July 2026 · Fusion blanket · Full 156Q
Nine published FLiBe liquid configurations × seventeen contextures on ibm_fez —
full 156 qubits, one auditable Heron job (d9a06m52su3c739l0eeg).
The ORNL–IBM study shows fragmentation error (~110 kcal/mol) dwarfs the quantum solver residual (~0.7 kcal/mol);
QPC couples methods, reliability signals, and engineering objectives in a single polycontextural layer —
without pretending to redo the unpublished chemistry campaign.
Plain English
ibm_fez · 8192 shots · 3 pre-registered variants in one job: coupled · intracontext-matched control · separable · job d9a06m52su3c739l0eeg.Lane A · Architecture. Open joint-structure gap on predeclared bridge pairs (coupled minus intracontext control).
Lane B · Domain readout. Provisional recovery score ranks encoded published outputs — engineering KPI, not ICC proprietary score.
flibe_binding_database.csv).| Variant | ISA depth | 2Q (CZ) | Mean |bridge corr| | Role |
|---|---|---|---|---|
| flibe_coupled | 246 | 1573 | 0.0180 | Polycontextural architecture arm |
| flibe_intracontext_matched | 144 | 838 | 0.0085 | Depth/gate-matched control |
| flibe_separable | 5 | 0 | 0.0090 | Separable baseline |
Lane A pass signal: open joint-structure gap = +0.0095 (coupled − intracontext). Cross-contexture bridges retain measurably stronger correlations on NISQ hardware.
Job ID: d9a06m52su3c739l0eeg · Backend: ibm_fez · Shots: 8192 per circuit · Status: completed
Provisional recovery score from coupled-circuit marginals. Ranks encoded published outputs — not physical extraction rates.
| Rank | Conformation | Recovery score | Published ext-SQD binding (kcal/mol) | Note |
|---|---|---|---|---|
| 1 | 9 | 0.691 | −134.76 | Weakest binding in table |
| 2 | 1 | 0.579 | −175.39 | — |
| 3 | 4 | 0.573 | −138.13 | 3rd-weakest binding |
| 4 | 7 | 0.555 | −188.84 | — |
| 5 | 6 | 0.521 | −157.88 | 6th-weakest binding |
Engineering warning: less-negative electronic binding is not the same as extraction rate, diffusion, T₂/T⁺ speciation, corrosion, or finite-temperature chemical potential. The paper itself requires ensemble averages over hundreds of liquid configurations; this pilot uses nine.
| Dimension | ORNL–IBM ext-SQD campaign | QPC FLiBe pilot (this job) |
|---|---|---|
| Primary goal | Sample LUCJ circuits per embedded fragment; recover energies via ext-SQD + selected CI | Couple already computed binding energies across methods and reliability contextures in one polycontextural inference layer |
| Hybrid structure | Many fragment jobs (classical FCI <13 orbitals; QPU ≥13) + HPC EWF assembly + classical recovery per fragment | One 156Q IBM job with three pre-registered architecture variants — no Python merge of independent fragment results |
| What quantum solves | Individual fragment correlation (66 logical qubits max per fragment) | Simultaneous representation of method disagreement, solver reliability, fragmentation conflict, and extraction/retention objectives |
| Dominant error source (paper) | Fragmentation/bath (~110 kcal/mol family gap) | Same — QPC attacks the model-risk layer, not the 0.7 kcal/mol fragment-solver residual |
| Public reproducibility | Blocked — geometries, FCIDUMPs, LUCJ QPY, job IDs not released | Full pipeline public — CSV database, scripts, IBM job ID, counts JSON |
| Pilot | Domain | 156Q use | Lane A (ICC) | Lane B (decoder KPI) |
|---|---|---|---|---|
| Manufacturing | Assembly-line scheduling | 3 line contextures + padding | Implicit in coupled layout | 0 late jobs, schedule cost 39.79 |
| Logistics | Route assignment HUBO | 3 route contextures | — | Decoder beats greedy (−49.80 vs −49.52) |
| FLiBe / Tritium (this) | Fusion blanket model risk | 17 contextures × 9 conformers + 3 junctions | Gap +0.0095 vs intracontext | Conformation ranking 9·1·4 top |
| MCGS | Gauge-sector ICC | 3 sector contextures | 3/3 raw pass, gaps 0.35·0.24·0.13 | Physics-motivated instance |
QPC advantage shared across all pilots: multiple objectives live in one coupled quantum submission with pre-registered controls — not separate cloud jobs stitched together classically. FLiBe extends this to scientific model risk (method families, fragmentation conflict, data completeness) at full Heron width.
vendor_benchmarks/flibe_tritium/flibe_binding_database.csv (nine conformations, eight methods from Table S3)..venv/bin/python3 vendor_benchmarks/flibe_tritium/qpc_flibe_tritium.py analyzed9a06m52su3c739l0eeg on ibm_fez — retrievable via IBM Quantum Platform.results/flibe_tritium_ibm_fez.json, results/flibe_tritium_run/submission.json, counts.json, decoded.json.qpc_flibe_tritium.py build → logical QPY + manifest.qpc_flibe_compare.py --mode ibm --backend ibm_fez --shots 8192 --wait.--mode aer (MPS after transpile; ~72 s at 512 shots locally).--compiler-plugin.The ORNL study’s own numbers establish a hierarchy of errors:
Replacing ext-SQD with another SQD variant alone would attack the smaller error. QPC’s polycontextural treatment targets the simultaneous representation of local correlation, long-range polarization, method disagreement, and engineering objectives — the layer where classical hybrid workflows currently lack a single auditable quantum object.