Authorized Access Only
This platform executes live quantum simulations on a governed research infrastructure (FDA 21 CFR Part 11-aligned, not a certified-compliant system). Access is granted to credentialed researchers only.
⚛
SOLANGE™
Scientific Oncology Legacy: Adjudication of Necessity through Governed Evidence · CGU 2026
DC
Doron Cohen · CGU
Scientist View

Clinical Target Dashboard

NSCLC Non-Druggable Mutation Profiles · Source: TCGA · COSMIC · ClinVar · NGS Report MI25-0349

Non-Druggable Target Registry — Active Queue
Class Gene Mutation Source Local e⁻ · Phase 3A Full e⁻ · Phase 3B Status
★ NGS = detected in patient report MI25-0349 · 📚 Research = TCGA/COSMIC/ClinVar literature target · Sibling = same gene/domain as NGS finding · Class A/B rows are quantum simulation targets · Class C rows filtered out after NGS load
Case NGS Profile — No report loaded
BiomarkerResultImplication
EGFR / ALK / KRAS / RET——
TMB——
MSI——
TP53——
KEAP1——
i
 Classical Tractability Plane — Size vs. Measured Entanglement
Why S_max alone is not a hardness measure — supporting evidence (C275F negative-control round, M=125) ▾
i
 S_max vs. Correlation Energy Recovered — Matched Bond Dimension (M=125)
Scientist View

Orchestration Engine

Quantum kernel · HPC dispatch · AI binding affinity pipeline

Gate 1 — Structural Resolvabilityi
Pick a variant straight from the loaded NGS report (fills gene + mutation below automatically), or type one by hand:
Or type exactly what an NGS report gives you — gene + mutation notation (e.g. "DNMT3A" / "p.Arg882His") — no PDB/chain/amino-acid-code knowledge needed:
Target/PDB/residue fields below — filled in by "Resolve" above, or fill in by hand for a chemist's own manual check. Chain is optional: leave blank to auto-check every chain in the structure and see which one has the residue.
Pending promotion (run scripts/laguna/promote_gate1_checks.py end of day):
Ladder · Rung 1 — Laptop smallest rung: prove the whole pipeline in-browser before climbing to HPC and quantum
N₂ Sanity Checki
⚛
N2 self-test — Awaiting run
Quantum Kernel · 8-qubit JW Hamiltonian (329 Pauli terms) · PySCF CASSCF(4e,4o) / STO-3G
Idle
Ladder · Rung 2 — HPC classical CASSCF/VQE dispatched to the Laguna cluster — beyond the laptop's reach
HPC Ladder · Laguna (USC CARC)i
Ladder · Rung 3 — DMRG + SHCI last classical resort — the A/B/C classifier that decides whether quantum is even needed
The Classifier — PDB to a Two-Method Decisioni
Part 1 — cluster (cached if this exact site was built before, otherwise built from scratch) · Part 2 — classify with both DMRG and SHCI, and report agreement
Classifier results — Method 1: DMRG
DMRG A/B/C Classifications · Laguna (solange_dmrg.py)i
Gate-2 heavy runs · SOLANGE → Slurmi
VERDICTS
Tick the runs that make up one target (both legs, every rung extension, the three controls), then Propose.
Classifier results — Method 2: SHCI
SHCI A/B/C Classifications · Laguna (solange_shci.py)i
Ladder · Rung 4 — Quantum bottom of the ladder — real IBM Heron QPU, where the classical rungs run out
Quantum Hardware · IBM Heron (real QPU)i
Gate 2 — Mechanism Category & Chemist Sign-offi
Custom Model Compound i
Uses the target/category/sign-off fields above. Paste a real .xyz geometry below — this platform does not generate molecular structures.
Stabilizer Comparison — RETIRED i
History only — no new comparisons can be computed here. Use Local Tractability Shift instead.
Local Tractability Shift i
Redox / Spin-Gap Comparison i
Requires a complete Gate 2 sign-off (category = metal_redox_center) for the target, and two DMRG records that match on everything except spin.
Catalytic Loss-of-Function — Status Lookup i
Scientist View

Results & Benchmarking

DMRG / SHCI Classification · Rung 3-4 Escalation Evidence

VQE Energy Convergence — awaiting simulation
Run a simulation to see the live convergence curve from PennyLane
— — — Iter 0 Iter 20 Iter 40 Iter 60 Iter 80 E₀ HF ref Run simulation to plot live VQE convergence
Energy in Hartree (Hₐ) · X axis: Adam optimizer iterations · Y axis: total ground-state energy · Convergence = energy plateau
All-Target Summary — 1 NSCLC Mutation · BQP Classification & Simulation Status
Reference data · These energies are computed on a small stand-in molecule for each target, not on the target's own active site, and the electron counts shown are literature estimates rather than measurements. A BQP class stays an expectation until a DMRG run at the target's own scale confirms it.
TargetMutationActive e⁻Qubits (JW)BQP ClassStatus
★ marks a Class A (provisional: unresolved classically at the declared budget) target. JW = Jordan-Wigner encoding. Load an NGS report to add patient-specific targets.
Live Results — Supabase Database (Real Runs)
Loading live results from Supabase...
Quality / Regulatory View

Compliance & Audit Trail

FDA 21 CFR Part 11 · Reproducibility Engine · Provenance Metadata

No simulation run yet — load an NGS report and run a VQE/sqDRIFT simulation to generate a compliance record.
Audit Trail — Full Run Log
No run completed yet — run a simulation to populate the audit trail.
P1–P9 Provenance Schema
ElementDescriptionLayerStatus
P1 Circuit fingerprint — ansatz type, gate count, depth, qubit map QC Kernel ○ Pending
P2 Compilation lineage — transpiler version, SWAP overhead, native gate set Hybrid ○ Pending
P3 Device & calibration — backend ID, calibration epoch, T1/T2/gate fidelity snapshot QC Kernel ○ Pending
P4 Error budget snapshot — per-gate error rates at execution time QC Kernel ○ Pending
P5 Raw outcome distribution — shot-level bitstring histogram before mitigation QC Kernel ○ Pending
P6 Error-mitigation lineage — ZNE noise factors, Pauli twirling config, mitigation version Hybrid ○ Pending
P7 Statistical estimator + 95% CI — mitigated energy, bootstrap interval, GNN-corrected value Hybrid ○ Pending
P8 Cryptographic seal — SHA-256 hash chain P1→P2→…→P7+P9→P8 · tamper-evident seal Governance ○ Pending
P9 ML-decoder lineage — Nvidia Ising 3D CNN model ID, weight checksum, training corpus hash, inference config (conditional: present when QEC uses a learned ML decoder) Hybrid ○ Pending
Platform Qualification Record — System Validation Status (VP-2026-001)
IQ — Installation
NOT FORMALLY EXECUTED
◐ Software stack — running in production, not formally verified
◐ Hardware connectivity — exercised informally, not a controlled test
? Dependency versions — not checked
? Environment isolation — not tested
No IQ protocol executed or signed yet — scoped as part of the Regulatory Validation roadmap.
OQ — Operational
NOT FORMALLY EXECUTED
✓ VQE convergence — N2 CAS(6,6)/cc-pVDZ reproduced to 0.00 mHa classically, 0.471 mHa on 12-qubit VQE
✓ Audit log integrity — append-only ledger, seals independently re-verifiable on demand
✓ TREX readout-error mitigation — Qiskit Runtime resilience_level=1, cheap (no extra QPU shots), always on
✓ ZNE error mitigation — Qiskit Runtime resilience_level=2, opt-in per job (multiplies QPU time/cost, so requested deliberately, not by default)
? Backend failover — not tested
No OQ protocol executed or signed yet — scoped as part of the Regulatory Validation roadmap.
PQ — Performance
NOT STARTED
◑ Cross-backend reproducibility (Phase 3B)
In progress — first cross-backend evidence: canonical H2 measured on a classical statevector simulator and on real IBM Heron r2 hardware (ibm_marrakesh), agreeing to Δ 0.02 mHa. This is a moving-parts smoke test on H2, not a qualified scientific result; full PQ requires the funded runs on the actual mutation targets.
Platform: SOLANGE v0.9.2-alpha Validation lead: D. Cohen · doron.cohen@cgu.edu Protocol: VP-2026-001 Next review: 2026-05-01
IT / Security View
🛡 Admin Console
Platform access log · Simulation runs · User management
—
Total Runs
—
Users
—
Mutations Run
—
Last Run
Security Events — Denied Access Attempts
Overreach only — not a request log. Every row here is a 403 the platform already enforced; this is the record that it did.
Loading…
All Simulation Runs
Loading…
Provenance Audit Log
Loading…
User Profiles
Loading…
Executive View

Executive View

Platform verified Reference data

Measured quantities only. Consumption is reported, not cost — the rate is your commercial term. Running on reference data, so results are not clinical findings; production runs use your own structures.

Nothing is stored unchecked

Every result is re-verified on arrival. Refusals and limits are recorded too.

Loading…
Quantum time used

From IBM's own billing figure — device time, not queue time.

Loading…
Your rate per QPU-second
$
Classical HPC used by governed heavy runs

Core-hours and peak memory as metered by the cluster's scheduler and sealed by LEON with each run. A run refused before it starts costs nothing; a run refused mid-way is counted at what it consumed.

Loading…
Your rate per core-hour
$
Where classical computing stops, per target
Classical compute at larger bond dimension
Extrapolated, not measured

Anchored to one real run — TP53 C275F mutant, CAS(36,34), this project's own HPC allocation: M=500 took 1826.5s wall-clock. Everything below is that single point raised to a chosen power — it is a projection, not a second measurement, and three different, genuinely disagreeing exponents are offered rather than one presented as settled (see note).

Target bond dimension M
Scaling law
Cores (this allocation)
Your rate per core-hour
$

Core count for the 1826.5s anchor run was not itself logged; 8 reflects this project's standard SLURM allocation (--cpus-per-task=8, USC Laguna/CARC) and is editable. The three laws are shown because they genuinely disagree — "M^2.21" is this specific three-point ladder (M=250/500/1000) fit by this instrument; "M^1.91" was reported elsewhere in this project from different runs and has not been independently re-derived this round; "M^3.05" is a published third party's own exponent on their own hardware, applied here only as a hypothetical to this project's anchor, never as their absolute time. Extrapolation beyond the calibrated range (M=250–1000) is unvalidated by this instrument. Quantum-side cost (Heron r3 / sqDRIFT) is not shown here — there is no current access channel to that hardware, so no number would be real.

What this screen deliberately does not tell you
Shown as missing rather than estimated.
Time saved
Needs the same task run manually to compare against.
Decision quality
Not judgeable by the system that produced the result.