Derived data, code, figures and source-study navigation for: Research Intent and Pediatric Applicability in Sepsis Host-Response Literature: A Bibliometric and Topic-Modeling Study

Version 5.0.0 archives the revised pediatric sepsis host-response bibliometric and topic-modeling study prepared for Journal of Clinical Medicine. It preserves the version 4.0.0 record (DOI 10.5281/zenodo.23063877) and the version 3.0.0 baseline (DOI 10.5281/zenodo.23036618). Frozen eligible corpora remain 238 Web of Science publications, 243 PubMed publications and a deduplicated union of 267 (214 shared; 29 PubMed additions). Publication cutoff: 29 August 2026; 2026 is incomplete. New in version 5: a source-kind and population audit of all 267 included publications corrects inherited source labels and identifies a 141-publication primary patient-research subset; the historical 98-publication subset remains separate. Six WoS branch comparisons report overlapping diagnostic, prognostic, mechanistic, therapeutic, safety and resource/implementation intents with yes/no/unclear counts and upper-yes sensitivity. Seven-corpus topic matching and five-seed prevalence diagnostics retain unstable results. The 72.3% two-topic share is an unchanged prespecified seed-17 description. Eight DOI/PMID-linked navigation examples connect population applicability, research intent, measured targets and sampling times; seven examples are eligible and one unresolved example is outside the analyzed corpus. The release contains author-created aggregate results, configurations, provenance hashes, data dictionaries, Python analysis code, R figure code, source-study annotations, and six main plus six supplementary figures in 600-dpi RGB TIFF, SVG, PDF and PNG formats. Ten figures reproduce exactly from public tables. Figures 3 and S1 require excluded licensed network tables; their final figures and attribution key are supplied. Model fitting and saved-state diagnostics reproduce from legally accessible exact frozen inputs/models using documented dependencies, row order, hashes and inference sequence. Original frozen probabilities remain distinct from later saved-state re-inference. Licensed database records, titles/abstracts, third-party full text, internal screening/evidence ledgers, publication-level tokens/posteriors and fitted models are excluded. AI-assisted eligibility/intent review covered 315 previously selected publications; earlier decisions are carried separately. The additional source audit is AI-assisted. Coding consistency and topic correspondence are methodological summaries, without a claim of clinical expert, patient-level or biological validation. Publication is the analysis unit, and underlying cohorts can overlap. Derived data, documentation and figures are CC BY 4.0; author-created code is MIT. SHA-256 manifests cover all 155 public files and the uploaded package. All-author manuscript/submission and exclusivity consent is confirmed separately from the actual source-coding method. This dataset release does not represent journal submission or acceptance.