Skip to content

[FEATURE] Establish AURA's SWE-bench non-degradation baseline #443

Description

@charlesjohnson

Outcome

Determine whether AURA preserves model capability on long-horizon software-engineering tasks.

Scope

  • Run AURA through a current SWE-bench Verified-compatible harness.
  • Compare the same model and materially equivalent settings with and without AURA.
  • Attribute failures and latency to the model, AURA, tools, or harness where possible.
  • Decide whether the result supports a public claim.

Complete when

The paired methodology and results are reproducible, the delta is explained, and the publication decision is recorded.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions