The conditional audit in the current model card shows weak evidence below room temperature. The right response is more independent test data, not a looser interval or a tuned split.
A useful next contribution would be one bounded, carefully checked tranche of cryogenic steel fracture-toughness measurements from public primary sources. NIMS structural-material data sheets and the DTIC plane-strain toughness handbook are promising starting points, but another openly accessible source is welcome.
Data requirements
- Aim for at least 15 usable records from two independent source documents or publication groups.
- Prefer specimen
K_IC, J_IC, or clearly documented elastic-plastic toughness measurements below 293 K. Do not mix in Charpy energy or indentation-derived estimates for this tranche.
- Record composition, material condition, processing or heat treatment, test temperature, specimen geometry, toughness measure, original units, and the converted project units.
- Give every row a durable citation plus the exact page, table, or figure needed to find the number again.
- Treat OCR as a transcription aid only. Every numeric value should be checked against the rendered source.
- Document any J-to-K conversion and the assumed elastic constants row by row or source by source.
Integration and verification
- Add the raw or manually transcribed records through the existing provenance-preserving ingest path.
- Add schema, unit-conversion, duplicate-detection, and deterministic-ingest tests.
- Keep publication grouping intact so records from the same source cannot leak across grouped validation folds.
- Rerun the fixed conditional audit and update the model card with the result, whether coverage improves or not.
- Summarize exactly which temperatures, alloys, heats, and measurement types were added.
The goal is a defensible evidence increment, not a larger row count for its own sake. A smaller tranche with excellent traceability is more valuable than a large opaque scrape.
The conditional audit in the current model card shows weak evidence below room temperature. The right response is more independent test data, not a looser interval or a tuned split.
A useful next contribution would be one bounded, carefully checked tranche of cryogenic steel fracture-toughness measurements from public primary sources. NIMS structural-material data sheets and the DTIC plane-strain toughness handbook are promising starting points, but another openly accessible source is welcome.
Data requirements
K_IC,J_IC, or clearly documented elastic-plastic toughness measurements below 293 K. Do not mix in Charpy energy or indentation-derived estimates for this tranche.Integration and verification
The goal is a defensible evidence increment, not a larger row count for its own sake. A smaller tranche with excellent traceability is more valuable than a large opaque scrape.