Capabilities

Features

Every production path documented; every roadmap item dated. Honest about what's shipped, partial, or pending.

Quality

Five-nines σ-profile fidelity

Pearson r vs reference CPU ground truth = +0.9998. Energy fidelity 1.1 × 10⁻⁹ H.

SHIPPED

Quality

Smart Conformer Funnel

ETKDG embed → UFF preopt → xtb rank → DFT top_k → Boltzmann avg.

SHIPPED

Free

Rosetta Stone IO

7 vendor formats with auto-detect, chameleon writers, batch migration.

SHIPPED

Format

.mfsig Apex format

SHA-256 audit trail + reproducibility metadata + legacy_vault.

SHIPPED

Throughput

CPU Hyper-Scaling

Thread caps + MINAO init + coarse-fine grid + r2SCAN fast-preset.

SHIPPED

Cavity

SES Phase 3 — analytical reentrant

Closed-form Connolly probe-touch-3 patches. Girard's-theorem area + great-circle inclusion. 5/5 unit tests at machine ε.

SHIPPED

Geometry

Dual-tier geometry optimisation

Platinum-standard geometry relaxation driven by analytical quantum gradients. Σ∇E = 2 × 10⁻¹⁶ on H₂ — machine ε.

SHIPPED

Access

Production delivery pipeline

Reference-grade σ-profile generation at fp64 precision end-to-end. σ-profile fidelity at five-nines (Pearson r against the reference ground truth +0.9998).

SHIPPED

Tier 2

Industrial-scale cohort throughput

100-drug Platinum cohort delivered in under 40 minutes with the same audit trail per drug. Suitable for large-cohort screening.

ROADMAP

Tier 3

Custom solvent / radii calibration

Reference-tier offering: cavity radii fitted per solvent family, COSMO-RS-aligned scaling, peer-reviewed methodology.

ROADMAP

Tier 3

MolForge Ground-Truth cohort (100 drugs)

Reproducible, signed σ-profile reference. 99/100 drugs PASS all audit gates. Published geometries, σ-moments, and SHA-256 trust hashes per entry.

ROADMAP

Tier 4

ML latent-vector encoder

Cavity → 512-d embedding for ML feature stores. .mfsig pillar 3.

ROADMAP

Tier 4

Parquet dual-scale

.mfsig.parquet for 10⁸+ molecule libraries. Vectorised scan.

ROADMAP

Audit tiers

Three audit tiers (v0.29)

Standard production · Compatibility (drop-in for legacy pipelines) · Premium pure-physics (R&D partner tier). Same signed .mfsig schema across all three — what changes is the audit context, not the engineering.

SHIPPED

Schema

v2.5-multivariant

One .mfsig holds results for multiple recipes side-by-side under physics_variants. Switch downstream pipelines without re-computing the geometry, cavity, or DFT density. Backward compatible with v2.4.

SHIPPED

Chemistry

Chem-aware σ-moments

sigma_moments split by chemical role: hb_donor_mass · hb_acceptor_mass · pi_negative_mass · polar_area_aa2 · halogen_mass. Bond-table classification, not just σ-threshold. Restores H-bond donor/acceptor signal across the cohort.

SHIPPED

Lineage

Git-for-σ-profile (v2.4+)

SHA-chained derivation history under audit_and_trust.lineage. Tier upgrades (Pro → Platinum → Reference) and recipe enrichments append to the chain — no re-compute, no double pay, no lost provenance. Fully signed, fully auditable.

SHIPPED

Atlas

Drug × polymer × solvent atlas

Thousands of drug-like molecules pre-computed at v0.29-directqm-svp. Pair-score matrix available by subscription. Use for screening, lead-finding, ASD formulation candidate ranking, ternary rescue patent leads.

SHIPPED

Q4 2026

Next-generation engine class

A new calibration approach is in validation. Public release Q4 2026 (after the current embargo). We don't detail the architecture yet — see roadmap.

ROADMAP

quality

How we measure quality

Every cohort entry passes four independent gates: SCF convergence, σ-profile non-empty, .mfsig SHA-256 verification, and geometry RMS-grad ≤ 10⁻³ H/Bohr. 99/100 drugs PASS all four — and the .mfsig records the per-drug numbers for forensic audit.

σ-profile correlation vs reference

Aspirin	+0.99991
Atorvastatin (76 atoms)	+0.99982
100-drug cohort mean	+0.9998
Energy fidelity \|ΔE\|	1.1 × 10⁻⁹ H
cohort PASS rate	99 / 100

What gets verified

• σ-profile Pearson r vs ground truth: +0.9998
• Energy fidelity: 1.1 × 10⁻⁹ H
• 5 / 5 cavity unit tests pass at machine ε
• 223 / 223 / 0 / 0 pytest (tests / pass / fail / err)

Tied to the SLA

Every delivered .mfsig.json is signed, gates-verified, and tied to the per-drug audit log. If any gate fails on a paid drug we re-run on our dime. The Reference tier additionally ships an automated 21 CFR Part 11 ALCOA stamp and a DOI-backed Zenodo deposit.

quality

Why the funnel saves ~10× DFT compute

xtb GFN2-xTB ranks low-energy conformers consistently with DFT for ~95 % of drug-like molecules. Surviving conformers contribute > 95 % of the Boltzmann weight; the rest add noise. 32 → 3 funnel preserves σ-profile accuracy.

Flexible-scaffold σ-profilebash

POST https://api.mfsig.com/v1/sigma_profile
{
  "smiles":    "CC(=O)Oc1ccccc1C(=O)O",
  "tier":      "platinum",
  "ensemble":  "auto",          // smart conformer compression
  "solvent":   "water"
}

# response (after webhook):
{
  "mfsig_url":   "https://mfsig.com/d/job_8K2....mfsig.json",
  "sha256":      "e3b0c44298fc1c14...",
  "tier":        "platinum",
  "n_conformers_used": 3,
  "boltzmann_weighted": true
}

Conformer funnel keeps the top-k by Boltzmann weight before DFT — same audit-grade σ-profile, a fraction of the compute. Pricing in the calculator already reflects the deal-of-the-day.

rosetta · free

Read every vendor. Write to any.

Turbomole / COSMOtherm: production fp64 round-trip. Gaussian / NwChem / ORCA: partial (best-effort on common cases). xtb native dump: production (disambiguated from Turbomole by $gfn header). MolForge CSV: native interchange format for ML pipelines.

One file — drag, drop, signedbash

# Web UI:  https://mfsig.com/converter

# CLI (free, no key, OSS):
$ mfsig convert legacy_run.cosmo
✓ vendor detected: Turbomole
✓ wrote out.mfsig.json (signed, SHA-256 audit)
✓ verify_integrity() → True

Batch — migrate a 10-year archivebash

$ mfsig convert /data/legacy_turbomole/ \
       --batch --out /data/mfsig/

  120 files · 7 vendors auto-detected · 0 errors
  signed .mfsig.json per file · idempotent · resumable

# Outputs a manifest with SHA-256 + provenance per drug.

.mfsig format

Tamper-evident σ-profile artifacts

SHA-256 hash of the canonical state (geometry + methodology + numerical results + solver args). Any edit to a hashed field invalidates verify_integrity(). 21 CFR Part 11 ALCOA-compatible by construction.

aspirin.mfsig.json — five pillars, one signaturejson

{
  "mfsig_version": "2.0-apex",
  "audit_and_trust": {
    "molecule_name": "aspirin",
    "sha256_checksum": "e3b0c44298fc1c14...",
    "timestamp_utc":   "2026-05-12T06:55:17Z",
    "reproducibility": { "lib_versions": {...}, "rng_seed": 42 }
  },
  "chemistry_and_geometry": {
    "smiles":       "CC(=O)Oc1ccccc1C(=O)O",
    "inchi_key_14": "BSYNRYMUTXBXSQ",
    "atomic_numbers": [...],
    "positions_aa":   [...]
  },
  "quantum_and_thermodynamics": {
    "methodology": { "functional":"B3LYP", "basis":"def2-SVP",
                     "solvent":"water", "tier":"pro" },
    "results":     { "e_total_h": -648.236489, "g_polar_h": -0.0149,
                     "sigma_profile_51": [...], "sigma_moments": {...} }
  },
  "ai_intelligence_layer": { "cavity_latent_vector_512d": null }
}

# any post-write edit invalidates verify_integrity() → False