1"issue_id","file","line","severity","type","quote","required_fix"2"K-01","manuscript/sections/part02_evolution.tex","62","major","citation_unverified","estimated the long-run growth of science at roughly three per cent per year with a doubling time of about a quarter of a century","Bornmann & Mutz (2015) abstract (Crossref, checked 2026-11-03): three phases, <1 % to mid-18th c., 2-3 % up to the inter-war period, 8-9 % to 2010. The 2-3 % figure is the pre-1940s phase, not the long-run/modern rate. Report 8-9 % (doubling about 9 years) or attribute the 2-3 % explicitly to the earlier phase."3"K-02","manuscript/sections/part02_evolution.tex","151 (also 55)","major","citation_unverified","found that a majority of computer-science preprints in their sample could not be matched to a published version","Abstract not retrievable through Crossref/OpenAlex (checked); the paper is widely quoted as finding about 66 % of arXiv CS preprints published. Verify against the full text and correct or delete the 'majority' claim; l.55 ('a large share ... never matched') depends on it. Part III l.369-370 cites the same paper neutrally."4"K-03","manuscript/sections/part02_evolution.tex","153","major","inconsistency","the seasonal component s_c(t) of equation (eq:p2-its) is modelled separately for the baseline and the research period for that reason","econometrics/structural_change.py `_design` fits one set of eleven month dummies common to both periods; no period-specific seasonality exists in any script or result. State that seasonality is common (or implement interacted dummies and report them)."5"K-04","manuscript/sections/part02_evolution.tex","49","major","inconsistency","The OpenAlex snapshot contains \LinkOaCsRecords{} works classified in computer science between 2018 and 2026","\LinkOaCsRecords = 16,592,966 is the union of pass 1 and pass 2 (oa_keys over the unioned view); the CS-topic filter itself yielded 15,490,159 (App H Table tab:appH-steps, RUNBOOK). About 1.10 M of the quoted works carry no CS topic. Use the pass-1 figure (add a macro) or reword."6"K-05","manuscript/sections/part02_evolution.tex","49","minor","citation_unverified","(a preprint finding; \citealp{culbert2024openalex}.","Unbalanced parenthesis. The work is published (Scientometrics 2025, cited in Part III as culbert2025openalex): cite the journal version, drop 'a preprint finding', and merge the duplicate bib entries (also martinmartin2021coverage/comparison, fortunato2018sos/science, kinney2023s2/s2ag, radev2013acl/aclanthology, openalex2024topics/2025topics; references_all.bib has 20 duplicated DOIs)."7"K-06","manuscript/sections/part02_evolution.tex","251","minor","overclaim","its diffusion among the public was by all accounts faster than that of any previous software product","Superlative with no independent source (the sentence itself says the usual figures are unverifiable). Delete or cite a verifiable source."8"K-07","manuscript/sections/part02_evolution.tex","190","minor","overclaim","became the first systems language to be adopted at scale on the strength of a safety guarantee","Unsupported 'first'; soften ('one of the first', or cite)."9"K-08","manuscript/sections/part02_evolution.tex","192","minor","inconsistency","the Kepler conjecture was formally proved in 2017 after a twelve-year effort","Flyspeck started in 2003 and was completed in August 2014; the Forum of Mathematics Pi paper appeared in 2017. Write 'completed in 2014 (published 2017)'."10"K-09","manuscript/sections/part02_evolution.tex","142","minor","overclaim","A break detected in one series and not in the other is informative about the mechanism, not merely about measurement.","It can equally reflect indexing lag, linkage failure or right truncation (Part III sec:data-dates). Add the competing explanation."11"K-10","manuscript/sections/part02_evolution.tex","336","minor","inconsistency","In a preprint, \citet{ott2022benchmarks} mapped the creation and saturation of benchmarks","Bib entry is Nature Communications 13 (2022), peer-reviewed. Remove 'In a preprint'."12"K-11","manuscript/sections/part02_evolution.tex","324","minor","unsupported_claim","the robustness checks of Appendix D re-estimate the main breaks with classifications that use no keyword signals","taxonomy/classify.py has no option to drop keyword signals and no venue-only classification exists; label as planned or implement."13"K-20","manuscript/sections/part03_data.tex","288","blocking","number_without_artifact","\LinkNeededOaIdsSecondPassDblpDoi{} identifiers from DBLP DOIs and \LinkNeededOaIdsSecondPassArxivDoi{} from arXiv DOIs","Macro undefined (linkage_report.json has no needed_oa_ids_second_pass.arxiv_doi key): the manuscript stops with 'Undefined control sequence' (verified by compiling a copy in /tmp). Emit the macro with value 0 from make_numbers.py or state in prose that no arXiv-DOI identifier was needed. Same macro in Part IV l.235 and App B l.182."14"K-21","manuscript/sections/part03_data.tex","517-531; 539 (caption)","blocking","inconsistency","the core and the penumbra partition the records that survive the exclusions","prisma_counts.json: after_exclusions 10,519,027 but cs_core_eligible 7,110,289 + cs_penumbra 3,402,475 = 10,512,764; 6,263 surviving records are neither (cs_any false, not DBLP-linked: pass-2 works reached through arXiv DOIs). Report the third class in the PRISMA flow and explain it, and reconcile with the membership rule (see K-47)."15"K-22","manuscript/sections/part03_data.tex","191-193","blocking","inconsistency","The filter yields \PrismaIdentifiedOpenalexCsTopicTwoZeroOneEightTwoZeroTwoSixPassOne{} records in the first pass over the snapshot","define_corpus.py counts oa_keys over the union of both passes, so the 'pass1' key equals the total loaded (16,592,966). The first pass retained 15,490,159 (RUNBOOK, App H, slim_cs_any_2018_2026). Record the true pass-1 count in prisma_counts.json and fix the prose here and in App B l.111/195-200."16"K-23","manuscript/sections/part03_data.tex","356-358","blocking","inconsistency","OpenAlex attributes to the preprint record the citations that cite the preprint DOI, and these are not added to the published node, a known under-count","bibliometrics/03_citation_impact.py (CTE CY, UNION ALL on preprint_work_id) adds the merged preprint's counts_by_year to the canonical node; App E l.604-606 documents the merge. Correct the prose (and state the merge in Part IV sec:m-citations)."17"K-24","manuscript/sections/part03_data.tex","664-665","major","inconsistency","re-estimates every break with the window ending in December 2025","run_structural_change.py --sensitivity defaults to 2026-03 and 2026-08 and numbers.tex carries \ItsSensitivityLastCompleteTotalTwoZeroTwoSixZeroThree/Eight; Part IV l.1047 says March and August 2026. Fix."18"K-25","manuscript/sections/part03_data.tex","313-316","major","unsupported_claim","At the time of writing the sampling script exists but the sample has not been drawn","data/metadata/linkage_sample.csv (450 pairs) and linkage_sample_summary.json exist: DOI stratum 235/250 auto_ok (0.94), title-year 200/200 (1.00). Report the automatic rates; note that the 15 DOI-step failures include chapter/book DOI collisions and retitled versions, which contradicts Part IV l.368 ('the only [stratum] where errors are plausible')."19"K-26","manuscript/sections/part03_data.tex","310 (also Part IV l.357)","minor","inconsistency","a 200-pair subsample reserved for the principal investigator","With 450 realised pairs the rule pi_review = (i % 5 < 2) flags 180 pairs. Say 180 (or draw 200)."20"K-27","manuscript/sections/part03_data.tex","443; 600-602","major","number_without_artifact","\pending{PrismaExcludedSequentialESevenGenericDeposit} ... \pending{PrismaEligibleDatePrecisionMonth} and \pending{PrismaEligibleDatePrecisionYear}","All three macros exist in tables/numbers.tex (473,549; 5,703,718; 1,406,571). Replace the pending markers by the macros."21"K-28","manuscript/sections/part03_data.tex","445-451","minor","inconsistency","The conflict of numbering is flagged here for correction in the amendments log","Amendment A-005 (2026-11-03) already resolves it (App B l.526-527, App H Table). Update the paragraph."22"K-29","manuscript/sections/part03_data.tex","816","minor","inconsistency","At the time of writing the log holds five entries","protocol/amendments.csv holds six (A-000 to A-005); App H says 'All six amendments'."23"K-30","manuscript/sections/part03_data.tex","277","minor","inconsistency","with the URL step contributing the remainder","linkage_report.json link_acl_oa has only doi (58,174) and title_year (15,004); the URL step linked no record. State that it contributed none."24"K-31","manuscript/sections/part03_data.tex","336-339","major","inconsistency","Two records are paired when their normalised titles are identical, both titles have at least 25 characters, and their publication years differ by at most two","define_corpus.py also pairs on a shared arXiv identifier (relation 'arxiv_id': \PrismaVersionRelationshipArxivId = 13,175 eligible pairs vs 84,776 by title) and requires the published twin to be in window, non-XPAC, titled and of a kept type (or DBLP-linked). Describe both routes and the twin conditions; same omission in Part IV eq. (eq:duppair)."25"K-32","manuscript/sections/part03_data.tex","251-253","minor","inconsistency","(iii) normalised title equality with a year difference of at most one","link_sources.py l.63 also requires equal first-author surnames when both are present (App B l.540-542 states it). Add the condition; same omission in Part IV l.198-201."26"K-33","manuscript/sections/part03_data.tex","123","minor","number_without_artifact","parsed into a table of 138{,}796 entries","Macro \AclParsedEntries exists; use it (policy of sec:data-integrity)."27"K-34","manuscript/sections/part03_data.tex","390-392","minor","inconsistency","this code catches only records with a missing date or a date beyond the extraction date","Pass-2 records are loaded by identifier without a year restriction, so E1 also removes out-of-window pass-2 works (part of the 59,453). Reword."28"K-35","manuscript/sections/part03_data.tex","596-600","minor","inconsistency","equal to year when the publication date is the first of January and month otherwise","Code sets 'month' when a merged-preprint or arXiv date exists even if the publication date is 1 January (App B l.515-516, App E l.573). Align the definition."29"K-36","manuscript/sections/part03_data.tex","32; 36; 689","minor","citation_unverified","\citep{visser2021comparison,martinmartin2021comparison} ... \citep{alperin2024openalex,culbert2025openalex,...}","Duplicate bibliography entries for the same works under two keys (see K-05). Keep one key per work."30"K-40","manuscript/sections/part04_methodology.tex","235","blocking","number_without_artifact","\LinkNeededOaIdsSecondPassArxivDoi{} from arXiv","Undefined macro; compile failure (see K-20)."31"K-41","manuscript/sections/part04_methodology.tex","766-777","blocking","inconsistency","the two diversity indices are not yet produced by a script; they are defined here because the protocol commits to them, and their values in Part IX are [line break] The yearly Rao--Stirling and Leinster--Cobbold indices ... are computed by \texttt{bibliometrics/04\_diversity\_cagr.py} ... [line break] until the computation is added and audited.","Garbled paragraph: two contradictory drafts spliced together, leaving an unfinished sentence. Rewrite: the script exists (B = 500, seed 44; dissimilarity from centroids when available, else co-classification Jaccard)."32"K-42","manuscript/sections/part04_methodology.tex","46-52 (eq:first-date)","blocking","inconsistency","d(w) = min{ d_pub(w), d_pub(pi(w)), d_arx(w) } ... each missing term is replaced by d_pub(w)","define_corpus.py l.58-60: when d_pub is YYYY-01-01 and a preprint or arXiv date exists, d_pub is dropped and the first date is the least of the other two (App B l.511-514 and App E l.566 state the code rule). The equation is wrong as written; a year-only 2022 record with a 2023 arXiv date is dated 2023 by the code. Fix the equation and state the consequence."33"K-43","manuscript/sections/part04_methodology.tex","76-87 (eq prec)","major","unsupported_claim","prec(w) = year if d_pub(w) falls on 1 January ... the loss is small and, more importantly, it does not depend on the period, so it cannot manufacture a break","Code: 'year' only when no preprint/arXiv date exists. The rescue of 1-January records by arXiv dates grows with arXiv linkage over time and Part III reports year-only shares of 18-28 % varying by source and type, so the exclusion is period-dependent. Correct the definition and replace the 'cannot manufacture a break' assertion by the sensitivity it calls for."34"K-44","manuscript/sections/part04_methodology.tex","250-261 (eq:duppair)","major","inconsistency","(p, b) is a candidate pair iff eta(p) = eta(b), |y(p) - y(b)| <= 2, p in P, b notin P, |title| >= 25","Omits the shared-arXiv-identifier route and the conditions on b (in window, non-XPAC, titled, kept type or DBLP-linked); see K-31. Also the length test is on the normalised title (length(ntitle))."35"K-45","manuscript/sections/part04_methodology.tex","64-70","minor","inconsistency","period(w) = B ... P ...","define_corpus.py also yields 'pre' (first date before 2018; 1,864 eligible works, Part III l.529) and 'future'. Add both and say how they are treated."36"K-46","manuscript/sections/part04_methodology.tex","198-201","minor","inconsistency","restrict each side to normalised titles that are unique within that side, join on equal eta, and keep the pair if the years differ by at most one","Add the first-author surname condition of link_sources.py l.63 (App B has it)."37"K-47","manuscript/sections/part04_methodology.tex","236-240","major","overclaim","it is the union of that set with what DBLP and arXiv call computer science","The third membership clause is cs_any AND arxiv_cs_primary, so an arXiv-declared CS work outside the topic filter is never admitted; the 6,263 orphan records of K-21 are such works. Either reword ('and, within the topic filter, what arXiv calls computer science') or amend the rule."38"K-48","manuscript/sections/part04_methodology.tex","368-369","major","unsupported_claim","the title--year stratum, which is the only one where errors are plausible, is weighted by its small share of the links","Contradicted by linkage_sample_summary.json (DOI stratum auto-precision 0.94; title 1.00) and by the failing DOI pairs (book-level DOI, retitled versions). Rewrite; with weights L_s/L the DOI stratum dominates p-hat."39"K-49","manuscript/sections/part04_methodology.tex","353","minor","inconsistency","a string similarity between the two raw titles","linkage_sample.py compares normalised titles (lower-cased, ASCII-folded, alphanumeric). Say so."40"K-50","manuscript/sections/part04_methodology.tex","1067-1068","major","inconsistency","Donut windows (dropping the months adjacent to the event) are not implemented","run_structural_change.py implements --donut (default 3 months each side) and numbers.tex carries \ItsSeriesTotalItsChatgptDonut* (n = 94, level change 0.049, p = 0.056). Describe and report it."41"K-51","manuscript/sections/part04_methodology.tex","1568-1572","major","number_without_artifact","it is the percentile interval of the non-parametric bootstrap over works, with 1,000 replications and seed 44, computed by the table-making script at reporting time","tables/make_numbers.py contains no bootstrap and no script bootstraps the Gini or the cross-area shares; the only bootstrap is 04_diversity_cagr.py (multinomial over categories, B = 500, l.772). Remove the claim or implement it, and reconcile 500 vs 1,000."42"K-52","manuscript/sections/part04_methodology.tex","588-589","minor","inconsistency","The CAGR is computed at table-making time from the annual counts rather than by a dedicated script","bibliometrics/04_diversity_cagr.py computes cagr_by_category.csv (2018->2022 and 2023->2025), as l.771-776 says."43"K-53","manuscript/sections/part04_methodology.tex","690-695","minor","unsupported_claim","the tables report the index with and without the preprint type","02_collaboration.py writes venue_hhi_yearly.csv over all types only; no preprint-excluded variant exists. Implement or delete."44"K-54","manuscript/sections/part04_methodology.tex","1379-1390","major","inconsistency","With h = 24 the last training label uses data to December 2023 ... the implementation stops at December 2021 so that the training labels do not extend into the test period at h = 24","Contradicts l.1379-1380 (December 2020 for h = 24) and the code (last_train = 2020-12 for h > 12; label closes December 2022). Cutoffs are months with index = 0 mod 3 from January 2018, so the realised training cutoffs run January 2020 to October 2021 (h = 12) or October 2020 (h = 24), not 'December 2019 to December 2021'. App C l.772-776 records the contradiction instead of fixing it."45"K-55","manuscript/sections/part04_methodology.tex","1309","minor","inconsistency","to the last month (October 2026 by default)","emerging_detection.py --last-month defaults to 2026-06."46"K-56","manuscript/sections/part04_methodology.tex","1337-1338","minor","inconsistency","the one-month overlap between the z-score's trailing window and the quarter it is compared with is immaterial","The trailing window ends at T-3 and the quarter is T-2..T: there is no overlap. Delete."47"K-57","manuscript/sections/part04_methodology.tex","1427-1435","major","inconsistency","The script reserves a table for this mapping and it is empty in the current version; the lead-time statistic is therefore \pending{...}","experiments/known_set.py exists and produced results/emerging/known_set_lead_times.csv with a different design: keyword-regex monthly series per concept; detection = trailing-3-month z > 3 and 12-month doubling with >= 30 works; peak = month of maximum 12-month log growth over the whole window. That is the z-score baseline rule applied to hand-written regexes, not the trained detector of H8, and the peak uses full-sample information. Describe it, label it a deviation from the preregistered lead-time definition, and do not present it as the H8 test."48"K-58","manuscript/sections/part04_methodology.tex","1622","minor","inconsistency","Five checkpoints (H-1 to H-5, ...)","Six checkpoints H-0 to H-5 (Part III l.876, reports/HUMAN_REVIEW_CHECKPOINTS.md)."49"K-59","manuscript/sections/part04_methodology.tex","37-38","minor","inconsistency","a creation date d_cr(w) (the date at which the record entered OpenAlex)","Part III l.626-634 shows created_date reflects a 2025 re-ingestion for most records; qualify here and in App E l.253."50"K-60","manuscript/sections/part04_methodology.tex","1181-1183","minor","unsupported_claim","with a trimming of eight or more months per segment this cannot create a break by itself","Asserted without argument; the Bai-Perron input is already seasonally adjusted with baseline month means (l.54 of the driver). Say that instead."51"K-61","manuscript/sections/part04_methodology.tex","662-663; 637-638","minor","inconsistency","the Gini coefficient of c^(36) among the n eligible works of that year ... the set of eligible works sharing year, type and subfield","03_citation_impact.py excludes retracted works (NOT e6_retracted) from both; say 'eligible, non-retracted' (the CitationWindowsYearly N differs from VolumeYearly N by the retracted count)."52"K-62","manuscript/sections/part04_methodology.tex","1131-1139 (and App C l.122-125)","minor","missing_uncertainty","the implementation ranges over the baseline and the post period","The baseline-only placebo (reported: \ItsSeriesTotalPlaceboChatgptBaselineOnly*) excludes the true event from the candidate set, so its value can be 0 (slope: 0.000), unlike the full-range rank whose floor is 1/|T'|. State the convention next to the result."53"K-63","manuscript/sections/part04_methodology.tex","1044-1046","minor","unsupported_claim","from February 2018 (the default start of the driver script, which drops one month and is immaterial)","Give the reason for dropping January 2018 or start at 2018-01; 'immaterial' is asserted, not shown."54"K-70","manuscript/appendices/appA_taxonomy.tex","396-399; 405; 422-446","minor","number_without_artifact","the mapped streams cover 2{,}017{,}910 of 3{,}319{,}884 records (60.8\,\%)","DBLP record counts are data-derived quantities typed from ONTOLOGY.md, not ontology properties; generate macros from dblp_streams_2018_2026.csv or label the table as 'ontology-build statistics (taxonomy/ONTOLOGY.md, generated section)'."55"K-71","manuscript/appendices/appA_taxonomy.tex","675","minor","inconsistency","The ontology was validated, after the classification run, on a stratified sample","taxonomy/validation/ does not exist in the repository mirror and Part IV l.370/App H step 13 mark the validation as not yet run; write 'is validated' / 'will be validated'."56"K-80","manuscript/appendices/appB_search_and_sources.tex","182","blocking","number_without_artifact","\LinkNeededOaIdsSecondPassArxivDoi{} through an arXiv DOI","Undefined macro (see K-20)."57"K-81","manuscript/appendices/appB_search_and_sources.tex","195-200","major","inconsistency","The PRISMA counts of Part III were generated from the first pass (\PrismaIdentifiedOpenalexTotalRecordsLoaded{} records loaded); ... \pending{PrismaIdentifiedOpenalexTotalRecordsLoaded after the second pass}","prisma_counts.json (generated 2026-11-03T23:03) already counts the union of both passes (16,592,966 = 15,490,159 pass 1 + 1,102,807 pass 2); the 'pass 1' macro name is a code labelling error (K-22). Rewrite with the two counts."58"K-82","manuscript/appendices/appB_search_and_sources.tex","111","major","inconsistency","about 20~GB in total for \PrismaIdentifiedOpenalexCsTopicTwoZeroOneEightTwoZeroTwoSixPassOne{} records","The 20 GB CS subset holds 15,490,159 records (App H Table tab:appH-raw); the macro value is the two-pass total."59"K-83","manuscript/appendices/appB_search_and_sources.tex","238-247","major","number_without_artifact","(\pending{IncrementApiCalls}, \pending{IncrementRecordsRetrieved}, \pending{IncrementTruncatedSlices})","Macros exist: 1,002 calls; 193,369 records; 1 truncated slice; 64 slices done; USD 0.100. Report them, say how many of the 66 day-slices (33 days x 2) are done, and quantify the truncated slice (2026-10-16 cs_secondary: count 37,712, 10,000 retrieved). \IncrementNewWorks has no source file yet (increment_report.json absent) and may stay pending."60"K-84","manuscript/appendices/appB_search_and_sources.tex","297; 365; 501","minor","number_without_artifact","\pending{DblpParsedRecords} ... \pending{AclParsedEntries} ... \pending{PrismaExcludedESevenGenericDeposit}","All three macros exist (8,840,866; 138,796; 1,277,584). Replace."61"K-85","manuscript/appendices/appB_search_and_sources.tex","84-86","minor","inconsistency","the aggregate over eight streams about 110~MB/s","App H l.76 and the RUNBOOK say about 108 MB/s; harmonise."62"K-86","manuscript/appendices/appB_search_and_sources.tex","102; 350; 384","minor","citation_unverified","\citealp{openalex2024topics} ... \citep{radev2013aclanthology, acl2024anthology} ... \citep{kinney2023s2ag}","Duplicate keys of openalex2025topics (Part II), radev2013acl and kinney2023s2 (Part III). One key per work."63"K-90","manuscript/appendices/appC_derivations.tex","760-762","minor","inconsistency","The implementation sets T_max^tr to December 2021 for h = 12 and to December 2020 for h = 24, so that T_max^tr + h is December 2022 in both cases","Cutoffs are months with index = 0 mod 3 from January 2018: the last training cutoffs are October 2021 and October 2020, so T_max + h = October 2022 (the gap inequality still holds). State the realised cutoffs."64"K-91","manuscript/appendices/appC_derivations.tex","772-776","major","inconsistency","stricter for h = 24 than the description of Part IV, which gives December 2021 as the last training cutoff for both horizons---the code's rule is the one that guarantees (eq) and the manuscript follows the code","An appendix must not document a live contradiction with the main text; fix Part IV (K-54) and delete this remark."65"K-92","manuscript/appendices/appC_derivations.tex","47-51","minor","method_weakness","at least one post-event month with t > t_e, guaranteed by the driver's rule that twelve months follow every event","Separating beta_2 from beta_3 needs at least two post-event months (t_post is identically zero with one); the driver's twelve-month rule suffices, so state 'at least two'."66"K-93","manuscript/appendices/appC_derivations.tex","86-87","minor","inconsistency","With n of the order of one hundred and fourteen regressors, the difference between the normal and the t_{n-15} reference distribution","X has 15 columns including the constant (l.32); write 'fifteen columns' to match t_{n-15}."67"K-100","manuscript/appendices/appE_data_dictionary.tex","253","minor","inconsistency","Date the record entered OpenAlex (used for indexing-lag diagnostics)","Part III l.626-634 shows the field reflects a 2025 re-ingestion and was not usable for the lag profile; add the caveat."68"K-101","manuscript/appendices/appE_data_dictionary.tex","130","minor","inconsistency","Record type: article, inproceedings, proceedings, book, incollection, phdthesis, mastersthesis","The parser also emits 'data' records (dblp_parse_report: 30,631; dblp_keys admits type data). Add it."69"K-110","manuscript/appendices/appH_reproducibility.tex","24-25","minor","number_without_artifact","M3U96b (Mac Studio, Apple M3~Ultra, 28~cores, 96~GB of unified memory, 103{,}079{,}215{,}104 bytes as recorded by the environment capture)","The environment capture holds no core count (lines: Python 3.12.15 / 26.4 / Apple M3 Ultra / 103079215104); '28 cores' comes from the RUNBOOK header, while the cluster inventory lists M3U96b at 32 cores. Record hw.ncpu / hw.perflevel in the capture and quote it."70"K-111","manuscript/appendices/appH_reproducibility.tex","107-109; 287-292","minor","number_without_artifact","\texttt{dblp\_publications.parquet} (8{,}840{,}866 records) ... (915{,}115 records) ... (138{,}796 entries)","These provenance counts are also macros (\DblpParsedRecords, \LinkArxivCsRecords, \AclParsedEntries); use them so that a regenerated run cannot leave the appendix stale."71"K-112","manuscript/appendices/appH_reproducibility.tex","32-33","minor","unsupported_claim","M5M36 for the second snapshot pass and M4M64c and M3U96a for embedding shards; they run the same locked environment","No environment capture exists for those nodes under reproducibility/; add captures or state it as an assumption."72"K-113","manuscript/appendices/appH_reproducibility.tex","269-271 (vs Part III Table tab:data-sources)","minor","inconsistency","arXiv metadata obtained through OAI-PMH is under the CC0 terms of that feed","Part III's source table says 'arXiv metadata terms'; harmonise the licence statement across the two."73