XolverOBSERVATORY

Analysis

OverviewDomainsContent mapEvolutionCompareUnique capabilityExperiments

Solver domains

SMTOMTCPMILPSAT · MaxSATQBFSyGuS · CHC
Design dataset7 domains · content map v2

Loading experiment data

Resolving capability…

Contacting the configured Observatory adapter.

∀x⊢Γ ⊨ φφ

Research in progress

Proof in progress_

Observatory is preparing and verifying real benchmark data. Public numbers will be released only after that verification is complete. Placeholder and test results stay hidden until then.

XolverObservatory

Γ ⊢ φ  ·  results withheld until ⊨ verified

Mathematical optimization

MILP

Linear relaxations, branch-and-bound, cutting planes, and mixed discrete-continuous models.

Locate in content mapCompare experiments
Mixed-Integer Linear Programming5 model classs mapped
Preview
Design datasetScenario data · representative multi-domain content for interface and information-architecture validation
Campaign complete
Artifactxolver-mip 0.2.6
Commit95e8b50
BenchmarkXolver MILP portfolio · 2026.08
Completed19 Aug 2026, 18:35
Solved
15,183
81.0% coverage
Benchmark instances
18,750
5 model classs
Regressions
8
vs comparable previous run
Wrong / invalid
2
correctness gate

MILP capability tracks

Domain-native results grouped by model class

Xolver MILP portfolio · 2026.08
model classFamilySolvedCoverageΔ solvedMedianBest baselineStatus
MIPLIBGeneral mixed-integer models4,256of 5,60076.0%
+6915.80 sSCIPstable
0–1 ILPBinary decision models3,546of 4,30082.5%
+745.24 sHiGHSstable
Network flowStructured linear networks2,907of 3,15092.3%
+18810 msHiGHSstable
SchedulingTime-indexed MILP2,406of 3,00080.2%
+4712.60 sSCIPexperimental
PackingKnapsack and placement2,068of 2,70076.6%
+369.43 sCBCresearch

Capability evolution

Coverage on comparable experiments only

Latest research change

What moved this domain

portfolio-2026.08

0–1 ILP capability update

Presolve and network detection close most of the structured-model gap.

Newly solved+74
Regressions−8

Research focus

→Presolve and relaxation quality
→Branching, cuts, and primal heuristics
→Numerical stability and certificate checking

Result composition

Domain-native outcome classes

Proven optimal6,984
Feasible / bounded8,199
Unknown / resource limit3,565
Wrong / invalid2

Evidence & baselines

What makes this view reproducible

HardwareAMD EPYC 9654 · 2 cores / task · Linux x86-64
Limits3600s · 16 GiB
SCIPHiGHSCBCGurobi
Inspect experiment identity