The same input should always produce the same output. No stochastic search, no random seeds, no instance-specific tuning — every result below is publicly reproducible.
One unified configuration — no per-instance parameter tuning — executed across every CVRP benchmark family, from classical 1960s–1990s test sets through the 10,001-node Set XL.
Set A
27
Set B
23
CMT
14
Set E
13
Set F
3
Set Golden
20
DIMACS
12
Set M
5
Set P
24
Tai
13
Li
12
AGS
2
Set X
100
Set XL (hyper-scale, up to 10,001 nodes)
100
XML Stress Test
10,000
Scale Validation, 100 → 10,001 Nodes — Built Alone, On a PhoneLive demonstration of GSL Solver CVRP instances at 100, 1,001, 9,571 and 10,001 nodes, each independently confirmed Feasible using the official CVRPLIB / PUC-Rio solution checker. Developed and executed natively on Android hardware via Python/Pydroid 3.
Comparative Evaluation vs. Published Heuristics
GSL Solver evaluated against ALNS and I1 — established reference algorithms for VRPTW — on the standard Homberger benchmark family.
Across the Homberger 800 benchmark set, GSL Solver matched or came within a few vehicles of the best-known vehicle count on all 60 instances, while the ALNS reference baseline showed substantially larger gaps (e.g. instance R2_8_1: GSL +1 vehicle vs. ALNS +38 vehicles). Full instance-by-instance logs are available in the linked repositories below.
Evidence Gallery
Benchmark reports published throughout the GSL Solver evaluation program.
Solomon 100 — Execution Logs56 instances across Clustered, Random and Mixed distributions, with real timestamped runtime logs.
Distribution TypesRoute visualizations for the three Solomon customer distributions: Clustered (C), Random (R), and Mixed (RC).
Independent Verification ArchitectureEvery generated route is checked by a verification engine separate from the routing core before being authorized as feasible.
Full-Scale Coverage MilestoneProgressive validation from 100 to 1,000 customers per instance, culminating in 13 exact best-known-solution matches.
VRPTW Benchmark Program — Final Summary354 public benchmark instances evaluated across the complete Solomon and Homberger benchmark families (100–1,000 customers).
Why This Milestone Matters
Rather than optimizing each routing variant independently, GSL Solver is a unified deterministic optimization architecture capable of solving multiple routing problem classes while remaining reproducible, verifiable, and deployable on ordinary hardware.
Every benchmark and validation presented in this portfolio was developed and executed using Python on an Android smartphone via Pydroid 3 — a 100% mobile-first engineering workflow.
This benchmark portfolio establishes the first public technical foundation of GSL Solver and provides a basis for future routing modules and commercial deployment.
Reproducible Evidence
Every benchmark instance, execution log, and validation result referenced above is publicly available. The index below links to the complete GSL repository network.
