c
70
- data_extraction/results/41835126_HRISymptoms.csv,6db55ed06cc167e4629a491f20661985074cee55d3de4fa56824eba3aa318316
71
- data_extraction/results/41888952_HDL-C.csv,5228ae5484833de34463edde58a50bdf82289ed20da6be4de567c9eb0e2bf984
72
- data_extraction/results/41888952_LDL-C.csv,d8901e6c0ea950747295a79ecb4a05451758bb25b1e4c31dad95436a17a005a5
73
- data_extraction/results/41888952_TC.csv,a7b89b3c9eaa220462d55704656f142d8d44b15a5616597754237a1cc027cdda
74
- data_extraction/results/41888952_TG.csv,ab2b7ee72479e9bc054a4c589f5a48791147629a6b82a1fcf0ae0c9a38fdeb4e
75
- data_extraction/results/41901064_DBP.csv,cc04ae22200aee8625a491dd2607aad2ccd6f0571dfcf3a647cbf063dd256c9f
76
- data_extraction/results/41901064_SBP.csv,6609ad4efd4363b6adaeebdceeefc0eea042d195bf67bc56825dd07bac6065e1
77
- data_extraction/results/41938012_CO.csv,14f2393fa1c4e0505a234aed399619c7734a3735fd3fa55dc596c1a4f10cb560
78
- data_extraction/results/41938012_HighTemperature.csv,46240c9a817c09a857f6afd93ed5166eda0137098083869c9406cb3b27182600
79
- data_extraction/results/41938012_NO2.csv,55cb4af1d7546fbd2fef28ee2d8af37a236b5047368e90d1ec53d258300e34ca
80
- data_extraction/results/41938012_Ozone.csv,c11e558d1e162d9108d6215bb5f126f8535e13ce5f429d1a7bac7670f93272d7
81
- data_extraction/results/41938012_ParticulateMatter.csv,7897d2e0aede1ce841af9ae1e942ead79e1599201feb137639f0bf5c71060bf8
82
- data_extraction/results/41938012_SO2.csv,5bf1362ee432be5ae0c501ee4ad7f770c774ce674d8e32bcb6804b5e3cfb0bdf
83
- data_extraction/results/41939286_CVDAdmission.csv,da7e4abbe74b9d7f415d131c541c5c6458fcc45553e126ae7320fc02c01a1674
84
- data_extraction/results/41939286_CVDMortality.csv,feeb0cba638e3203029ec6bb1edee109ede99278602d1a7a507c2db37fa32f67
85
- data_extraction/results/41939286_CVDOverall.csv,48b5a100853cf9cfa9c859ae67a5715ca48ad160c4bcc2f70a8eaaa8440416e7
86
- data_extraction/results/41970007_DBP.csv,f0f4af571ed53cac6541f1d68cb550965e0032a8b7c8d048bbae90aaddd6f6dd
87
- data_extraction/results/41970007_Ozone.csv,9f7b4c107bb745b48e9e5f8f0fbaa4aa0ff2febe766e157620b330bd2f82f4ae
88
- data_extraction/results/41970007_PM1.csv,1e38eacf2fae9080243ff7ffc469bf34d520f76621627d80129d61888c91fffd
89
- data_extraction/results/41970007_PM10.csv,520e3d0c3d1b4c1c69c199f92ef13eb94bd78337412a3c3ec445dc893a607d21
90
- data_extraction/results/41970007_PM25.csv,eed6cf59684adbb412cd476112c8202b27c898f5bb6dc5e6787799e389cf16d8
91
- data_extraction/results/41970007_RespiratorySymptoms.csv,a9ee3e25fe2ed85ecdf7f0154a2160d51fcd3cf8651fb5123be0641141ce2948
92
- data_extraction/results/41970007_SBP.csv,396dc28284a7f9f939df7037dcec90d316a863b250f59735fdb92571c65fd706
93
- data_extraction/results/42003421_DBP.csv,f92c2189652e5568054ddad4ac1d5d97133960effb4c27e210ade09375d63ff6
94
- data_extraction/results/42003421_SBP.csv,26e7447c29bcbbb45c2c2e4bc761df5868f9ca4022ebddc27ccf7328c34cc73e
95
- data_extraction/results/42015424_ESelectin.csv,8df30cf4ac2000dcd81fb5fc234b71d8b7bdd92621dbd790f5c4b46bf724fd30
96
- data_extraction/results/42015424_ICAM.csv,3869ab53fdd62251c3c4f7e095ac189584495cbbdbd3cebc95d87bcf6ae643e3
97
- data_extraction/results/42015424_VCAM.csv,7270929d1a50e1303345cca9d94726270bd26cdb495e157b687970a30bb74da1
98
- data_extraction/results/42032055_Dyspareunia.csv,4f4508a645c0437a76eb4cd822226e6923f4db0f44e6605acbafac9023108b6c
99
- data_extraction/results/42032055_VaginalDryness.csv,2bd2d8c09564bca8711a26976aafc0288d711b32d4703eecdb02292eabd075ca
100
- meta_analysis/units.csv,9ae4859ed914a1322aa426170606160712f510669c820d34781b4e72a1114178