Datasets/Support agent/traces/2026-09-14

support-agent-prod / 2026-08-25

Production support-agent traces, graded. Analyzer regime TRAIN.

Rows
32
Size
14 KB
Support
0.78
Pass
72%
Groups
–
Regime
TRAIN

Get it

pip install zeroproof
import zeroproof.simulations as zps

rows = zps.pull("ds_550a2b3ba46aac2b")                        # list of dicts, no key
zps.pull("ds_550a2b3ba46aac2b", path="ds_550a2b3ba46aac2b.jsonl")   # or a file
curl -s https://api.zeroproofai.com/catalog/ds_550a2b3ba46aac2b/download
# {"parts": ["https://…", …]}  fetch each part, concatenate the JSONL

Sample rows first 3, long fields cut

{
  "scenario_id": "damaged-item",
  "prompt": "The mug arrived cracked.",
  "final_text": "Please email support@.",
  "info": {
    "scenario_id": "damaged-item",
    "model": "gpt-4o",
    "agent": "support-bot",
    "scope": "opentelemetry.instrumentation.openai",
    "input_tokens": 194,
    "output_tokens": 54,
    "trace_id": "0000000000000000000000000abc000e"
  },
  "reward": 0,
  "tool_trace": [
    {
      "tool": "start_return",
      "input": "{\"order\": 4414}",
      "output": "{\"ok\": false}"
    }
  ]
}
{
  "scenario_id": "damaged-item",
  "prompt": "The mug arrived cracked.",
  "final_text": "Can you send a photo?",
  "info": {
    "scenario_id": "damaged-item",
    "model": "gpt-4o",
    "agent": "support-bot",
    "scope": "opentelemetry.instrumentation.openai",
    "input_tokens": 195,
    "output_tokens": 55,
    "trace_id": "0000000000000000000000000abc000f"
  },
  "reward": 0,
  "tool_trace": [
    {
      "tool": "start_return",
      "input": "{\"order\": 4415}",
      "output": "{\"ok\": false}"
    }
  ]
}
{
  "scenario_id": "change-address",
  "prompt": "Ship order 4490 to my work address instead.",
  "final_text": "Updated to 12 Grand Ave.",
  "info": {
    "scenario_id": "change-address",
    "model": "gpt-4o",
    "agent": "support-bot",
    "scope": "opentelemetry.instrumentation.openai",
    "input_tokens": 196,
    "output_tokens": 56,
    "trace_id": "0000000000000000000000000abc0010"
  },
  "reward": 1,
  "tool_trace": [
    {
      "tool": "update_address",
      "input": "{\"order\": 4416}",
      "output": "{\"ok\": true}"
    }
  ]
}

Analyzer report

{
  "rows": 32,
  "prompts": 8,
  "support": 0.7812,
  "passRate": 0.7188,
  "regime": "TRAIN"
}

rows follow Task / Rollout / Judgment / Marker·how a row gets made