{
  "kind": "ainglish.research.agent-task-benchmark",
  "version": "0.1",
  "frozen_at": "2026-08-28T21:18:10Z",
  "status": "design-only-no-results",
  "source_release": {
    "version": "0.35.0",
    "register_digest": "ee8978f9ab5adb252aa244dc1a0dbb5abaa81f499758ec18c95caf5dcfa863b8",
    "url": "https://ainglish.org/releases/ainglish-core-v0.35.0/register.json"
  },
  "design": {
    "items": 22,
    "constructs": 11,
    "arms": ["bare", "careful", "ainglish"],
    "tracks": ["cold", "one_exposure"],
    "maximum_repair_turns": 1,
    "tasks_canonical_sha256": "e29e18435ed8ab7ac2144a1ce57ca943b2d82c6568b32a8c1bae7935690f5bdf"
  },
  "files": {
    "README.md": "6c196a45630ae86201e5f1f05b99de94769862ebcf179feab2e401a57ff9f964",
    "SCORING.md": "18bcd8bcc507fd5e4be1f0b473a3f3ba426f0d5bf217362b3bed88ffb7337c30",
    "benchmark.py": "f08a9951598612da8408d87b712eed84b518f09c89755d3104b560e6b1101e5a",
    "tasks.json": "9c586c37b66a23abe0252ef793393682e89f6ac3cdcf45acaaa289db049b78c9"
  },
  "claim_boundaries": [
    "This packet contains no model result.",
    "A project-operated run is internal evidence even when reader identifiers differ.",
    "Cold means no definition in the served prompt; it does not establish absence from training data.",
    "One-exposure results are not zero-shot results and are not evidence about a tokenizer adapted for Ainglish.",
    "Task success cannot establish human intuitiveness, external adoption or superiority of Ainglish overall."
  ]
}