The concept of success or failure is crucial for AI agent observability. Atla Insights provides explicit functions to mark trace outcomes, enabling better error analysis and performance tracking.

​
Usage

Use mark_success() and mark_failure() within instrumented functions to indicate trace outcomes: 
from atla_insights import (
    configure,
    instrument,
    instrument_openai,
    mark_failure,
    mark_success,
)
from openai import OpenAI

configure(token="<MY_ATLA_INSIGHTS_TOKEN>")
instrument_openai()

client = OpenAI()

@instrument("Math solving agent")
def solve_math_problem(problem: str) -> str:
    result = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {
                "role": "user",
                "content": f"Solve this math problem: {problem}. Reply with only the answer.",
            }
        ]
    )
    response = result.choices[0].message.content

    # Custom success criteria
    if is_correct_answer(response, problem):
        mark_success()
    else:
        mark_failure()
    
    return response

def is_correct_answer(response: str, problem: str) -> bool:
    # Your validation logic here
    if "2 + 2" in problem:
        return "4" in response
    # Add more validation logic
    return True
Use mark_success() and mark_failure() within an instrumented function (decorated with @instrument).