RILEM flexural specimens with their test context

Correction made 15 September 2026

The previous nine-row excerpt pooled incompatible test methods, specimen scales and conditions. Its averages cannot support the old 32–55% comparison. The old CSV is retained only as superseded history. The published manuscript v1.2 describes that frozen historical excerpt; its counts do not describe this new export.

What one row means

The corrected CSV contains 200 individual flexural specimens, each with its own source sample identifier in provenance_notes. n_specimens is 1. Strength and age are copied from that specimen; no averages, standard deviations, cast/printed ratios or effect estimates are calculated.

The companion provenance ledger maps CSV row order to source sample_id, full sample name, compound orientation, bending method, source object, process condition, age, dimensions, support/load spans, loading rate, dates and original units. These source identifiers are not specimen-to-print-to-batch validation. The extractor does not guess a parent print or material from a mix prefix. Fresh rheology and print-process settings are therefore omitted rather than copied from the first matching record.

Inclusion and units

Selection retains the three original mix-name prefixes 01_a, 13_a and 19_a, a finite nonnegative flexural result in N/mm² or MPa, and age from 20 through 40 days. The units are numerically equivalent. Exact prefix matching avoids SQL wildcard ambiguity. All qualifying individual specimens are retained; there is no top-three-orientations rule.

Of 292 source records in those prefixes, 200 are emitted and 92 excluded. The inclusion ledger records every decision and its source strength/unit. Dispositions: included: 200, missing_nonfinite_or_negative_strength: 1, outside_20_to_40_days: 51, unrecognized_strength_unit: 40.

Corrupted source unit strings are not interpreted as MPa or kg/m³. Unsupported strength units exclude a specimen; unsupported optional dimension/density units leave that output field blank while preserving the source value/unit in the context ledger. Zero is distinct from missing. No conversion based on an assumed density is performed.

Orientation and comparison limits

Full source codes such as U.W remain intact in the ledger and notes. No first-letter X/Y/Z projection is made. CAST is retained as source preparation; printed compound codes identify printed preparation. Unknown orientation leaves preparation unresolved. Bending-method descriptions remain source context rather than an invented standard code.

Individual records are not automatically comparable or independent. A future analysis must explicitly establish compatible test geometry, method, age, condition and source-linked printing/batch relationships. This export is a traceable conversion example, not a replacement RILEM study.

Source and reproduction

Bos, F., Robens-Radermacher, A., Muthukrishnan, S., Versteegen, J., Wolfs, R., Santhanam, M., Menna, C., & Mechtcherine, V. / RILEM TC 304-ADC (2024). Database of the RILEM TC 304-ADC interlaboratory study on mechanical properties of 3D printed concrete (ILS-mech) [Dataset]. Zenodo. DOI 10.5281/zenodo.12200570. CC BY 4.0.

License: CC BY 4.0. Original database/workbook is not re-hosted. Download the source from its repository. The expected SHA-256 is 674839356d64bf9444b5b1eac8b198f68559129b571da9e82c53ae73c2a3e0df; a changed source file requires a new audit before replacement.

Run from the repository root:

python examples/rilem-tc304-ils-mech/build/extract.py "/path/to/2024-06-21_openbis.db" --output-dir /path/to/output

The input is opened read-only; the extractor checks its hash again after export. The default output directory is this example folder. Extraction report · CSV · Attribution.