Geospatial agent benchmark · adapted

Axis Synthetic Migration Cases

Also known as: Axis Synthetic Migration Cases

Evaluates: complete agent systems

The synthetic cases can exercise migration-code and guardrail verification.

At a glance

3Tasks
YesAgents run real code
NoData + code public
YesChecked by recomputation
Axis original syntheticData licence

Sources

every claim on this page traces to these records

In shortAgents run real code inside this benchmark, and answers are checked by recomputation, not by another model's opinion. The underlying data and code are not published at a verified link. Broad migration competence, production readiness, or superiority of one configuration.

Evidence boundaries

what a score on this benchmark does and does not tell you

A good score shows

The synthetic cases can exercise migration-code and guardrail verification.

A good score does not show

Broad migration competence, production readiness, or superiority of one configuration.

Arena status

adapted — Only provenance-clear original-synthetic cases may enter a new study; historical results are context, not current ranking truth.

All recorded facts (22 fields)
Evaluates
complete-system
Task families
migration code generation,guardrail compliance,platform-specific deployment
Task count
3
Difficulty
mixed (T1 single-transform through T3 multi-script site suitability with AWS platform constraints)
Geography
original synthetic, non-client-specific
Modality
mixed
Input types
synthetic migration requirements,synthetic source examples
Output types
generated Python code with guardrail checks
Execution environment
code generation with automated syntax and guardrail checks
Ground truth method
AST parse success, guardrail scan pass, prohibited-import absence, dry-run guard presence, and portable-path usage
Verifier method
deterministic automated code checks
Deterministic verification
yes
Scoring dimensions
syntax,guardrails,prohibited imports,dry-run safety,portable paths
Aggregation formula
binary verifier outcomes per case and trial; aggregation must be preregistered
Trials policy
historical evidence only; pinned repeated trials required
Variance reporting
not yet sufficient for a public comparative claim
Data licence
Axis original synthetic; publication rights must be recorded per task
Code licence
Axis original synthetic; publication rights must be recorded per task
Contamination concerns
exclude all historical or uncertain-rights material
Reproducibility status
candidate for pinned reproduction
First published
2026-04
Latest known update
2026-04-02
where this benchmark overlaps Arena's own task pack

Machine-readable record

This entry is available as JSON at /api/registry#axis-migration-benchmark. See the registry endpoint.