1% ============================================================================2% Part III — Data and Systematic Review Protocol3% Written by Agent L (Part III instance), 2026-11-03. Every corpus number is a macro from tables/numbers.tex4% (generated by tables/make_numbers.py from data/metadata/prisma_counts.json and linkage_report.json).5% Sources of truth: RESEARCH_PLAN.md §2–§3, protocol/PROTOCOL.md, protocol/amendments.csv,6% reports/PHASE0_FEASIBILITY.md, data/raw/README.md, ingestion/*.py, systematic_review/define_corpus.py.7% ============================================================================89\section{Information sources and conditions of access}10\label{sec:data-sources}1112A field-level mapping study stands or falls with the bibliographic databases on which it is built. The choice13of sources determines which works can be observed at all, how reliably they can be dated and typed, whether14citation links are available, and under what licence derived data may be redistributed. This part documents15these choices for WP44 and reports, stage by stage, how the mapping corpus was assembled from the raw16sources. The exposition follows the order of the pipeline (identification, linkage, deduplication,17eligibility, scope, dating) and closes with the quality indicators and the research-integrity procedures that18govern every later analysis. The reporting structure is modelled on the PRISMA 2020 items that apply to a19mapping study \citep{page2021prisma,page2021prismaee}; items concerned with risk of bias and effect measures20have no counterpart in a bibliometric design and are replaced, as PRISMA-ScR does for scoping reviews21\citep{tricco2018prismascr}, by indicators of metadata quality and coverage (Section~\ref{sec:data-quality}).22The search-reporting recommendations of PRISMA-S \citep{rethlefsen2021prismas} are followed to the extent23that a database-scope strategy without keyword queries permits.2425\subsection{Why a multi-source design}2627No single database indexes computer science completely, and the gaps are not random. The venue-centred28coverage of DBLP, the publisher-centred coverage of Crossref, the preprint-centred coverage of arXiv and the29aggregating coverage of OpenAlex each miss a distinct part of the literature and type the part they do cover30in distinct ways. Comparative studies of bibliographic data sources have documented large differences in31coverage between Web of Science, Scopus, Dimensions, Crossref and Microsoft Academic32\citep{visser2021comparison,martinmartin2021comparison}, and the journal-level coverage of the two commercial33indexes is known to under-represent conference proceedings, which carry a large share of computer-science34output \citep{mongeon2016coverage}. For the open successor of Microsoft Academic, OpenAlex, the first35independent assessments report strong coverage of works with DOIs together with weaker and uneven metadata on36affiliations, document types and references \citep{alperin2024openalex,culbert2025openalex,37scheidsteger2023openalex,delgadoquiros2024completeness}. These findings motivate the design adopted here: one38source supplies the backbone (OpenAlex), a second supplies the definition of what counts as computer science39(DBLP), a third supplies the preprint layer and its dates (arXiv), and a fourth checks the coverage of a40subfield with a dense, curated venue structure (the ACL Anthology). Crossref is used for verification only.4142\subsection{OpenAlex}4344OpenAlex \citep{priem2022openalex} is the backbone of the corpus. It supplies, for every work, the45identifiers (OpenAlex~W identifier, DOI, arXiv identifier where present), the publication date and type, the46title and an inverted-index abstract where available, the venue (source) and its type, the authorships with47institutions and countries, the topic assignment at three levels (field, subfield, topic) with a primary48topic, the list of referenced works (the citation graph), the open-access status and licence, and the49field-weighted citation impact together with yearly citation counts. OpenAlex is released under the CC050public-domain dedication, which permits redistribution of record-level derived data.5152Two access routes exist and both were used, for different purposes. The public snapshot is a versioned bulk53release in gzipped JSON Lines, published in a public S3 bucket; the release used here is dated542026-10-14 and comprises 1{,}898 part files for the \emph{works} entity, 675.3~GB compressed, with a manifest55listing the record count of each file. The snapshot was streamed from the primary compute node, each file's56content length checked against the manifest and a SHA-256 of the compressed bytes recorded in a SQLite ledger57before parsing; every one of the 1{,}898 files parsed without error and the sum of the per-file record counts58equals the manifest total of \LinkSlimTotal{} works. Two outputs were written during this single pass. The59first is the full JSON record of every work carrying \emph{any} topic in OpenAlex field~17 (Computer Science)60with a publication year between 2018 and 2026, stored as Zstandard-compressed JSON Lines and partitioned by61the snapshot's own date partitions (\LinkOaCsRecords{} records, about 20~GB). The second is a \emph{slim}62table of every work in the snapshot (identifier, DOI, year, type, primary topic, subfield, field and domain,63citation count and a few flags), kept in Parquet so that works cited by computer-science papers but published64in other fields can be resolved for the interdisciplinary analyses of Part~VIII (decision D2 of65\texttt{RESEARCH\_PLAN.md}). Records flagged by OpenAlex as XPAC expansion records are retained at this stage66with their flag so that the exclusion decision can be taken, and counted, later (Section~\ref{sec:data-eligibility}).67The smaller entity snapshots (topics, subfields, fields, domains, sources, institutions, publishers, funders,68concepts) were downloaded the same day with a per-file SHA-256 ledger; author entities were not downloaded69because the authorship data embedded in each work record suffice for the analyses planned.7071The second route is the REST API. Without an API key the public endpoint operates under a shared daily budget72of USD~0.10, list-and-filter calls are priced at USD~0.0001 each, and cursor paging is unavailable, so that a73single filtered query cannot return more than 10{,}000 results through basic paging. Single-record lookups by74identifier are free. The API was therefore used for two narrowly defined tasks. First, an incremental harvest75of works with any computer-science topic and a publication date between 2026-10-01 and 2026-11-02, intended76to capture records added after the snapshot was cut; the harvest is sliced by publication day and, where a day77exceeds the paging limit, split into records with a computer-science primary field and records without one.78Raw JSON pages are stored verbatim under a dated directory with a ledger of counts and cost, and the job79resumes on the following day when the budget is exhausted. Second, free single-record lookups by DOI serve as80one of the verification routes for the bibliography of this manuscript (Section~\ref{sec:data-integrity}).81A free API key, which would raise the daily budget, requires the principal investigator to act on an e-mail82and was deliberately not obtained without instruction; the mandate forbids assuming credentials that have not83been tested.8485\subsection{DBLP}8687The DBLP computer science bibliography \citep{ley2002dblp,ley2009dblp} is a curated, venue-centred index that88has for three decades defined, in practice, the set of journals and conference series regarded as computer89science. It is used here in that capacity: a work indexed by DBLP is in computer science by construction, and90DBLP venue streams feed the taxonomy of Part~IV. The complete XML dump (\texttt{dblp.xml.gz}, dated 2026-11-029122:39~UTC, 1.12~GB) was downloaded together with its DTD and MD5 checksum, the checksum verified, and the file92parsed with a validating XML parser into one row per publication record (articles, conference papers,93proceedings volumes, books, book chapters, doctoral and master's theses, data records; the \texttt{www}94person records are skipped). The whole archive was kept so that per-year counts are available for the95historical baseline and for context. The DBLP dump is released under the Open Data Commons Attribution licence96(ODC-BY~1.0). DBLP contributes, in addition to the scope definition, author person identifiers, venue streams,97and DOI links for most records; the \texttt{ee} fields were also scanned for arXiv URLs, with a result reported98in Section~\ref{sec:data-linkage}.99100\subsection{arXiv}101102The arXiv e-print repository \citep{ginsparg2011arxiv} is the main preprint channel of computer science and,103for the period under study, the place where a large share of the most cited work first appeared. Its104metadata are harvested through the OAI-PMH interface (\texttt{set=cs}, \texttt{metadataPrefix=arXiv}) in105monthly datestamp windows from January 2018 to 2 November 2026. The harvester waits at least 3.5~s between106requests and honours \texttt{Retry-After} headers; every raw XML page is kept verbatim as the immutable107snapshot and parsed afterwards into a single Parquet table. Two properties of OAI-PMH matter for what follows.108The \texttt{from}/\texttt{until} parameters filter on the record's \emph{datestamp}, the time of its last109update, so that a monthly window contains every record updated in that month, not every record first110submitted then; duplicates across windows are resolved at parse time with the latest datestamp winning. The111\texttt{created} element gives the first submission date, which is the date used throughout this study for112preprints. The harvest returned \LinkArxivCsRecords{} distinct records, of which113\LinkArxivCreatedTwoZeroOneEightTwoZeroTwoSix{} were first submitted in 2018--2026; the remainder are older114e-prints updated during the window. The arXiv metadata are released for reuse under the repository's metadata115terms; e-prints themselves are governed by per-paper licences and are downloaded, for the high-priority set116only, within those licences and at the repository's rate limits. The use of arXiv as a research dataset, and117the caveats attached to its category labels and versioning, are discussed by \citet{clement2019arxiv}.118119\subsection{ACL Anthology}120121The ACL Anthology is the curated archive of the computational-linguistics community and has itself served as a122bibliometric corpus \citep{radev2013acl}. Its BibTeX export with abstracts (\texttt{anthology+abstracts.bib.gz},12346.7~MB, SHA-256 recorded) was downloaded on 2026-11-02 and parsed into a table of 138{,}796 entries, of which124\LinkAclRecords{} are conference papers or journal articles and \LinkAclTwoZeroOneEightTwoZeroTwoSix{} fall in1252018--2026. The Anthology plays a diagnostic role: because its venue structure is complete and hand-maintained,126the share of its entries that the backbone can be matched to is a direct measure of OpenAlex coverage for a127subfield that publishes overwhelmingly in conference proceedings. It also supplies abstracts for those entries128in which OpenAlex has none. The export is licensed CC~BY~4.0.129130\subsection{Crossref, Semantic Scholar and other sources}131132Crossref \citep{hendricks2020crossref} is the registration agency behind most DOIs in the corpus and offers133open metadata through a REST API with a polite pool. It is not used to build the corpus, whose backbone134already carries Crossref-derived metadata through OpenAlex, but for three verification purposes: resolution of135every DOI cited in this manuscript, a cross-check of publisher-assigned document types on samples, and a136cross-check of citation counts (\texttt{is-referenced-by-count}) on a sample of highly cited works in137Part~IX.138139The Semantic Scholar Academic Graph \citep{kinney2023s2} was listed in the mandate among the priority sources140because of two signals it alone provides, the influential-citation classification and citation intent. At the141time of the feasibility audit, anonymous requests to its API returned HTTP~429 with an instruction to apply for142a key, and the bulk-dataset endpoint requires a key outright. In keeping with the rule that inaccessible data143be recorded rather than circumvented, Semantic Scholar is recorded as unavailable (decision D3) and no analysis144in this manuscript uses influential-citation or citation-intent signals; where Part~VIII discusses the145strength of citation links, it does so with OpenAlex reference lists and the knowledge-graph extraction of146Part~IV, whose limits are stated there. The Microsoft Academic Graph \citep{wang2020mag}, from which OpenAlex147inherited much of its historical coverage, was discontinued at the end of 2021 and is not an independent source148for the period studied. Kaggle credentials for the arXiv snapshot distributed there were absent and were not149needed, OAI-PMH being sufficient. Benchmark repositories and code-hosting platforms enter only in the150reproducibility and benchmark analyses scoped in Part~IV and are not part of the bibliographic corpus.151152Table~\ref{tab:data-sources} summarises the sources, the access route, the licence and the role of each.153154\begin{table}[htbp]155\centering\footnotesize156\caption{Information sources of the WP44 corpus (all accessed 2026-11-02/03).}157\label{tab:data-sources}158\begin{tabularx}{\textwidth}{>{\raggedright\arraybackslash}p{2.5cm}>{\raggedright\arraybackslash}X>{\raggedright\arraybackslash}p{2.1cm}>{\raggedright\arraybackslash}X}159\toprule160Source & Access and snapshot & Licence & Role \\161\midrule162OpenAlex works snapshot & public S3 bucket, release 2026-10-14, 1{,}898 JSONL files, \LinkSlimTotal{} works & CC0 & backbone: identifiers, dates, types, topics, authorships, references, abstracts, OA status \\163OpenAlex REST API & no key; USD~0.10 per day shared budget; increment 2026-10-01 to 2026-11-02; single lookups & CC0 & post-snapshot increment; reference verification \\164DBLP & XML dump 2026-11-02 22:39~UTC, MD5 verified & ODC-BY~1.0 & computer-science scope; venue streams; author identifiers \\165arXiv & OAI-PMH set \texttt{cs}, 2018-01 to 2026-11, raw pages kept & arXiv metadata terms & preprint layer; first-submission dates; categories \\166ACL Anthology & BibTeX export with abstracts, 2026-11-02 & CC~BY~4.0 & coverage check for NLP venues; abstracts \\167Crossref & REST API, polite pool & open metadata & verification of DOIs, types and citation counts \\168Semantic Scholar & API key required; anonymous calls refused & --- & \emph{unavailable, recorded} \\169\bottomrule170\end{tabularx}171\end{table}172173\section{Identification: what each source contributes}174\label{sec:data-identification}175176There is no keyword search in this study. A mapping of an entire discipline cannot be defined by a query177string without importing the vocabulary of whoever wrote it; it is defined instead by database scope filters178that are themselves documented and reproducible, which is the strategy recommended for mapping studies in179software engineering and adopted here for computer science as a whole180\citep{petersen2008mapping,petersen2015mapping}. The identification stage (S0 in the notation of the research181plan) therefore consists of four scope filters, one per source, applied to immutable snapshots whose dates and182hashes are recorded in \texttt{data/raw/README.md}.183184\subsection{Scope filters}185186The OpenAlex filter retains every work with at least one topic, primary or secondary, in field~17 and a187publication year in $[2018, 2026]$. The choice of \emph{any} topic rather than the primary topic is deliberate188and conservative at this stage: a work whose primary topic lies in, say, statistics or electrical engineering189but which carries a secondary computer-science topic is a candidate, and the decision whether it belongs to190the core or to the penumbra is deferred to the eligibility stage, where it is made explicit and counted191(Section~\ref{sec:data-core}). The filter yields192\PrismaIdentifiedOpenalexCsTopicTwoZeroOneEightTwoZeroTwoSixPassOne{} records in the first pass over the193snapshot. The DBLP filter retains the publication records of type article, inproceedings, incollection, book,194phdthesis and data with a year in $[2018, 2026]$: \PrismaIdentifiedDblpTwoZeroOneEightTwoZeroTwoSix{} records.195The arXiv filter retains the records of set \texttt{cs} whose first submission falls in the window:196\PrismaIdentifiedArxivCsTwoZeroOneEightTwoZeroTwoSix{} records. The ACL filter retains the conference papers197and journal articles of the Anthology with a year in the window:198\PrismaIdentifiedAclTwoZeroOneEightTwoZeroTwoSix{} records. The four counts overlap heavily, since most DBLP199and ACL records and a majority of arXiv records also exist in OpenAlex; the overlap is resolved by record200linkage, not by summation, and no total of identified records is reported as if the sources were disjoint.201202\subsection{The OpenAlex topic classifier as a scope instrument}203204Because the OpenAlex filter relies on an automated topic classifier, its properties bear on everything205downstream. OpenAlex assigns to each work up to three topics from a fixed list of 4{,}516, organised in 252206subfields, 26 fields and four domains; the assignment is made by a model applied to the title, abstract and207venue, and the field label of a work is inherited from its topics rather than assigned directly. Two208consequences follow. A work can be labelled computer science because one of its topics belongs to field~17209even when its content is only marginally computational, and conversely a work in a computer-science venue can210receive no computer-science topic when its abstract is missing or its vocabulary is domain-specific. The first211error inflates the candidate set with what will be called the penumbra; the second removes genuine212computer-science works from the first-pass candidate set. DBLP is used to correct the second error, which is213why the linkage stage has two passes (Section~\ref{sec:data-linkage}). The literature on OpenAlex has so far214examined its coverage and metadata completeness rather than the accuracy of its topic labels215\citep{alperin2024openalex,delgadoquiros2024completeness}; the present study treats the labels as a scope216instrument whose errors are bounded by the DBLP and arXiv memberships and quantified in217Section~\ref{sec:data-quality}, not as ground truth.218219\subsection{Verification of the scale against the mandate}220221The mandate set planning targets of more than one million candidate records and more than one hundred222thousand deeply classified records, subject to availability. The feasibility audit established, before any223ingestion, that the OpenAlex API alone reported more than six million works with a computer-science primary224field in 2018--2026 and that the DBLP dump held more than four million records in the same years; the targets225were therefore attainable without recourse to any restricted source. The audit also established what was226\emph{not} attainable: a Semantic Scholar key, an OpenAlex key, Kaggle credentials and the AWS command-line227tool were all absent, and the ingestion was designed around their absence (HTTPS streaming of the public228bucket, basic paging of the public API). The record of this audit is \texttt{reports/PHASE0\_FEASIBILITY.md}.229230\section{Record linkage across sources}231\label{sec:data-linkage}232233The four candidate sets describe overlapping populations with different identifiers. Record linkage (stage S1)234establishes, for each DBLP, arXiv and ACL record, the OpenAlex work it describes, so that one node per work235can be formed with the union of the metadata. The methods are deterministic and ordered by reliability; the236probabilistic framework of \citet{fellegi1969linkage} is not needed because the available keys (DOI, arXiv237identifier) are close to exact, and the residual title-based step is restricted to blocks where it is238unambiguous. The implementation follows the standard decomposition of data matching into key normalisation,239blocking, comparison and classification \citep{christen2012matching}; Part~IV gives the formal statement of the240keys and of the precision-estimation design, and this section reports what was done and what it produced.241242\subsection{Keys and matching order}243244All keys are normalised before comparison. DOIs are lower-cased. arXiv identifiers are lower-cased and245stripped of their version suffix (\texttt{v2}, \texttt{v3}, \ldots), so that all versions of an e-print map to246one identifier. Titles are reduced to a normalised string by lower-casing and deleting every character that is247not a letter or digit; the normalised title is used only when it has at least 25 characters, is unique within248its source, and is unique within OpenAlex, which is the blocking rule that keeps the title step from249producing spurious links between generic titles.250251For DBLP, the order is: (i) DOI equality, (ii) arXiv identifier equality, restricted to records not yet linked252on either side, (iii) normalised title equality with a year difference of at most one, restricted to records253not yet linked on either side. The DOI step linked \LinkLinkDblpOaDoi{} records and the title step254\LinkLinkDblpOaTitleYear{}; the arXiv step linked none, because the extraction of arXiv URLs from the DBLP255\texttt{ee} fields returned no candidate (\LinkDblpWithArxiv{} records with an extracted identifier). The256\texttt{ee} fields of this dump evidently do not carry \texttt{arxiv.org/abs} URLs in the form the extractor257expects; whether the CoRR entries are reached by the DOI step instead depends on DBLP's current conventions for258the \texttt{ee} field, which the audit did not characterise, and the question is left to the linkage-precision259review. The step is kept in the pipeline because other dumps may differ. In total \LinkDblpLinkedDistinct{} of the \LinkDblpTwoZeroOneEightTwoZeroTwoSix{} DBLP260records in the window were linked to an OpenAlex work after the first pass, leaving \LinkDblpUnlinked{}261unlinked. The composition of the unlinked set is informative: \LinkDblpUnlinkedByTypeArticle{} journal262articles, \LinkDblpUnlinkedByTypeInproceedings{} conference papers, \LinkDblpUnlinkedByTypePhdthesis{}263doctoral theses, \LinkDblpUnlinkedByTypeData{} data records, \LinkDblpUnlinkedByTypeIncollection{} chapters264and \LinkDblpUnlinkedByTypeBook{} books. Of these, \LinkDblpUnlinkedWithDoiInSlim{} carry a DOI that265\emph{does} exist in the slim table of all OpenAlex works, which means the work is in OpenAlex but was not266caught by the computer-science topic filter; this is the second error of the topic classifier described above267and the reason for the second pass.268269For arXiv, the order is: (i) arXiv identifier equality against the identifier OpenAlex records for the work,270(ii) DOI equality for e-prints that carry a journal DOI, (iii) normalised title equality with a year271difference of at most two, the wider tolerance reflecting the typical lag between a preprint and its published272version \citep{lin2020preprints}. The steps linked \LinkLinkArxivOaArxivId{}, \LinkLinkArxivOaDoi{} and273\LinkLinkArxivOaTitleYear{} records respectively, for \LinkArxivLinkedDistinct{} distinct e-prints linked.274For the ACL Anthology, the order is: (i) DOI equality, (ii) equality between the Anthology identifier and the275identifier embedded in an \texttt{aclanthology.org} landing-page URL among the work's OpenAlex locations,276(iii) normalised title equality with a year difference of at most one. The steps linked277\LinkLinkAclOaDoi{} records by DOI and \LinkLinkAclOaTitleYear{} by title, with the URL step contributing the278remainder, for \LinkAclLinkedDistinctTwoZeroOneEightTwoZeroTwoSix{} distinct 2018--2026 entries linked out of279\LinkAclTwoZeroOneEightTwoZeroTwoSix{}; the residual is examined in Section~\ref{sec:data-quality} as a280measure of OpenAlex coverage of conference-heavy subfields.281282\subsection{The second snapshot pass}283284A DBLP record whose DOI resolves to an OpenAlex work outside the topic filter is, by the scope rule, a285computer-science work that the first pass did not load. The same situation arises for arXiv e-prints whose286journal DOI exists in OpenAlex without a computer-science topic. The linkage script therefore writes the list287of OpenAlex identifiers that are needed but not loaded: \LinkNeededOaIdsSecondPassDblpDoi{} identifiers from288DBLP DOIs and \LinkNeededOaIdsSecondPassArxivDoi{} from arXiv DOIs. A second pass over the works snapshot,289restricted to this identifier list, retrieves the full records of these works; they are normalised into the290same Parquet tables with a \texttt{\_p2} suffix, the DuckDB views are rebuilt over the union of both passes,291and linkage is re-run so that the DBLP-linked set gains the works that the topic classifier had missed. The292second pass is what makes the DBLP membership rule of Section~\ref{sec:data-core} operational: without it,293``DBLP-indexed'' would silently mean ``DBLP-indexed and labelled computer science by OpenAlex'', and the DBLP294correction would be void precisely where it matters. DBLP records with no OpenAlex counterpart after both295passes (\PrismaLinkedDblpOnlyNoOpenalexRecord{} records before the second pass) are kept in a separate table296with their year and type and are reported as such; they enter the volume counts of Part~IX only in the297sensitivity analysis that adds DBLP-only records, because they lack the abstracts, references and affiliations298on which every other indicator depends. Most of them are journal articles and conference papers with a DOI299that OpenAlex does not hold at all, together with doctoral theses, which DBLP indexes from university300repositories that OpenAlex covers unevenly.301302\subsection{Precision of the linkage}303304Exact-key links are correct whenever the key is correct in both sources; the known failure modes are DOIs305reassigned by publishers between versions, DOIs registered for a proceedings volume rather than a paper, and306e-prints whose OpenAlex record points to a different version. Title-based links can fail on near-identical307titles of distinct works (extended abstracts and full papers, workshop and conference versions). The protocol308pre-specifies the estimation of linkage precision on a sample of 500 DBLP--OpenAlex pairs stratified by method,309with automatic agreement checks (normalised-title similarity, year difference, first-author surname) and a310200-pair subsample reserved for the principal investigator (checkpoint H-1). The sampling script draws 250 DOI311links, 200 title links and 50 arXiv links with a fixed seed; since the arXiv step produced no links in this312dump, the realised sample has 450 pairs, and the precision estimate for the arXiv step is undefined rather313than reported as perfect. At the time of writing the sampling script exists but the sample has not been drawn and the314human checkpoint is unsigned; the automatic agreement rates and the human verdicts will be reported in Part~IX315and Appendix~B, and every conclusion that depends on the title-based links is labelled as pending human review316until H-1 is signed.317318\section{Deduplication of preprint and published versions}319\label{sec:data-dedup}320321OpenAlex creates distinct work records for an arXiv e-print and for the journal or conference version of the322same paper whenever the two carry different DOIs, which is the normal case since arXiv registers DataCite DOIs323for its e-prints. In a corpus where preprinting is widespread this is the dominant form of duplication, and it324biases every count in the direction of fields that preprint most: a paper in machine learning that appears on325arXiv and then at a conference would be counted twice, a paper in a theory journal once. OpenAlex itself links326some preprint--publication pairs, but the links are incomplete, and prior work on discovering such links has327shown that neither DOIs nor titles alone suffice \citep{cabanac2021preprint}. Stage S2 therefore merges328preprint and published versions into one node per work, in the sense required by the mandate (one node per329\emph{work}, with both dates retained).330331\subsection{Pairing rule}332333The rule is conservative by design, favouring a small number of certain merges over a larger number of334probable ones. A \emph{preprint-like} record is one whose OpenAlex type is \texttt{preprint} or whose DOI is335an arXiv DataCite DOI (prefix \texttt{10.48550/arxiv}); a \emph{published-like} record is any other record.336Two records are paired when their normalised titles are identical, both titles have at least 25 characters,337and their publication years differ by at most two. When a published record matches several preprints, the338earliest preprint is kept as its preprint; when a preprint matches several published records, the earliest339published record is kept; all other pairs are discarded, so that the pairing is one-to-one. The pairing is340confined to the loaded OpenAlex records; preprints that are not computer-science candidates cannot be paired,341which is immaterial since they would not enter the corpus. No fuzzy title similarity is used at this stage, so342that a preprint whose title changed at publication is not merged; the consequence, a residual double count, is343discussed below.344345\subsection{What is kept from each member of a pair}346347The published record is the surviving node. The preprint record receives the flag \texttt{duplicate\_of} with348the identifier of the published work and is excluded with reason code E5 (Section~\ref{sec:data-eligibility}).349The published node receives \texttt{preprint\_work\_id} and, through the arXiv linkage, the first-submission350date of the e-print, so that for every work the \emph{first date} is defined as the earliest of its OpenAlex351publication date, the publication date of its preprint record and the arXiv \texttt{created} date. This first352date is the date variable of all monthly series in Part~IX; the publication date of the version of record is353kept alongside it. The distinction matters for structural-change analysis: dating a paper by its journal issue354would displace much of the research activity of 2023 into 2024 and 2025 and would blur any monthly series355around the November 2022 boundary, whereas the first date places the work at the moment it entered the public356record. Citation counts are those of the surviving published node; OpenAlex attributes to the preprint record357the citations that cite the preprint DOI, and these are not added to the published node, a known358under-count for works whose preprint remained the cited version. Part~IV states the resulting measurement359assumption and Part~IX reports, for the high-priority set, the share of citations that accrue to preprint360records.361362\subsection{Outcome and residual duplication}363364The pairing identified \PrismaExcludedEFiveDuplicate{} preprint records as duplicates of a published work,365of which \PrismaExcludedSequentialEFiveDuplicate{} remained after the earlier exclusions had been applied.366Set against the \PrismaEligibleByTypePreprint{} preprints that remain in the eligible corpus as works with no367published counterpart, the figure shows that most e-prints in the candidate set were not matched to a368published version within OpenAlex. Three explanations coexist and cannot be separated without manual review:369many e-prints are genuinely unpublished, consistent with the publication rates reported for computer-science370preprints \citep{lin2020preprints}; some published versions carry a modified title; and some published371versions are not in OpenAlex or lack a computer-science topic. The first explanation is benign; the second and372third leave a residual double count whose size is bounded in Part~IX by re-running the pairing with a title373similarity threshold rather than exact equality and reporting the difference as a sensitivity, not as a374correction. Because arXiv-linked works carry their e-print identifier, the sensitivity analyses can also375remove every preprint node whose e-print has a journal DOI in the arXiv metadata, whether or not the pairing376found the published record.377378\section{Eligibility criteria and exclusions}379\label{sec:data-eligibility}380381Stage S3 applies the eligibility criteria of the preregistered protocol (\texttt{protocol/PROTOCOL.md}, P2) to382every loaded record and records, for each, a set of Boolean exclusion flags with reason codes. The flags are383computed independently so that each code can be counted on its own and also sequentially, in the order384E1--E7, so that the PRISMA flow (Figure~\ref{fig:prisma}) sums correctly. Reporting both views is deliberate:385the independent counts show how large each problem is in the raw data, the sequential counts show how many386records each criterion removed given those before it.387388\subsection{Reason codes}389390\textbf{E1, outside window.} The publication date is missing, earlier than 2018-01-01 or later than3912026-11-02. Because the first-pass filter already restricted the publication year, this code catches only392records with a missing date or a date beyond the extraction date. Forward-dated records, that is works whose393publication date lies in the future relative to the snapshot, are a documented feature of OpenAlex, where394publishers register articles for forthcoming issues; the window end removes those dated after the extraction395date, and the remaining 2026 records are discussed in Section~\ref{sec:data-dates}.396\PrismaExcludedEOneOutsideWindow{} records are flagged.397398\textbf{E2, non-research document type.} The protocol lists the document types that constitute research399output: article, conference paper, preprint, book chapter, book, dissertation, report, review and software400paper. The exclusion operates asymmetrically. A record that is \emph{not} DBLP-linked is excluded when its401OpenAlex type is outside the list. A record that \emph{is} DBLP-linked is kept whatever its OpenAlex type,402unless that type belongs to a hard-exclusion list of types that are never research outputs (editorial,403erratum, retraction notice, peer review, paratext, conference abstract, reference entry, dataset, software,404supplementary material, library guide, letter, grant, standard). The asymmetry reflects amendment A-001 and405is explained below. Independently of order, \PrismaExcludedETwoType{} records carry this flag. The largest406contributors are records typed as datasets (\PrismaExcludedETwoByTypeDataset{}), followed by the407residual type \texttt{other} (\PrismaExcludedETwoByTypeOther{}), library guides408(\PrismaExcludedETwoByTypeLibguides{}), software (\PrismaExcludedETwoByTypeSoftware{}), conference409abstracts (\PrismaExcludedETwoByTypeConferenceAbstract{}) and peer reviews410(\PrismaExcludedETwoByTypePeerReview{}); paratext, reference entries, editorials, supplementary materials,411errata, standards, book reviews, retraction notices, data papers and letters make up the remainder. The412prominence of datasets and software among records carrying a computer-science topic is itself a finding about413the backbone: OpenAlex indexes the DataCite registry, where repositories mint DOIs for research objects that414are not publications, and the topic classifier labels them by their descriptions.415416\textbf{E3, XPAC expansion records.} OpenAlex marks as XPAC the records it added from non-Crossref sources in417its coverage expansion; these carry sparser metadata and a higher share of non-research objects. The418protocol excludes them, and \PrismaExcludedEThreeXpac{} records carry the flag independently. Amendment A-001419introduced one exception: an XPAC record that is linked to a DBLP record is a computer-science publication420whatever its provenance in OpenAlex, and it is kept with the flag \texttt{xpac\_kept\_dblp}421(\PrismaExcludedXpacKeptDblp{} records). Sequentially, after the type criterion,422\PrismaExcludedSequentialEThreeXpac{} records are removed under E3, far fewer than the independent count423because most XPAC records are also non-research types.424425\textbf{E4, no title.} Records with a missing or empty title cannot be linked, deduplicated or embedded and are426excluded: \PrismaExcludedEFourNoTitle{} independently, \PrismaExcludedSequentialEFourNoTitle{} sequentially.427428\textbf{E5, duplicate version.} The preprint member of a merged pair (Section~\ref{sec:data-dedup}):429\PrismaExcludedEFiveDuplicate{} independently, \PrismaExcludedSequentialEFiveDuplicate{} sequentially.430431\textbf{E6, retracted.} Works flagged \texttt{is\_retracted} by OpenAlex, which draws on the Retraction Watch432data now distributed through Crossref, are \emph{not} removed from the mapping corpus: they are part of the433published record and of its volume, and their presence is itself an object of study given the record numbers434of retractions reported for 2023 \citep{vannoorden2023retractions}. They are kept with the flag and excluded435from every impact analysis (citation rankings, landmark selection, field-normalised scores).436\PrismaExcludedESixRetracted{} loaded records carry the flag, of which437\PrismaEligibleMembershipRetractedFlagged{} are in the eligible corpus.438439\textbf{E7, generic deposit platform.} Introduced by amendment A-002, this code excludes records whose primary440source is a generic deposit platform (Zenodo, Figshare, the Japanese Institutional Repositories DataBase,441LA~Referencia, Dialnet, the DOAJ-as-source entries and the OSTI OAI feed) unless the record is DBLP- or442arXiv-linked. The number of records removed under this code is reported by the macro443\pending{PrismaExcludedSequentialESevenGenericDeposit} once the corpus has been rebuilt under A-002.444445One inconsistency of numbering must be recorded. In the frozen protocol, code E7 designates the446penumbra of works with a secondary computer-science topic only; the implementation reuses the number E7 for447the generic-deposit exclusion of amendment A-002 and carries the penumbra as a separate flag,448\texttt{cs\_penumbra}. The penumbra is not an exclusion in the same sense as E1--E7: it is a scope boundary,449reported separately and used in sensitivity analyses, and is treated in Section~\ref{sec:data-core}. The450conflict of numbering is flagged here for correction in the amendments log before the protocol is cited in451its final form.452453\subsection{Amendments A-001 and A-002 and their evidence}454455Two eligibility decisions were taken after the protocol freeze, on the evidence of the first run of the456corpus-definition script, and both are logged in \texttt{protocol/amendments.csv} with their date, their457reason and their status (executing agent, pending confirmation by the principal investigator at458checkpoint H-0). They are reported here in full because an eligibility rule changed after preregistration is459exactly the kind of decision a reader must be able to audit \citep{nosek2018preregistration}.460461\emph{A-001, document types.} The protocol's type list was written from the OpenAlex documentation available462at the time; the 2026 snapshot uses the types \texttt{conference-paper} and \texttt{software-paper}, which the463list did not contain. The first run of the eligibility stage excluded under code E2 a very large number of records typed464\texttt{conference-paper}, close to half of them DBLP-linked; the figures of that superseded run are recorded in465the amendment entry and are not repeated here, since the run was not retained. Conference papers are the principal publication form of several areas of466computer science, and their exclusion would have invalidated the composition analyses of Part~IX. The467amendment adds the two types to the list and, more consequentially, makes DBLP linkage decisive: a468DBLP-linked record is a computer-science research output whatever OpenAlex calls it, unless its type is one469that can never be a research output. The same amendment keeps DBLP-linked XPAC records. The corrected run470yields the counts reported above, in which \PrismaEligibleByTypeConferencePaper{} eligible works are471conference papers.472473\emph{A-002, generic deposit platforms and date precision.} Inspection of the eligible corpus by source474(\texttt{systematic\_review/inspect\_sources.py}) showed that a single generic repository, Zenodo, contributed475several times more eligible records dated 2026 than dated 2025, that almost none of them carried a reference476list, and that random samples were slide decks, datasets, software releases and project reports typed as477articles because the depositor had chosen that type; the counts are recorded in the amendment entry. Left in the478corpus, such records would have produced a spurious surge of 2026 output concentrated in a repository that479does not publish research articles, and would have done so in the months where right truncation already480makes the series fragile. The amendment excludes records whose primary source is a generic deposit platform481unless DBLP or arXiv vouch for them, and introduces the \texttt{date\_precision} flag discussed in482Section~\ref{sec:data-dates}. The rule is a source-level rule, not a content judgement: a genuine research483report deposited only on Zenodo is lost, which is accepted because the alternative, keeping every deposit,484contaminates the primary estimand of RQ1. The sensitivity analyses of Part~IX report the monthly series with485and without the generic-deposit exclusion.486487The amendments illustrate a general point about preregistered bibliometrics. The protocol can freeze the488\emph{rules} but not the data model of a database that changes between the time of writing and the time of489extraction; when the data model changes, the honest course is an amendment with its evidence, not a silent490adaptation.491492\section{The computer-science core, the penumbra and the strict core}493\label{sec:data-core}494495The decision of what counts as computer science is the single most consequential modelling choice in a496discipline-level mapping, and the protocol makes it with a three-part membership rule (decision D4) rather497than with any one source. A record that survives the exclusions E1--E7 belongs to the \emph{core} if at498least one of the following holds:499500\begin{enumerate}501\item it is linked to a DBLP record (venue-based membership);502\item its OpenAlex primary field is Computer Science (classifier-based membership);503\item it carries a secondary computer-science topic \emph{and} is linked to an arXiv e-print whose primary504 category is in the \texttt{cs} archive (author-declared membership).505\end{enumerate}506507A record that survives the exclusions but satisfies none of these, which means it carries a secondary508computer-science topic without venue or author support, belongs to the \emph{penumbra}. The three conditions509correspond to three independent judgements about disciplinary membership: the judgement of DBLP's editors510about venues, the judgement of OpenAlex's classifier about content, and the judgement of the authors who chose511a primary arXiv category. Requiring agreement of any one of them with the first-pass topic filter removes the512over-inclusion that a secondary topic alone would produce, while keeping works that any single source would513have missed.514515\subsection{Counts}516517After the sequential exclusions, \PrismaAfterExclusions{} loaded records remain. Of these,518\PrismaCsCoreEligible{} satisfy the membership rule and constitute the eligible mapping corpus, and519\PrismaCsPenumbraEligibleOtherwise{} form the penumbra. The penumbra is thus a substantial fraction of the520records that a naive ``any computer-science topic'' filter would have called computer science; its exclusion521from the primary analyses is the main difference between the present corpus and what an OpenAlex query alone522would return, and it is the reason why no count in this manuscript should be compared directly with the counts523that the OpenAlex interface displays for field~17. Within the eligible corpus, the three membership524conditions overlap: \PrismaEligibleMembershipInDblp{} works are DBLP-linked,525\PrismaEligibleMembershipCsPrimary{} have a computer-science primary field, and526\PrismaEligibleMembershipArxivLinked{} are linked to an arXiv e-print; \PrismaEligibleMembershipInAcl{} are527ACL Anthology papers. The eligible corpus divides by period into \PrismaEligibleByPeriodBaseline{} works whose528first date falls in the baseline window, \PrismaEligibleByPeriodPost{} in the post-boundary window, and a529small residue of \PrismaEligibleByPeriodPre{} whose first date, taken from a preprint or an e-print, precedes5302018 although the version of record falls in the window; the residue is kept in the corpus and assigned to531the ``pre'' period, so that it enters no monthly series. Figure~\ref{fig:prisma} renders these counts as a532PRISMA-style flow, drawn from the same JSON file that generates the macros of this section533\citep[cf.][]{haddaway2022prismaflow}.534535\begin{figure}[htbp]536\centering537\includegraphics[width=0.9\textwidth]{fig_prisma_flow.pdf}538\caption{PRISMA-style flow of record identification, exclusion and the final mapping corpus. Identification539counts are per source and overlap; exclusions are sequential in the order E1--E7; the core and the penumbra540partition the records that survive the exclusions. Source: \texttt{data/metadata/prisma\_counts.json},541rendered by \texttt{figures/make\_figures.py}.}542\label{fig:prisma}543\end{figure}544545\subsection{The strict core}546547The membership rule is a disjunction, and a sceptical reader may object that a disjunction of three imperfect548judgements inherits the errors of each. The protocol anticipates the objection with a \emph{strict core},549defined as the conjunction of the first two conditions: works that are both DBLP-linked and labelled computer550science by the OpenAlex classifier. The strict core contains \PrismaStrictCoreDblpAndCsPrimary{} works. Every551headline result of Part~IX is re-estimated on the strict core, and the manuscript reports a result as robust552to scope only when the sign and the order of magnitude agree between the eligible corpus and the strict553core. The strict core is not a better corpus; it is a narrower one, biased toward established venues and554against preprints, dissertations and the newer conference series that DBLP has not yet adopted, and its role555is to bound the influence of scope decisions, not to replace them.556557\subsection{What the design does not do}558559Three limits deserve statement. First, DBLP membership is venue membership: a paper in a DBLP-indexed journal560of bioinformatics or operations research is in the core even when its content is not computational, and the561taxonomy of Part~IV, not the scope rule, is what distinguishes such works. Second, the rule has no language562component; works in any language are eligible, and the share of non-English records, which is concentrated563in dissertations and in regional journals, is reported among the quality indicators. Third, the rule is564applied to the records loaded by the two snapshot passes; a computer-science work that OpenAlex does not565index at all, or indexes without any computer-science topic and without a DBLP or arXiv DOI, is not observed.566The DBLP-only table and the unmatched ACL entries bound the size of this blind spot for two sources with567complete venue lists, and Section~\ref{sec:data-quality} reports them.568569\section{Dates, date precision, indexing lag and right truncation}570\label{sec:data-dates}571572Monthly series are the primary temporal unit of this study, and the November 2022 boundary falls inside a573year. Everything in the structural-change analysis of Part~IX therefore depends on the quality of dates at574the month level, which is considerably worse than at the year level in every bibliographic database. This575section documents three distinct problems, their diagnosis in the corpus, and the rule adopted for each.576577\subsection{Which date}578579Each eligible work carries up to three dates: the OpenAlex publication date of the surviving record, the580publication date of its merged preprint record where one exists, and the arXiv first-submission date where the581work is linked to an e-print. The \emph{first date} of a work is the earliest of the three582(Section~\ref{sec:data-dedup}), and the \emph{first month} is its calendar month. The first date measures583the entry of the work into the public record, which is the quantity relevant to questions about the pace and584composition of research activity; the publication date of the version of record measures the pace of the585publishing system and is kept for the analyses of venues and of preprint-to-publication lags. The two586coincide for works without a preprint, which are the majority outside the machine-learning areas.587588\subsection{Date precision and the January artefact}589590A large share of publication dates in OpenAlex are year-only dates stored as the first of January. The591phenomenon is visible without any model: the raw monthly series of first appearances shows a spike in every592January that no publishing calendar could explain (Figure~\ref{fig:volume-monthly-raw}), and inspection of the593eligible corpus by year found that between about a fifth and somewhat over a quarter of records carry a date of the form594\emph{YYYY}-01-01, with the proportion varying by source and type; dissertations, book chapters and records595from repositories are the most affected, journal articles with publisher-registered issue dates the least.596Amendment A-002 introduced the flag \texttt{date\_precision}, equal to \texttt{year} when the publication597date is the first of January and \texttt{month} otherwise. The rule is a heuristic with a known cost: a work598genuinely published on 1 January is classed as year-precision, and since the rule is applied to the599publication date rather than to the first date, a year-precision work with an arXiv first date is correctly600dated by its e-print. The counts of eligible works by precision class will be reported by the macros601\pending{PrismaEligibleDatePrecisionMonth} and \pending{PrismaEligibleDatePrecisionYear} once the corpus has602been rebuilt under A-002.603604All monthly series in Part~IX use month-precision records only. Year-precision records enter the annual series,605where their date is correct, and the composition analyses that do not depend on the month. The exclusion of606year-precision records from monthly series is not innocuous: if the share of year-only dates differs between607categories or between periods, the monthly composition is biased relative to the annual one. Part~IX608therefore reports, for each category, the share of year-precision records by year, and treats a divergence609between the monthly and the annual growth estimates as a warning rather than as two independent results.610611\begin{figure}[htbp]612\centering613\includegraphics[width=0.95\textwidth]{fig_volume_monthly_raw.pdf}614\caption{Monthly first appearances in the eligible corpus before the precision filter. The January spikes615reflect year-only publication dates stored as 1 January and motivate the month-precision rule. Source:616\texttt{results/bibliometrics/volume\_monthly.csv}.}617\label{fig:volume-monthly-raw}618\end{figure}619620\subsection{Indexing lag and why \texttt{created\_date} cannot measure it}621622A work is observable only once the database has indexed it, and the delay between publication and indexing623determines how incomplete the most recent months are. The research plan proposed to estimate the delay624profile from the OpenAlex \texttt{created\_date} field, the date on which the record was created in OpenAlex,625and to use the profile to correct the right tail of the monthly series. The diagnostic run of this estimator626(\texttt{results/bibliometrics/indexing\_lag.csv}) refuted the proposal. For works published in 2018, the627median difference between \texttt{created\_date} and the publication date is close to eight years, and the628median falls by almost exactly one month for each later publication month, which is the signature of a629constant creation date: the field records the date on which the record was created in the current OpenAlex630data model, apparently during a re-ingestion in the second half of 2025, not the date on which the work was631first indexed. For the great majority of works published before that re-ingestion, \texttt{created\_date}632carries no information about indexing delay, and for works published after it the field measures the delay633only for the records created since. A lag profile estimated from it would be an artefact of database634maintenance. The field is therefore not used, and the plan's lag-corrected estimates are replaced by the635approach below. The episode is recorded in detail because a reader who reproduces the estimator on a later636snapshot may find the field meaningful again, and because it shows why the mandate's requirement to preserve637immutable raw snapshots is not a formality: the only way to measure indexing lag honestly is to compare two638snapshots of the same database.639640\subsection{Right truncation}641642The last months of the window are incomplete for three distinct reasons that must not be conflated: works643published but not yet indexed (indexing lag), works indexed but not yet linked to a DBLP record because DBLP's644own curation lags the publishers, and works whose citation counts have had little time to accumulate. The645first two shrink the counts of the recent months; the third affects only impact indicators and is treated in646Part~IV through fixed citation windows, which are the standard remedy for the age dependence of citation647counts \citep{wang2013window,waltman2016review}.648649For the counts, the diagnosis uses the sources against one another. The DBLP dump of 2026-11-02 holds, for6502026, a count well below its 2025 count although ten months of the year have elapsed, which shows that DBLP's651own coverage of 2026 is incomplete; and the eligible corpus for 2026 is of the same order as that for 2025652(\PrismaEligibleByYearTwoZeroTwoSix{} against \PrismaEligibleByYearTwoZeroTwoFive{}), which on its face653suggests that OpenAlex is \emph{ahead} rather than behind. The two observations are reconciled by the654composition of the 2026 records: a much lower share with a reference list, a much lower mean number of655references, a higher share of preprints and, before amendment A-002, the Zenodo contribution described656above. In other words, the recent months are not simply truncated; they are differently composed, with the657fast channels (preprint servers, repositories) over-represented relative to the slow ones (journals,658curated indexes). A correction factor estimated from totals would be wrong for every category whose channel659mix differs from the average.660661The rule adopted is conservative. The ``complete'' observation window for monthly inference ends in662June~2026; the months from July to November 2026 are reported descriptively, separately, and never enter a663breakpoint test. Within the complete window, the structural-change models include the last months with664their observed values and the robustness section of Part~IX re-estimates every break with the window ending665in December 2025, so that a result that depends on the 2026 months is identified as such. The API increment666harvested after the snapshot (Section~\ref{sec:data-sources}) adds the works registered between the snapshot667date and the extraction date, which narrows but does not close the gap. Finally, forward-dated records,668works carrying a 2026 publication date for an issue not yet released, are retained when their date precedes669the extraction date because they are, by the definition of the first date, in the public record; the670current build does not count them separately, and a count by month of the records whose publication date lies671after the snapshot date is listed among the quality indicators to be added before Part~IX is finalised.672673\section{Quality indicators and known biases}674\label{sec:data-quality}675676PRISMA 2020 asks for an assessment of the risk of bias in the included studies; a mapping study has no677studies in that sense, and the protocol replaces the item with indicators of the quality of the bibliographic678record itself, reported per year and per category in Appendix~B. This section defines the indicators,679reports the corpus-level values that are available at the time of writing, and sets out the known biases of680the design, each with the direction in which it is expected to push the results of Part~IX.681682\subsection{Metadata completeness}683684Four completeness indicators are computed for every work: presence of an abstract, presence of a non-empty685reference list, presence of at least one affiliation with an institution identifier, and presence of a686country for at least one author. In the eligible corpus, \PrismaEligibleMembershipHasAbstract{} works have687an abstract and \PrismaEligibleMembershipHasReferences{} have at least one resolved reference, out of688\PrismaCsCoreEligible{}. Both shares are lower than those reported for the DOI-bearing journal literature in689comparisons of OpenAlex with the commercial indexes \citep{culbert2025openalex,delgadoquiros2024completeness},690which is expected in a corpus that includes dissertations, book chapters and preprints, and both fall sharply691for the most recent year (Figure~\ref{fig:types-openness}). The consequences are concrete. A work without an692abstract cannot be embedded or classified by its content and is classified from its venue, its arXiv693categories and its OpenAlex topics alone, with a lower confidence score; a work without references does not694contribute out-citations to the knowledge graph, so that the citation flows of Part~VIII are estimated on the695subset with references and are weighted accordingly. The ACL Anthology abstracts fill part of the first gap696for one subfield; no source fills the second.697698\begin{figure}[htbp]699\centering700\includegraphics[width=0.95\textwidth]{fig_types_openness.pdf}701\caption{Document-type composition and metadata completeness of the eligible corpus by publication year702(2026 incomplete). Source: \texttt{results/bibliometrics/volume\_yearly.csv}.}703\label{fig:types-openness}704\end{figure}705706\subsection{Coverage overlap between sources}707708The overlap between sources is the study's main instrument for measuring its own coverage, because each709source has a complete list of what it holds. Three overlaps are reported. The DBLP--OpenAlex overlap, after710both passes, is the share of DBLP records in the window with an OpenAlex counterpart; its complement, the711DBLP-only table, is tabulated by year and type in Appendix~B. The arXiv--OpenAlex overlap is the share of712\texttt{cs} e-prints with an OpenAlex counterpart (\LinkArxivLinkedDistinct{} of713\LinkArxivCreatedTwoZeroOneEightTwoZeroTwoSix{} e-prints first submitted in the window). The ACL--OpenAlex714overlap is the share of Anthology papers linked (\LinkAclLinkedDistinctTwoZeroOneEightTwoZeroTwoSix{} of715\LinkAclTwoZeroOneEightTwoZeroTwoSix{}). An unmatched ACL paper is a documented computer-science publication716that the backbone either does not hold or holds under a title or identifier the linkage could not resolve;717the residual is therefore an upper bound on the sum of backbone non-coverage and linkage failure for one718conference-heavy subfield, and a manual review of a sample of unmatched entries, planned with checkpoint719H-1, separates the two.720721Figure~\ref{fig:volume-yearly} shows the eligible works by publication year and by source membership. The722widening of the arXiv-linked layer and the relative stability of the DBLP-linked layer are the first723descriptive facts of the study, and they are also a warning about right truncation: in the most recent724years, the fastest source dominates.725726\begin{figure}[htbp]727\centering728\includegraphics[width=0.95\textwidth]{fig_volume_yearly.pdf}729\caption{Eligible works per publication year by source membership (DBLP-linked, OpenAlex computer-science730primary field, arXiv-linked, ACL Anthology); memberships overlap. 2026 is incomplete. Source:731\texttt{results/bibliometrics/volume\_yearly.csv}.}732\label{fig:volume-yearly}733\end{figure}734735\subsection{Known biases and their expected direction}736737\emph{DBLP coverage bias.} DBLP indexes established venues and adopts new conference series with a delay;738within the core, the DBLP-linked subset over-represents long-standing journals and conferences and739under-represents workshops, regional venues and the newest series. Because the strict core is defined740through DBLP, every strict-core estimate inherits this bias, which runs against emerging topics and against741the most recent years. Results that are stronger in the strict core than in the eligible corpus are therefore742unlikely to be artefacts of over-inclusion; results that appear only in the eligible corpus may be.743744\emph{OpenAlex topic noise.} The topic classifier mislabels in both directions, and the membership rule745bounds only the over-inclusion, through the penumbra. Under-inclusion is bounded by DBLP and arXiv for the746venues and preprints they cover and is unbounded elsewhere. Within the core, the topic labels are used as747one of several signals in the taxonomy, never alone, and the taxonomy's validation on human-reviewed samples748(Part~IV, checkpoint H-2) is the only direct measure of how much label noise survives.749750\emph{XPAC records.} The expansion records excluded under E3 are not a random sample of the literature: they751are concentrated in repositories, in non-English regions and in recent years. Their exclusion removes some752genuine research, mostly outside the venue system, and biases the corpus toward the formal literature; the753DBLP exception restores the part that the venue system vouches for. The sensitivity analysis that reinstates754XPAC records of research types reports the difference.755756\emph{Generic-deposit platforms.} The exclusion under E7 removes, besides the deposits that motivated it, any757research output whose only home is a generic repository; the effect is largest in 2025--2026 and in the758areas, such as scientific computing and computational biology, where data and software deposits accompany759papers. The direction of the bias is toward under-counting recent output in those areas, which is the760opposite of the bias the rule corrects.761762\emph{Missing abstracts.} Abstract availability depends on publishers' metadata deposits and on OpenAlex's763sources; it is lower for conference papers of some societies and for older records, and it is strongly764correlated with the DOI registrant. Every content-based analysis is conditional on abstract availability, and765the appendix reports the abstract share by category and year so that the reader can judge where the766conditional population departs from the corpus.767768\emph{Forward-dated and year-dated records.} Both displace works along the time axis in a direction that is769not symmetric: year-only dates pull works toward January, forward dates push them toward the future. The770month-precision rule removes the first; the second is bounded by the window end and will be counted771explicitly.772773\emph{Retractions.} Retracted works are in the corpus and out of the impact analyses. The retraction flag774depends on the coverage of the Retraction Watch data through Crossref, which is incomplete for publishers775that do not deposit retraction notices; some retracted works are therefore in the impact analyses776unflagged. The direction of the bias is small but known: retraction is concentrated in a few venues and in777a few years, and the landmark selection of Part~V checks each candidate against the current retraction778status at the time of review.779780\emph{Preprint duplication.} The residual double count after the exact-title pairing inflates the areas that781preprint most, which are also the areas expected to show the strongest growth. The sensitivity with a782similarity threshold and the removal of preprints with a journal DOI bound the effect; a growth result for783machine-learning areas that does not survive these two variants is reported as not robust.784785None of these biases is corrected by reweighting. The study reports them, bounds them where a second source786permits, and re-estimates its results on alternative corpora (strict core; corpus with penumbra; corpus with787XPAC research types; corpus with DBLP-only records; corpus with generic deposits) so that the reader can see788how far each conclusion depends on each scope decision. This is the approach that the science-of-science789literature recommends for databases whose coverage is known to be uneven \citep{fortunato2018science}, and it790is the only one compatible with a mandate that forbids the fabrication of corrections.791792\section{Research-integrity procedures}793\label{sec:data-integrity}794795The mandate for this working paper lists integrity rules that are unusual for a bibliometric study: no796fabricated reference, identifier, count or result; no paper content invented from metadata; no abstract797treated as a full text; no model-generated summary treated as evidence; no temporal association read as798causation; no preprint described as peer-reviewed; inaccessible data recorded explicitly; and human review of799every scientific conclusion. The rules are a response to the way the study is conducted, by a principal800investigator directing autonomous software agents, and this section describes the procedures by which they801are enforced. The procedures are reported here, in the data part, because most of them operate on data and802artefacts rather than on prose.803804\subsection{Preregistration and the amendments log}805806The protocol (\texttt{protocol/PROTOCOL.md}) was frozen on 2026-11-02 before any analysis of the corpus. It807fixes the research questions, eight directional hypotheses H1--H8 with the sign of the expected effect808stated so that null and opposite results must be reported as such, the eligibility criteria and their codes,809the sources and extraction dates, the data items, the quality indicators, the synthesis methods, the810human-review checkpoints and the data-availability commitments. The preregistration is internal, in the811repository rather than in a registry, which limits its evidential value relative to a time-stamped public812record \citep{nosek2018preregistration}; the repository's version history and the dated amendments log are813the available substitute, and the release of Phase~9 will include both. Every departure from the frozen814protocol is an entry in \texttt{protocol/amendments.csv} with an identifier, a date, the phase, the change,815the reason and the approver, which until the principal investigator signs is ``executing agent (pending816PI)''. At the time of writing the log holds five entries: A-000 records that the mandate's acceptance817criteria were completed by the executing agent after the pasted text was truncated; A-001 and A-002 are the818eligibility amendments of Section~\ref{sec:data-eligibility}; A-003 and A-004 fix the budgets and the models819of the LLM-based taxonomy validation and knowledge-graph extraction. Analyses added after the freeze are820labelled exploratory in Part~IX.821822\subsection{Immutable raw snapshots and provenance}823824Nothing in the corpus can be traced to a result unless the raw data from which it was built can be825re-obtained or re-verified. The OpenAlex release is a public, versioned object; the study keeps its manifest,826the per-file content lengths and SHA-256 hashes, and the filtered subset, rather than a copy of the full827bucket (decision D6), so that a reader with the same release and the same scripts obtains byte-identical828inputs, and a reader with a later release can measure the difference. The DBLP dump is kept with its MD5,829the arXiv harvest as the raw OAI-PMH pages, the ACL export with its SHA-256, and the API increment as the raw830JSON pages with a ledger of calls and cost. Each ingestion job writes a ledger with attempts, timings, hashes,831counts and errors, which is the job-level provenance the mandate requires. The processing chain from raw832snapshot to PRISMA counts is a sequence of scripts with measured wall times in833\texttt{reproducibility/RUNBOOK.md}, executed in a locked Python environment whose package versions are834recorded; the chain was executed on a single node, deliberately, because it is network-bound and because a835single-node execution is easier to reproduce than a distributed one.836837\subsection{Numbers that cannot be typed}838839Every count, share and estimate quoted in this manuscript is a LaTeX macro generated by840\texttt{tables/make\_numbers.py} from the result files (\texttt{prisma\_counts.json},841\texttt{linkage\_report.json}, the bibliometric and econometric result tables). The macros cannot be edited842by hand, a macro whose source file is absent is undefined and stops the compilation, and a number the authors843need but the pipeline has not produced appears in the text as a visible pending marker rather than as a844figure. The procedure is the mechanical form of the mandate's requirement that every quantitative claim845trace to a script, an input artefact and a result file. It has a cost in prose, since the text cannot846anticipate a number, and a benefit in auditability, since a reader who regenerates the result files847regenerates the manuscript's numbers.848849\subsection{Verified references}850851The bibliography of this manuscript is built under a rule that no reference may be cited unless it has been852resolved to a bibliographic record by a single programmatic lookup against Crossref, OpenAlex or arXiv, with853the returned title compared with the intended one and the comparison recorded in a verification table854(\texttt{references\_verification*.csv}: key, identifier, service, returned title, match). References that855fail the lookup are removed from the bibliography and kept in the table with a failed status, and a second856agent re-verifies a sample in the audit phase. The rule exists because the model-based agents that draft the857prose are known to produce plausible but non-existent references when unconstrained; it is also why858preprints are cited as preprints, with their arXiv identifier and without a venue, and why the bibliography859distinguishes a technical report from a peer-reviewed article in its entry type.860861\subsection{Agents, artefacts and the register}862863The study is executed by a research-director agent that spawns specialised agents with written briefs;864each brief, the artefacts it produced and the agent that verified them are recorded in865\texttt{agents/REGISTER.md}. The rule that governs the register is that no agent approves its own findings:866an artefact enters the manuscript only after a different agent, the scientific critic or the reproducibility867auditor, has checked it, and the register records the checker. Agents exchange structured artefacts (tables,868JSON, Markdown with declared schema) rather than conversational summaries, so that what one agent passes to869another can be inspected after the fact. The writing agents, including the one that produced this part, work870from the source-of-truth files named in their briefs and are instructed to describe what the code does871rather than what the plan intended; where the two differ, as with the indexing-lag estimator of872Section~\ref{sec:data-dates}, the difference is reported.873874\subsection{Human-review checkpoints}875876Six checkpoints are defined (H-0 to H-5: plan and protocol; linkage sample; taxonomy validation sample;877landmark list and specialist reviews; structural-change results and their interpretation; manuscript878conclusions), each with a named artefact that the principal investigator reviews and signs879(\texttt{reports/HUMAN\_REVIEW\_CHECKPOINTS.md}). Until a checkpoint is signed, every conclusion that880depends on it is labelled ``pending human review'' in the text, with the same visible marker as a missing881number. At the time of writing no checkpoint is signed. The reader should take this literally: the882conclusions of this manuscript are, in its present version, the conclusions of a procedure, reviewed by its883own audit agents and awaiting the judgement of its human author.884885\subsection{What the procedures do not guarantee}886887The procedures guarantee traceability, not truth. A count traced to a script is still wrong if the script is888wrong; a verified reference is still miscited if the prose misreports what it says; a preregistered889hypothesis can still be tested with an inappropriate model. What the procedures make possible is detection:890every number, reference and decision has a place where an auditor can look. The remaining risks are the891ones this part has tried to state plainly: the coverage and labelling errors of the sources, the residual892duplication, the fragility of the recent months, and the dependence of every scope decision on the893judgement, recorded but not infallible, of the agents that took it.894895