PASS

Discovery · 17 logics · IBM Heron r2 · ibm_marrakesh

Seventeen ADMET logics.
One architecture.

Keep conflicting absorption, distribution, metabolism, excretion and toxicity logics co-resident in one 156-qubit job, and close a pre-registered architecture scorecard the same day.

What the case is

Lead selection is a multi-logic landscape. Permeability, CYP risk, hERG, DILI and the rest pull in different directions. Classical practice flattens them (weighted sum) or sequences them (hard filters). On the frozen 9×17 public-proxy table those two already disagree: acetaminophen ranks high under a weighted sum and drops at a sequential DILI gate.

Goal

Show that co-resident ADMET logics produce a measurable joint-structure gap versus no-cross and separable controls, with primary metrics on raw counts.

How it was prepared

How it was computed

IBM Heron r2 · ibm_marrakesh · 156 qubits · 8192 shots. Coupled jobs plus a separable control. Primary readout: raw counts. Resilience / SAR is a labelled diagnostic channel on the same jobs.

How it was evaluated

Architecture WOW on a frozen public panel. PASS when every Stage C2 bar clears the same day.

0.185
Raw junction gap (≥0.05)
0.749
Aer retention (≥0.60)
0.184
Coupled − separable (≥0.05)
PASS
Classical conflict
Junction gap
0.185
Retention
0.749
Coupled−sep.
0.184
Bar (0.05)
0.050
Bar (0.60)
0.600
Coupled jobsd9qubi7pemts73crsa2g, d9qubihdsedc73aftfj0, d9qubivpemts73crsa3g
Separable jobd9qubk8pdb6s73e45q3g

This is a measured architecture result on a public benchmark. Classical computers can audit the conflict table and controls. The finding is the joint structure the coupled job keeps.

Original WOW article ADMET desk report Atlas