Scattered kinetic data for permutation-invariant reaction-mechanism inference (M1–M20)

Simulated scattered kinetic observations for 20-class organic reaction mechanism classification (M1–M20). This upload accompanies the work “Permutation-invariant neural inference of reaction mechanisms with scattered kinetic data”. The archive scattered-kinetic-data.zip unpacks to: data/default/{train,val,test}.parquet data/combined/{train,val,test}.parquet plus the schema README files. Unpack at the Scattered_Set_Transformer repository root. Two splits share the same table schema. A model sample is one group (mech_id, k_hash): a variable-length set of kinetic points. One parquet row is one observation at a single time. - default/: catalyst loading cat0 in [0.01, 0.10] (1–10 mol%).- combined/: cat0 in [0.001, 0.20] (0.1–20 mol%), covering 0.1–1%, 1–10%, and 10–20%. Columns: mech_id, k_hash, t, tau, X, RP, cat0, S0, P0, St, Pt, t_ss. Apache Parquet. Code and trained weights will be released in a separate GitHub repository (Scattered_Set_Transformer).