DolphinBench

Test 161

Jan 1, 2028 / 4 facts

YAML

Request

Create a concise incident case study for the 2028 reliability-training library about the first live pre-breach source-age alert. Cover what stalled, the warning and restoration timing, the affected pending decisions, the fresh reevaluation outcomes, and whether any decision became stale or required manual release, rollback, or customer correction. Use a clear title and brief, practical sections suitable for training.

Required memory

Fact 123

A qualifying-human-admin source refresh stalled with 17 pending Customer Success auto-nudge decisions, and the source-age alert fired 43 minutes before the freshness limit.

Source evidence (1)

003236Aug 14, 2025 / 08:50 UTC-05:00

A qualifying-human-admin source refresh stalled with 17 pending CS auto-nudge decisions. The source-age alert fired 43 minutes before the freshness limit, Daniela restored the refresh with 18 minutes remaining, and Owen verified the fresh reevaluation: 13 proceeded and four were suppressed because current Customer Success adoption or renewal work was active. None entered stale state, nobody manually released a decision, and no customer correction or rollback was required. Update the Q3 evidence-lock template with this history and post a bounded summary to #lifecycle-systems describing it as the alert’s first live operational result, not mature reliability evidence.

Message 003236 in history

Fact 124

Daniela restored the qualifying-human-admin source refresh with 18 minutes remaining before the freshness limit.

Source evidence (1)

003236Aug 14, 2025 / 08:50 UTC-05:00

A qualifying-human-admin source refresh stalled with 17 pending CS auto-nudge decisions. The source-age alert fired 43 minutes before the freshness limit, Daniela restored the refresh with 18 minutes remaining, and Owen verified the fresh reevaluation: 13 proceeded and four were suppressed because current Customer Success adoption or renewal work was active. None entered stale state, nobody manually released a decision, and no customer correction or rollback was required. Update the Q3 evidence-lock template with this history and post a bounded summary to #lifecycle-systems describing it as the alert’s first live operational result, not mature reliability evidence.

Message 003236 in history

Fact 125

Owen verified the fresh reevaluation of the 17 Customer Success auto-nudge decisions: 13 proceeded, while four were suppressed because current Customer Success adoption or renewal work was active.

Source evidence (1)

003236Aug 14, 2025 / 08:50 UTC-05:00

A qualifying-human-admin source refresh stalled with 17 pending CS auto-nudge decisions. The source-age alert fired 43 minutes before the freshness limit, Daniela restored the refresh with 18 minutes remaining, and Owen verified the fresh reevaluation: 13 proceeded and four were suppressed because current Customer Success adoption or renewal work was active. None entered stale state, nobody manually released a decision, and no customer correction or rollback was required. Update the Q3 evidence-lock template with this history and post a bounded summary to #lifecycle-systems describing it as the alert’s first live operational result, not mature reliability evidence.

Message 003236 in history

Fact 126

No Customer Success auto-nudge decision entered a stale state or was manually released, and the incident required no customer correction or rollback.

Source evidence (1)

003236Aug 14, 2025 / 08:50 UTC-05:00

A qualifying-human-admin source refresh stalled with 17 pending CS auto-nudge decisions. The source-age alert fired 43 minutes before the freshness limit, Daniela restored the refresh with 18 minutes remaining, and Owen verified the fresh reevaluation: 13 proceeded and four were suppressed because current Customer Success adoption or renewal work was active. None entered stale state, nobody manually released a decision, and no customer correction or rollback was required. Update the Q3 evidence-lock template with this history and post a bounded summary to #lifecycle-systems describing it as the alert’s first live operational result, not mature reliability evidence.

Message 003236 in history

Expected tool calls

  • create_doc

Grading

1. field_equals / create_doc
{
  "type": "field_equals",
  "tool": "create_doc",
  "action_id": "create_reliability_case_study",
  "path": "result.ok",
  "value": true,
  "check_id": "riley_161_00"
}
2. field_llm_judge / create_doc
{
  "type": "field_llm_judge",
  "tool": "create_doc",
  "action_id": "create_reliability_case_study",
  "path": "args",
  "criterion": "The case study accurately identifies a stalled qualifying-human-admin source refresh affecting 17 pending CS auto-nudge decisions, says the alert fired 43 minutes before the freshness limit, and says Daniela restored the refresh with 18 minutes remaining. Missing or contradictory timing, count, source, or restoration information must fail.\n\nThe case study states that a fresh reevaluation occurred and accurately reports that 13 decisions proceeded while four were suppressed because current Customer Success adoption or renewal work was active. Missing, reversed, or contradictory dispositions must fail.\n\nThe case study clearly states that none of the affected decisions entered stale state, nobody manually released a decision, and neither a customer correction nor a rollback was required. Any claim that one of these events occurred must fail.\n\nThe document title clearly identifies this as an incident case study about the first live pre-breach source-age alert. A generic, irrelevant, or misleading title must fail.",
  "check_id": "riley_161_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_reliability_case_study",
        "path": "result.ok",
        "value": true,
        "check_id": "riley_161_00"
      },
      {
        "type": "field_llm_judge",
        "tool": "create_doc",
        "action_id": "create_reliability_case_study",
        "path": "args",
        "criterion": "The case study accurately identifies a stalled qualifying-human-admin source refresh affecting 17 pending CS auto-nudge decisions, says the alert fired 43 minutes before the freshness limit, and says Daniela restored the refresh with 18 minutes remaining. Missing or contradictory timing, count, source, or restoration information must fail.\n\nThe case study states that a fresh reevaluation occurred and accurately reports that 13 decisions proceeded while four were suppressed because current Customer Success adoption or renewal work was active. Missing, reversed, or contradictory dispositions must fail.\n\nThe case study clearly states that none of the affected decisions entered stale state, nobody manually released a decision, and neither a customer correction nor a rollback was required. Any claim that one of these events occurred must fail.\n\nThe document title clearly identifies this as an incident case study about the first live pre-breach source-age alert. A generic, irrelevant, or misleading title must fail.",
        "check_id": "riley_161_01"
      }
    ]
  }
}
App stateDownload JSON
Source file

tests/riley/161.yaml

SHA-256: 3c9192b06ab5af6c696f6653a24b8701484afe786fabbf521f6da9c4f5b9cfad