{
  "schema": "local-evidence-selection/v1",
  "frozen_at": "2026-09-23T11:25:24.199573+00:00",
  "dataset_revision": "c104f840cc67f8b6eec6f759ebc8b2693d585d4a",
  "dataset_sha256": "a45b1fe4e2f0c8390b2b2938ac83e92ed5979000856808f3679c07812e9e6dcd",
  "seed": "next-model-20260923-swe-scout-v1",
  "rule": "Sort every Verified test instance by sha256(seed + vertical bar + instance_id), retain first from each repo until five; exclude previously observed smoke instance psf__requests-2317. Selection uses IDs/repo only, no task text or outcomes.",
  "instances": [
    {
      "instance_id": "sphinx-doc__sphinx-9658",
      "repo": "sphinx-doc/sphinx"
    },
    {
      "instance_id": "sympy__sympy-12096",
      "repo": "sympy/sympy"
    },
    {
      "instance_id": "django__django-13512",
      "repo": "django/django"
    },
    {
      "instance_id": "pydata__xarray-4356",
      "repo": "pydata/xarray"
    },
    {
      "instance_id": "pylint-dev__pylint-8898",
      "repo": "pylint-dev/pylint"
    }
  ],
  "profile": "scout",
  "max_steps": 60,
  "max_completion_tokens": 16384,
  "required_comparators": [
    "GLM EXL3 incumbent",
    "GLM Spark",
    "Qwen selected supported reasoning policy"
  ],
  "interpretation": "Small stratified scout, not full SWE-bench score; matched budget, report all outcomes and aggregate agent duration separately from official grading."
}