DolphinBench

Test 125

Jan 1, 2028 / 1 fact

YAML

Request

Create a compact, reusable cohort-cut template for the analytics archive based on the operating pattern we used while the two experiments were live concurrently. Include the two historical experiment lanes, the weekly owner and cadence, and the required summary-plus-raw-data packaging. Make it explicit that this documents a historical operating pattern rather than assigning current work.

Required memory

Fact 335

Owen now prepares separate weekly cohort cuts every Monday for the pricing-page and onboarding live reads; each uses a one-page summary followed by the attached raw data.

Source evidence (1)

000095Mar 14, 2023 / 11:00 UTC-05:00

Owen's splitting it cleanly across both live reads now. Mondays he does the pricing-page weekly cohort cut and the onboarding weekly cut, same format on both -- 1-pager up top, raw attached underneath. i like it better this way tbh. less thrash than ad hoc pulls, and I can just scan the summary first then drop into the file if Marcus starts poking at a segment.

Message 000095 in history

Expected tool calls

  • create_doc

Grading

1. field_equals / create_doc
{
  "type": "field_equals",
  "tool": "create_doc",
  "action_id": "create_historical_cohort_template",
  "path": "result.ok",
  "value": true,
  "check_id": "riley_125_00"
}
2. field_llm_judge / create_doc
{
  "type": "field_llm_judge",
  "tool": "create_doc",
  "action_id": "create_historical_cohort_template",
  "path": "args",
  "criterion": "The reusable template identifies the pricing-page experiment and onboarding experiment as separate historical lanes, with Owen producing a separate cohort cut for each every Monday. Each lane uses a one-page summary followed by the underlying raw data. The document clearly describes the past operating pattern while both experiments were live and does not assign current work.",
  "check_id": "riley_125_01"
}
Complete grading specification
{
  "type": "tool_trace",
  "config": {
    "check_version": 2,
    "today": "2028-01-01",
    "semantic_judge_version": 2,
    "assertions": [
      {
        "type": "field_equals",
        "tool": "create_doc",
        "action_id": "create_historical_cohort_template",
        "path": "result.ok",
        "value": true,
        "check_id": "riley_125_00"
      },
      {
        "type": "field_llm_judge",
        "tool": "create_doc",
        "action_id": "create_historical_cohort_template",
        "path": "args",
        "criterion": "The reusable template identifies the pricing-page experiment and onboarding experiment as separate historical lanes, with Owen producing a separate cohort cut for each every Monday. Each lane uses a one-page summary followed by the underlying raw data. The document clearly describes the past operating pattern while both experiments were live and does not assign current work.",
        "check_id": "riley_125_01"
      }
    ]
  }
}
App stateDownload JSON
Source file

tests/riley/125.yaml

SHA-256: 784553b2e60040c778b06d69b78b0e3de42c57f7ae423b58d06f89a91502fe15