paperclip-runner evalsstatic evidence

← Overview

Agent requests structured confirmation

request-human-confirmation

The agent creates exactly one durable confirmation interaction.

Prompt

Ask the human for a structured confirmation titled Confirm eval scope with prompt Proceed with the context-before-action slice? Use confirmation kind and no continuation wake.

Fixture

../fixtures/companies/orientation.json

Authority

{
  "actorId": "actor-engineer",
  "capabilities": [],
  "explicitClaims": [],
  "taskId": "task-orientation"
}

Current results

ModelConfigurationResultEstimated costRun time
claude-sonnet-5live-acpx-0.13.1-claude-acp-0.70.0-sonnet-5PASS$0.036017 estimated2026-09-06T21:22:00.355Z
claude-sonnet-5live-claude-managed-agents-2026-04-01-sonnet-5PASS$0.014948 estimated2026-09-06T21:22:25.448Z
global.anthropic.claude-sonnet-4-6live-aws-agentcore-global-anthropic-claude-sonnet-4-6PASS$0.024912 estimated2026-09-06T21:22:25.841Z
gpt-5.4-minilive-codex-pinnedPASS$0.001268 estimated2026-09-06T21:24:04.483Z
gpt-5.6-sollive-acpx-0.13.1-codex-acp-1.6.2-controlPASS$0.007414 estimated2026-09-06T21:22:18.115Z
openrouter/anthropic/claude-sonnet-5live-opencode-1.18.17-openrouter-claude-sonnet-5FAIL$0.049684 estimated2026-09-06T21:24:06.482Z
openrouter/deepseek/deepseek-v4-flash-0731live-opencode-1.18.17-openrouter-deepseek-v4-flash-0731PASS$0.000574 estimated2026-09-06T21:28:34.954Z
openrouter/google/gemini-3.8-flashlive-opencode-1.18.17-openrouter-gemini-3.8-flashPASS$0.021872 estimated2026-09-06T21:24:46.374Z
openrouter/openai/gpt-6-astralive-opencode-1.18.17-openrouter-gpt-6-astraPASS$0.032792 estimated2026-09-06T21:26:10.293Z
openrouter/qwen/qwen3.8-max-0902live-opencode-1.18.17-openrouter-qwen3.8-max-0902PASS$0.033918 estimated2026-09-06T21:27:21.311Z
openrouter/z-ai/glm-5.3live-opencode-1.18.17-openrouter-glm-5.3PASS$0.054221 estimated2026-09-06T21:25:24.802Z

Assertions

request-input
semantic_operation
{
  "id": "request-input",
  "kind": "semantic_operation",
  "maxCalls": 1,
  "minCalls": 1,
  "operationId": "request_human_input",
  "requireSuccess": true
}
one-interaction
state_effect
{
  "collection": "interactions",
  "id": "one-interaction",
  "kind": "state_effect",
  "maximumAdded": 1,
  "minimumAdded": 1
}
terminal-completed
terminal
{
  "id": "terminal-completed",
  "kind": "terminal",
  "status": "completed"
}
transcript-complete
transcript_complete
{
  "id": "transcript-complete",
  "kind": "transcript_complete"
}
mock-only
network
{
  "id": "mock-only",
  "kind": "network",
  "realPaperclipRequests": 0
}
Full authored test definition
{
  "authority": {
    "actorId": "actor-engineer",
    "capabilities": [],
    "explicitClaims": [],
    "taskId": "task-orientation"
  },
  "checks": [
    {
      "id": "request-input",
      "kind": "semantic_operation",
      "maxCalls": 1,
      "minCalls": 1,
      "operationId": "request_human_input",
      "requireSuccess": true
    },
    {
      "collection": "interactions",
      "id": "one-interaction",
      "kind": "state_effect",
      "maximumAdded": 1,
      "minimumAdded": 1
    },
    {
      "id": "terminal-completed",
      "kind": "terminal",
      "status": "completed"
    },
    {
      "id": "transcript-complete",
      "kind": "transcript_complete"
    },
    {
      "id": "mock-only",
      "kind": "network",
      "realPaperclipRequests": 0
    }
  ],
  "description": "The agent creates exactly one durable confirmation interaction.",
  "fixture": "../fixtures/companies/orientation.json",
  "id": "request-human-confirmation",
  "prompt": "Ask the human for a structured confirmation titled Confirm eval scope with prompt Proceed with the context-before-action slice? Use confirmation kind and no continuation wake.",
  "schema": "paperclip-runner/eval-case/v1",
  "title": "Agent requests structured confirmation"
}