Coding dataset for a targeted critical review of buckling-driven optimisation of variable-stiffness composite structures
Description
This dataset contains the coding tables underlying a targeted critical review of the relation between the analysis formulation and the design outcome in buckling-driven optimisation of variable-stiffness thin-walled composite structures. The review distinguishes prediction fidelity, the agreement of a model with a reference where the comparison was made, from decision fidelity, the property of the mapping from model to selected design. For every study read, the tables record not only whether model levels or criteria were compared, but what was compared: predicted responses or resulting design decisions. Coverage: a preassembled thematic corpus of 1928 studies, of which 450 were classified under a screening scheme fixed before reading began, 116 were shortlisted against a coverage matrix, and 108 records corresponding to 106 unique publications were verified against their full texts. Eight files: the screening and routing log; the protocols and codebook with the version history of the coding scheme; the pilot calibration; the selection and targeted-completion log; the 116-row extraction matrix; the full-text verification report; the provenance record and bibliography; and a post-hoc audit of the approximation metadata flag. Every extracted value carries its comparison baseline, its conditions and a status of Reported, Inferred or Not reported. Values without a stated baseline were not extracted. Limitations. The corpus was assembled before this work began and its search strategy is not available, so no statement about the absence of studies in the literature can be supported by these tables. Selection was targeted, not exhaustive: 443 studies on the priority route and 525 on the fast-exclude route were not read. The counts characterise the selected and verified evidence set and are not estimates of prevalence. Provenance. A large language model was used for relevance triage, for assisted reconciliation of the coding against full texts, and for assembling tables from fields already coded. No inclusion decision, validation target or numerical value was assigned by a model without verification by the author against the primary source.
Files
Steps to reproduce
1. Screening (S1). Counting rows by route gives 775 FULL, 609 CONTEXT, 544 FAST_EXCLUDE. Rows with a non-empty category give 450 classified: Core 121, Relevant 112, Peripheral 128, Exclude 74, indeterminate 15. Rows flagged F_IMPERF give 63, of which 17 carry object_class VS-core or VS-ext. 2. Approximation axis (S5). Restrict to rows whose evidence_set does not begin with "reserve" and filter comparison_axis on approximation: 21 rows. Tabulating validation_target reproduces Table 2 of the article. The error evaluation domain gives five studies assessing agreement at the optimum already obtained and zero quantifying the error in the region of the active constraint or near the stability boundary. Rows whose computational cost field begins with "yes" give 20 of 21. design_outcome_compared is "no" in all 21. 3. Reported discrepancies (Table 3). Eight of the 21 report a numerical discrepancy against a stated reference; the identifiers are listed in S2 section 8, since the split follows from no single field. 4. Comparison baselines (Section 6.2). Rows of S5 with a non-empty baseline_of_comparison: 82. No count of gain-reporting baselines is derived from this field, which mixes optimisation-gain baselines with validation references, experimental measurements and convergence states. The withdrawal of the earlier counting rule is documented in S2 section 8. 5. Records and publications (Section 2.6). S6 records 108 verified records; S7 section 4a documents two duplicate pairs, leaving 106 unique publications, which is the length of the bibliography in S7 section 5. 6. Depth of extraction (Section 2.7). The two conditions defined in S2 section 8 give classes A 52, B 1, C 29, D 26, plus 8 reserve records without a full text: 116. 7. Scope of the flag (S8). Rows of S1 with fidcomp_type = approximation: 54, of which 21 form the analytical subset present in S5. S8 accounts for the remaining 33 record by record. 8. Null result (Section 5.6). Rows of S1 with F_NULL in flags: one, S0278. In flags_at_selection: four. The difference is the post-hoc correction documented in S2 section 5.3 and S6 section 4.
Institutions
- Bauman Moscow State Technical UniversityMoscow, Moscow