# Explicit Solvation Benchmark Methodology (Public, 2026-03-05)

## Scope

- Module: `Explicit Solvation`
- Endpoint type: hydration free energy in water
- Dataset: `FreeSolv v0.52 screening subset`
- Cases: `642`
- Objective: evaluate external-experimental agreement for production explicit-solvation hydration predictions.

## Evaluation Setup (public summary)

- Benchmark mode: production screening configuration
- Solvent environment: explicit water
- Solvent molecules per case: `50`
- Trajectory sample frames: `8`
- Equilibration blocks: `8`
- Production stride: `2`

## Ground Truth and Metrics

- Ground truth source: external experimental hydration free energies from FreeSolv v0.52.
- Primary metrics:
  - MAE: `0.3295 kcal/mol` (`1.3787 kJ/mol`)
  - RMSE: `0.4593 kcal/mol` (`1.9219 kJ/mol`)
  - Pearson r: `0.9928`
  - R squared: `0.9857`
  - Median absolute error: `0.2427 kcal/mol`
  - Max absolute error: `2.4019 kcal/mol`
  - Within-uncertainty fraction: `0.7492`

## Solvent Coverage Policy

- Externally benchmarked models in this packet:
  - Water (TIP3P)
  - Water (TIP4P)
- Additional production native explicit carriers (regression-covered in this packet):
  - Methanol
  - Ethanol
  - Acetonitrile
  - DMSO

## Interpretation Policy

- This packet is the official public benchmark for water hydration in the explicit-solvation route.
- Non-water carriers are included as production capabilities but are not claimed as externally benchmark-equivalent in this packet.
- Case-level outputs and a public error-breakdown note are included to support independent review.

## Public Export Note

This public export is sanitized to remove internal implementation details such as local paths, internal runner names, and internal script references.