Datasets:

---

deep-principle

/

science_physics

Like 2

Follow

Deep Principle 2

Provenance and label fixes from the SDE revision (reviewer items D5 / R3-6).

Everything here is additive: the original provenance column is preserved,
and two new columns are appended.

1. Scenario label fixes

   • materials-safety-classification: 139 of 140 rows carried the
     corrosion-prediction task label, and 1 row carried safety-prediction.
     As released, the 140-question safety set was merged into the corrosion
     scenario and was not addressable as its own scenario.
   • mw2formula: 'Molecuclar' -> 'Molecular'
   • polymer-tg-prediction: 'poylmer' -> 'polymer'

2. Provenance column split
   The released data overloads one free-text column with three kinds of
   value: item sources, contributor names, and the studied entity. Two
   columns are appended:

   • source_ref : upstream databases, DOIs, URLs, benchmarks
   • contributor : the curator, normalised to the full name used in the
     manuscript author list (e.g. Jieyu -> Jieyu Lu, Xiu-Zhe Luo ->
     Xiuzhe Luo, Yifan Zhang -> Yifan F. Zhang)

Values that are neither a source nor a contributor (studied compounds such
as cisplatin) are left out of both columns rather than guessed at.