Roadmap — beyond the σ-profile

Phase 3 analytical cavity ✓ shipped. fp64 reference pipeline ✓ shipped. Ground-Truth reference cohort ✓ shipped. The next horizon: ML σ-profile calibration, parquet at 10⁸ scale, and the downstream COSMO-RS thermodynamic predictor itself.

ML calibration layer: train (xtb σ-profile → DFT σ-profile) on the Ground-Truth cohort + DrugCentral cohort. Deploy as a one-line API kwarg that delivers DFT-grade σ-profiles at xtb wall time. The training set is already in hand from our reproducible reference work.

Parquet at 10⁸ scale: .mfsig.parquet for vectorised scan of 100-million-molecule libraries. The .mfsig.json format is forward-compatible — same schema, columnar layout. Enables retrieval-augmented σ-profile workflows and ML feature stores at industrial scale.

COSMO-RS predictor

The σ-profile is the input; downstream COSMO-RS predicts activity coefficients, vapor pressures, partition coefficients. We ship the σ-profile generator today and the predictor on top of it next — closing the loop from SMILES to thermodynamic property in one API call.

Funding the roadmap

Pro / Platinum / Reference tier revenue funds the engineering. No outside dependencies, no external partnerships required. The methodology is open; the audit-graded compute, signed delivery, and re-run guarantee are what customers pay for.