Quality at scale is meaningful only if the answer is identical to the reference. We hold ourselves to that standard. Energy fidelity |ΔE| ≤ 1.1 × 10⁻⁹ H is three orders of magnitude tighter than the SCF convergence floor most commercial stacks deliver. σ-profile relative error stays below 6 × 10⁻⁵ on every drug we have measured.

The 100-drug Ground-Truth cohort is the proof: under 40 minutes of wall time, 99 of 100 drugs delivered with a SHA-256-audited .mfsig.json file. The artefacts are open; anyone with the same library versions can re-run and obtain bit-identical output.

Measured fidelity

Aspirin σ-profile Pearson r vs reference ground truth: +0.99991. Atorvastatin (76 atoms, 749 atomic orbitals): +0.99982. |ΔE| ≤ 1.1 × 10⁻⁹ H across the cohort. Every cohort entry passes the .mfsig integrity verification.

Industrial-scale throughput

The 100-drug Platinum cohort delivers in under 40 minutes. The 389-drug DrugCentral cohort runs in ~2.5 hours. Throughput scales linearly to thousands of drugs without per-molecule quality drift.