NP-hard organ logistics with stochastic prediction and QAOA pathfinding

StochVRP-Mesh · Organ Transplant Research Platform

Kryptur Research presents StochVRP-Mesh — a quantum-ready operating system for time-critical medical cargo. Quantile-regression risk estimates for dwell time, travel time, and locker diversion feed classical VRPTW routing and QUBO/QAOA shortest-path solvers under ischaemic cold-chain deadlines.

10.5281/zenodo.20944392 Raja Ram ColdMesh 1.0

Locker diversion ROC-AUC

0.849

Success probability model · 52% base rate

Travel PI coverage

0.771

80% prediction interval · 80,000 training samples

Dwell PI coverage

0.747

22-feature advanced dwell quantile regression

VRPTW scenario

15×4

Stops × vehicles with hard ischaemic windows

Recent news

StochVRP-Mesh: Stochastic Prediction and NP-Hard Routing for Time-Critical Organ Transplant Logistics

Research Paper

Raja Ram M · Main Author · Kryptur Quantum R&D · Zius Research Center

July 2026

This report documents StochVRP-Mesh: quantile-regression models for facility dwell time (q10/q50/q90), inter-site travel, and locker diversion success, fused into a cold-chain risk synthesis against remaining ischaemic budget. Classical OR-Tools VRPTW returns Pareto-feasible multi-vehicle routes; a directed-edge QUBO shortest-path formulation is solved with COBYLA-tuned QAOA (reps=2) on IBM ibm_fez.

NP-hardVRPTWOrgan Logistics

Predictive layer model status — dwell, travel, locker

Figure 1

80% PI coverage and ROC-AUC for the three inference endpoints

QuantileAUC

Classical VRPTW ischaemic time windows across organ types

Benchmark

15-location · 4-vehicle hard window scenario (OR-Tools)

OR-ToolsCold chain

Open access archive on Zenodo Read more research on our hub

Stay up to date

with the latest from

Kryptur Research on

Results & figures

Table 1. StochVRP-Mesh predictive endpoints and validation metrics

Endpoint	Model	Primary metric	Notes
POST /predict/dwell_advanced	Quantile regression (22 features)	80% PI coverage 0.747	Pinball q10/q50/q90 = 0.47 / 1.03 / 0.46
POST /predict/travel_time	Quantile travel model	80% PI coverage 0.771	80,000 training samples · cyclic time encodings
POST /predict/locker_advanced	Binary success classifier	ROC-AUC 0.849	Base success rate 52% · weather & slack features

Table 2. NP-hard routing and quantum pathfinding stack

Layer	Method	Role
Classical VRPTW	OR-Tools · Guided Local Search	Multi-vehicle organ routes with hard ischaemic windows
Voxel path graph	64³ volume · 8 waypoints · radius connect	Facility / corridor geometry for shortest-path QUBO
Quantum sub-solver	QAOA reps=2 · ibm_fez SamplerV2	Directed-edge QUBO with flow penalties · COBYLA params
Risk synthesis	Travel + dwell vs ischaemic budget	Best / median / worst cold-chain margin read-out