Coding dataset for a structured review of automated layup constraints, manufacturing deviations and structural response in curvilinearly reinforced composite structures
Description
This dataset contains the complete coding tables underlying a structured review of how the literature on automated fibre placement and automated tape laying connects manufacturing constraints, the deviations they give rise to, the structural consequences of those deviations, and feedback into the design problem. Two independent layers are provided. The prevalence layer records six implementation indicators and five causal links for a corpus of 167 publications, of which full texts were obtained for 147; proportions are accompanied by assumption-free bounds rather than confidence intervals, because full-text availability was non-random. The mechanism-and-effect layer is a purposively assembled matrix of 91 comparisons extracted from 44 publications, coded by effect class, reference state, direction of comparison, response level, deviation type and origin, fidelity level and the provenance of each reported value. The dataset also contains the screening log for 814 publications, the frozen analytical set of 476, the pilot validation blocks that established the accuracy of coding from bibliographic records, the claim ledger mapping every generalising statement onto the identifiers of the comparisons that support it, the discrepancy register, the metric harmonisation map, the comparison-direction adjudication, the gap-test log, a changelog of two corrections made after the data freeze, and an audit trail from which every numerical value reported in the article can be recomputed. The dataset is entirely in English: identifiers, codes, flags, numerical fields, column names and all free-text provenance, adjudication and interpretation fields. A README with a per-file catalogue and a column glossary is included. Provenance of the values. Every numerical value and every classification in these tables was extracted and verified by the author from the primary sources. A large language model was used only to move information already present in bibliographic records into the fields of the extraction sheet and to assemble comparative tables from fields that had already been coded. No effect class, reference state, comparison direction, fidelity level or numerical magnitude was assigned by a model, and no inclusion decision was made by a model. The accuracy of the dataset does not depend on any tool other than the author's verification against the primary sources.
Files
Steps to reproduce
1. Prevalence figures. Open S5_CODING_COVERAGE_v3.tsv. Rows whose excluded_reason field is empty form the denominator of 147 publications. Counting the value 1 in the columns I0_need_stated, I1_operationalised, I2_defect_represented, I3_effect_quantified, I4_effect_validated, I5_fed_back and link_1_2 to link_5_6 reproduces every proportion reported in the article. Assumption-free bounds for a count y are y/167 as the lower bound and (y+20)/167 as the upper bound. 2. Implementation profiles. S10_PROFILES_FULL.tsv gives all 23 observed profiles with their counts and the publication identifiers in each; the counts sum to 147. 3. Effect figures. Open S11_MASTER_MATRIX_v2.tsv, which holds 91 comparisons from 44 publications. Filtering on effect_class gives the class counts; within class E2, filtering additionally on effect_synthesis_eligible = 1 gives the denominator of 30, and the comparison_direction field splits it into 22 canonical, 5 reverse and 3 between-configuration comparisons. Filtering the canonical subset on response_level = structural_response gives the 19 comparisons used for the sign analysis. 4. Adjudications. S13_D56_COMPARISON_DIRECTION.tsv records the direction decision for every class E2 comparison, S12_E2_HARMONISATION_MAP_v2.tsv the normalisation of metrics and reference states, and S18_E1_CASES_FULL.tsv the canonical and non-canonical split within class E1. 5. Traceability of statements. S14_CLAIM_EVIDENCE_LEDGER_v2.tsv maps each generalising claim onto the identifiers of the comparisons that support it, together with the limitation attached to that claim. S15_CONTRADICTION_LOG_v2.tsv records each registered discrepancy with its two sides and its conditioning factor. S16_GAP_TEST_LOG_v2.tsv records the gap tests. 6. Provenance of every number. S17_METHODS_AUDIT_TRAIL.md states, for each figure reported in the Methods and Limitations sections of the article, which file and which field it comes from and how it is computed. S7_CHANGELOG_POST_FREEZE.md lists the two corrections made after the data freeze and every quantity they affected; S6_CODING_COVERAGE_frozen.tsv is the pre-correction version, retained for comparison against S5.
Institutions
- Bauman Moscow State Technical UniversityMoscow, Moscow