Use cases and boundaries

This page describes behavior implemented by STOMS 0.2.0.

Good fits

Local research ledgers

Use a SQLite ledger for an experiment, project, or research collection. The file can be backed up, transferred, and copied to node-local scratch after writes complete.

Provenance-aware structure and label ingestion

Ingest ASE structures with a StructureSourceSpec and Provenance, then add DFT, MLIP, or other evaluations for their exact structure IDs. Standard energy, forces, and node_energy are shape-validated; additional properties may also be stored.

Duplicate-aware collection

Default ingestion detects equivalent calculator-free ASE content. Exact repeats reuse a record; changed immutable provenance or source specification is reported as a conflict. Distinct equivalent records require explicit opt-in.

Dataset preparation

Discover data by formula, purpose, time, method metadata, property availability, and effective tags. Strictly select one evaluator per structure before export. Use a stored query for a reusable dynamic definition and an aggregation for stable ordered workflow membership.

Traceable export and transfer

Export strict selections to extxyz with labels and traceability metadata. Transfer complete SQLite ledgers additively while preserving IDs, timestamps, and post-ingestion tags.

VASP evaluation adaptation

With the optional vasp extra, parse a VASP directory into properties, units, MethodSpec, and Provenance. The caller supplies and validates the matching ASE structure before ingestion.

Not provided

  • A database server, multi-user coordination, authentication, or access control.
  • A workflow scheduler, job submission system, retry engine, or workflow store.
  • Analysis, plotting, metrics, model training, or scientific-equivalence inference.
  • Mutation/deletion of structures, evaluations, trajectories, or aggregation membership. Stored queries can be hidden/deleted; tags can only be added.
  • Queries over numerical values, atom counts, array contents, or geometry.
  • Unit conversion.
  • A general input/output artifact store or automatic VASP ingestion.
  • HDF5 support, including transfer or import/export.

Operational constraints

Coordinate writers externally and never copy an active SQLite database. The default DELETE journal mode leaves a quiescent database in one file. If using WAL mode, call SQLiteRepository.checkpoint() and coordinate access before copying. Broad structure, evaluation, and catalog iteration is cursor-backed; trajectory, stored-query, and aggregation listing materializes results in memory.