Seventeen conflicting ADMET logics. One quantum architecture. A same-day pass on IBM Marrakesh.

Quantum Polycontextural Computing (QPC) kept multi-logic drug-property decisioning co-resident in one 156-qubit job — and cleared a pre-registered architecture scorecard on ibm_marrakesh without changing the geometry lock.

Dr. Ilan Kreitmann · quantumpolycontextural.ai · Polycontextural architecture evidence with public IBM job IDs

We congratulate IBM on its success in establishing a certified path to quantum advantage for hard sampling. We have gone another way on the same Heron-class hardware: keep seventeen conflicting ADMET logics co-resident in one coherent job, and close a falsifiable architecture scorecard that classical computers can still audit — same-day pass on ibm_marrakesh, with public job IDs.

What we set out to prove

Classical practice flattens or sequences conflicting logics: a weighted score here, a hard filter there. On our frozen public-proxy ADMET table, those methods already disagree — acetaminophen ranks high under a weighted sum and is removed by a sequential DILI gate.

QPC asks a different question: can those logics live together inside one coherent quantum job, with a measurable joint-structure gap versus controls that destroy cross-context coupling — including the industry habit of running contexts separately and merging later?

The Stage A → C ladder

• Stage A — Classical conflict: same frozen 9 × 17 table; weighted-sum vs sequential filters disagree.
• Stage B — Architecture on Fez: locked junction_ryy geometry; raw junction gap ≈ 0.18 vs no_cross.
• Stage C — Problem-class closure: add separable / merge-fail; require retention vs Aer reference; clear all bars same day.

Marrakesh Stage C2 scorecard (same day)

Backend: ibm_marrakesh · 156 qubits · shots 8192 · primary metric on raw counts (mitigation / SAR secondary).

Coupled jobs: d9qubi7pemts73crsa2g, d9qubihdsedc73aftfj0, d9qubivpemts73crsa3g
Separable job: d9qubk8pdb6s73e45q3g

What this is

A public, falsifiable polycontextural architecture demonstration: seventeen ADMET logics co-resident in one coherent 156-qubit job; predeclared coupled / no_cross / separable controls; primary metrics on raw counts; IBM job IDs; and a Stage A→C scorecard that clears same-day on Marrakesh.

Classical computers can still check our controls and conflict table. The win is architectural: classical flatten / filter / merge cannot keep the joint structure that the coupled QPC job realizes under the locked geometry.

Why Marrakesh mattered

The same locked geometry had already cleared a strong gap on ibm_fez (Stage B). Same-day Fez retries for Stage C cleared gap and separable but missed retention under device drift. Moving the optional Stage C2 track to ibm_marrakesh — without retuning junction parameters — produced a clean same-day full pass. That is how the WOW closed.