Same-value cluster audit of Taiwan's public works carbon emission factor compilation: data, programs and operating manual

Data, programs and operating manual for the manuscript "Numerical agreement is not source verification: tracing emission factors shared across agencies in Taiwan's public works compilation" (submitted to Cleaner Environmental Systems). Version 1.17 contains all files of version 1.16 unchanged, plus the programs and outputs added for manuscript revision 3.7 (Supplementary Material S7.10–S7.14): matching-rule sensitivity on one population; populations and set relations of the counts; the compilation's life-cycle scope field and its labels by body; rule-based concrete comparators; a key variant ignoring punctuation in names; a same-key test of tracing upstream revisions; annotations against values; a corrected S7 program (one 2024 factor supplied to the F3 substitution); number checks for revisions 3.2–3.7; and a requirements file. The source documents are agency publications and are not redeposited. See README.md in the archive. 〈Numerical agreement is not source verification: tracing emission factors shared across agencies in Taiwan's public works compilation〉(投 Cleaner Environmental Systems)之資料、程式與操作手冊。 第 1.17 版含第 1.16 版全部檔案(未改動),另加修訂稿 v3.7(補充資料 S7.10–S7.14)所依之程式與輸出:同一母體下之比對規則敏感度、各計數之母體與集合關係、彙編生命週期範疇欄與其機關層級、程式化選擇之混凝土比較係數、名稱忽略標點之寬鍵變體、上游修訂追蹤之同鍵檢驗、註記與數值之關係、S7 程式修正版(F3 補入一個 2024 年係數)、稿件 v3.2–v3.7 數字比對閘,以及套件版本檔。來源文件為機關出版品,不再存放。詳見壓縮檔內 README.md。 Description of version 1.16 (kept for reference) Data, programs and operating manual for the manuscript "Source Transparency, Cross-Agency Consistency, and Comparability of Emissions Estimates in Taiwan's Public Works" (submitted to Cleaner Environmental Systems). Version 1.16 contains all files of version 1.15 unchanged, plus the programs and outputs behind Supplementary Material S7 of manuscript revision 3.0: a row-level check of the 42 note-5 rows (and all 308 rows carrying note 5) against the Water Resources Agency's web table; precision differences between that agency's compiled and published values; source dependence in the 49 clusters by row, cluster, statement and document; factor substitution on fixed quantities for 36 standard water works; document and edition evidence; control cases; and a check of the manuscript's numbers against the outputs. The source documents are agency publications and are not redeposited. See README.md in the archive. 〈Source Transparency, Cross-Agency Consistency, and Comparability of Emissions Estimates in Taiwan's Public Works〉(投 Cleaner Environmental Systems)之資料、程式與操作手冊。 第 1.16 版含第 1.15 版全部檔案(未改動),另加修訂稿 v3.0 補充資料 S7 所依之程式與輸出:註 5 之 42 列(及註 5 全部 308 列)與水利署網頁表之逐列查核、該署彙編值與公開值之精度差異、49 群之來源依賴(列、群、聲明、文件四種單位)、36 項標準水利工程固定數量下之係數替換、文件與版次證據、正反對照,以及稿件數字比對閘。來源文件為機關出版品,不再存放。詳見壓縮檔內 README.md。 Description of version 1.15 (kept for reference) Data, programs and operating manual for the manuscript "Source Transparency, Cross-Agency Consistency, and Comparability of Emissions Estimates in Taiwan's Public Works" (submitted to Cleaner Environmental Systems). Version 1.15 contains all files of version 1.14 unchanged, plus an addendum (addendum-2026-10/) with the programs and outputs of the analyses added in October 2026, including those behind Table 3 and Supplementary Material S6; the input and output of an AI reading of 42 annotation contexts; the generator of Supplementary Material S4 with a list of its hand revisions; checks of manuscript numbers; a corrected manifest with LF line endings; and a transcription of the Water Resources Agency web table of item carbon emission factors (Open Government Data License, version 1.0, Taiwan). The source documents are agency publications and are not redeposited. See README.md in the archive. 〈Source Transparency, Cross-Agency Consistency, and Comparability of Emissions Estimates in Taiwan's Public Works〉(投 Cleaner Environmental Systems)之資料、程式與操作手冊。第 1.15 版含第 1.14 版全部檔案(未改動),另加補充資料夾 addendum-2026-10/:2026 年 10 月新增分析之程式與輸出(含 Table 3 與補充資料 S6 所依者)、42 個註記情境之 AI 判讀輸入與輸出、補充資料 S4 之產生器及其人工修訂清單、稿件數字比對閘、改以 LF 換行之 MANIFEST,以及經濟部水利署網頁「工項碳排係數」表之轉錄(政府資料開放授權條款-第1版)。來源文件為機關出版品,不再存放。詳見壓縮檔內 README.md。 Description of version 1.14 (kept for reference) Data, programs and operating manual for the manuscript "Auditing source attribution in a construction carbon emission factor compilation: a same-value cluster procedure applied to Taiwan's public works table" (manuscript version 0.40). Contents: extracted coefficient tables of six Taiwanese agency carbon reduction guidelines and a transcription of the national compilation (edition of 16 June 2026); alignment, cluster, two-state and robustness outputs; the statistics file; the eleven programs with their tests; the operating manual (v1.11); the annotation census and evidence tables; identifiers and source counts of the ÖKOBAUDAT data stock OBD_2024_II (retrieved 2026-10-09); and process records, with the two programs that produced them. The source documents are agency publications and are not redeposited. The programs were written and run with Claude (Anthropic) under the author's direction. Integrity: MANIFEST-sha256.txt inside the archive. Changes from version 1.13: the operating manual moves to version 1.11 (file operating_manual_v1.11.md), whose history no longer names manuscript versions; the README states that stats.py also reads the compilation's earlier edition (113.05), not redeposited, for two checks of the manuscript, so those counts cannot be recomputed from this deposit alone; the tests follow manuscript version 0.40. No cluster, direction or member count changes. Changes from version 1.12 to 1.13: program comments in count.py, make_figures.py and test_e11.py no longer name persons or refer to a manuscript reviewed for another journal; the README states that the transcription of the national compilation keeps its printed values and units verbatim, including rows whose value is of a per-kilogram magnitude under a per-tonne unit (e.g. PCC-0822 and PCC-0823), which are not corrected and enter no same-value cluster under the strict key; the tests follow manuscript version 0.39. No cluster, direction or member count changes. Changes from version 1.11 to 1.12: the tests follow manuscript version 0.38 (version 1.11 carried the tests of version 0.36); the header of robustness.py no longer attributes the two robustness forms to a named person; the version string in the statistics file is updated (it had not been since an earlier round, though it said it would be). The operating manual is unchanged (version 1.10). No cluster, direction or member count changes. Changes from version 1.10 to 1.11: the operating manual moves to version 1.10 (file operating_manual_v1.10.md), stating that a project name alone is read as a self-estimate on any body's rows (reading another agency's project name as a citation is a sensitivity) and naming the fifth source class as type word only or otherwise unresolved; stats.py adds two counts splitting the compilation rows classed as the submitter's own document into those reached only through the alias of the low carbon building alliance (112) and the rest (1,630); the README gives the sewerage and railway guidelines' titles as reference guidelines; the tests follow manuscript version 0.36. No cluster, direction or member count changes. Changes from version 1.9 to 1.10: the operating manual moves to version 1.9 (file operating_manual_v1.9.md), stating the shared-name condition as the code applies it (among two or more bodies whose annotations yield an institution, platform, database name or document-type word, all share one) and setting aside annotations marking no coefficient before the six source classes; evidence_tables.py and its outputs (空白註記核對.csv, 第十五輪證據表.json) no longer assert that the two blank railway cells continue a merge from the previous page, and compute the judgement for every source text visible on the row's page and the previous one (none changes an established count); the tests follow manuscript version 0.35. No cluster, direction or member count changes. Changes from version 1.8 to 1.9: the operating manual moves to version 1.8 (file operating_manual_v1.8.md), stating the external-source conditions as shared by all bodies rather than any two, and removing a withdrawn count of text-layer mismatches and a page count with no deposited record; stats.py recognises the compilation by its title as well as its abbreviation when counting guideline annotations that name it (railway: 29 rows, previously 8); the tests follow manuscript version 0.34. Version 1.9 was published with its version field blank and the description of version 1.8. No cluster, direction or member count changes. Changes from version 1.7 to 1.8: the operating manual moves to version 1.7 (file operating_manual_v1.7.md), stating that the earlier compilation edition may also be read for the values of clusters without an attribution and that the second extraction route uses another library on the same files; the header of render_check.py gives ten, not eleven, rendered pages with at least ten rows; the tests follow manuscript version 0.33. No cluster, direction or member count changes. Changes from version 1.6 to 1.7: the operating manual moves to version 1.6 (file operating_manual_v1.6.md), correcting the rendering-check count to 165 rows (172 values or text fields) and recording its changes; the README states that the earlier compilation edition is also used for the values of clusters without an attribution; the tests follow manuscript version 0.32. No cluster, direction or member count changes. Changes from version 1.5 to 1.6: the comparison of the 34 clusters with the earlier compilation edition distinguishes editions listing every body of a cluster from those listing only some; the statistics and robustness files carry updated version strings; the operating manual states which of its recommended checks the reference corpus did not meet and that the reading sample was withdrawn; the text-layer parses used as the second extraction route, and the program that made them, are added under records/second_route; the tests follow manuscript version 0.31. No cluster, direction or member count changes. Changes from version 1.4 to 1.5: the statistics file adds the number of the 34 clusters whose value an earlier edition lists under a body outside the cluster; the tests follow manuscript version 0.30; the figure program marks the guideline annotations in Figure 1 as mostly not carried into the compilation; the operating manual header no longer names a manuscript version; the README describes which process records the paper cites. No cluster, direction or member count changes. Changes from version 1.3 to 1.4: the earlier (May 2024) compilation edition is now also compared for the 34 clusters with neither an established direction nor a candidate, in a new output file, and the candidate comparison adds a lookup for names without the prefix meaning product; the statistics file adds the corresponding counts and a note on one key; the tests follow manuscript version 0.29; the figure program relabels the guideline annotation in Figure 1; the operating manual header no longer names a manuscript version; a code comment on the frozen rules is corrected. No cluster, direction or member count changes. Changes from version 1.2 to 1.3: the statistics file reports differing values by material class separately for the two populations (values held by different bodies; several values within one body), no longer adding them; the tests follow manuscript version 0.28, including a count of the established clusters whose origin row names a source outside the material; the counting gate also runs in the review-package layout; Figure 3 marks the number of clusters entering the established category; the README notes that key names in the statistics file are internal labels. No cluster, direction or member count changes. Changes from version 1.1 to 1.2: the explanation strings of the cluster output now state the single self-estimate rule as applied (other members may carry annotations that name no body in the cluster) and count direct namings separately; the description of one direction-rule reading no longer states where candidate clusters move, which no output records; the figure program anchors the Figure 3 arrow to the category the changing clusters start from and relabels the data source field in Figure 1; tests follow manuscript version 0.27; export_r9_lists.py and sample_adjudication.py are added under records. No cluster, direction or member count changes. Changes from version 1.0 to 1.1: the annotation census locates each string on the row that carries it (nine, not ten, contexts establish directions); source level L4 is relabelled to include own-document rows; freeway coverage is stored to six decimals; the withdrawn reading sample and the process records are added; operating manual v1.5. The member table lists every table a member's rows sit in, and the footnote check reads the guideline's text layer. No cluster, direction or member count changes.