Reuse-Readiness of Open Additive-Manufacturing Mechanical Data: A Reproducibility Audit of a Continuous-Fiber FDM Dataset

Openly deposited mechanical datasets are increasingly promoted as low-cost resources for secondary analysis in additive manufacturing (AM), yet their fitness for reuse is rarely tested. This study reports a reproducibility audit of a widely cited, openly licensed monotonic dataset for continuous fiber-reinforced thermoplastic composites, comprising nylon specimens reinforced with carbon, glass, or Kevlar fibers, alongside matrix-only references. We conducted a multi-response reanalysis of 11 archived force–displacement records, deriving peak load, initial stiffness, and toughness to peak. Rather than producing a reliable material ranking, the analysis revealed four structural deficiencies that constrain reuse. First, only one representative curve is archived per configuration, preventing estimation of within-condition variance and inferential comparison. Second, the sole nominally duplicated condition—a 60° fiberglass configuration represented by 12- and 969-point files—is internally discordant: Peak loads differ by only 6% (959 vs. 1020 N), whereas displacement to peak differs 7.5-fold and computed toughness 12-fold, causing the top-ranked configuration to change depending on which file is analyzed. Third, the two records already differ about two-fold in stiffness between 20% and 50% of peak load, so the discordance reflects how the records were acquired, not a failure mechanism; the raw files also show three different end-of-test conditions and timing and strain columns that cannot be taken at face value. Fourth, force-only archiving without specimen cross-sectional geometry prevents conversion to stress or specific energy and may conflate geometry with material behavior. A controlled downsampling experiment showed that trapezoidal energy integration is robust to sampling density, with less than 1% error when a 969-point curve was reduced to 11 points. Thus, reuse risk arises primarily from replicate discordance and protocol heterogeneity rather than numerical processing. We propose a transferable four-test reuse-readiness audit and a six-item minimum metadata standard for AM mechanical datasets, linking each requirement to a specific failure mode. Screening six further AM deposits showed that each deficiency recurs elsewhere, while one recent deposit met all six items. Four of the items already appear in the report clause of ASTM D3039; what is missing in practice is their transfer from the test report to the deposited files. For AM researchers, the practical rule is to check replicate structure, displacement channel, and specimen geometry before comparing, pooling, or modeling open mechanical data.