[Bug] Evaluation - Unable to run evaluations when an agent flow has a null total_cost

[jbueza-railtownai]

Description

1. Use the eval.py below
2. Run it uv run eval.py
3. select an agent from the list
4. Notice t will stall and then fail when an agent session contains an agent with null as total_cost

Select agent index(es) (comma-separated), or -1 to evaluate all: Traceback (most recent call last):
  File "C:\Users\Jaime\Code\github\railtracks-visualizer\eval.py", line 29, in <module>
    results = evals.evaluate(
        data=agent_data,
    ...<3 lines>...
        payload_callback=upload_agent_evaluation,
    )
  File "C:\Users\Jaime\Code\github\railtracks-visualizer\.venv\Lib\site-packages\railtracks\evaluations\runners\_evaluate.py", line 157, in evaluate
    data_dict, agents = _setup_agent_data(data, agent_selection, agents)
                        ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Jaime\Code\github\railtracks-visualizer\.venv\Lib\site-packages\railtracks\evaluations\runners\_evaluate.py", line 125, in _setup_agent_data
    agents = _select_agent(
        {agent_name: len(data_dict[agent_name]) for agent_name in data_dict.keys()}
    )
  File "C:\Users\Jaime\Code\github\railtracks-visualizer\.venv\Lib\site-packages\railtracks\evaluations\runners\_evaluate.py", line 37, in _select_agent
    user_input = Prompt.ask(
        f"\n[{COLORS['prompt']}]Select agent index(es)[/{COLORS['prompt']}] (comma-separated), or [{COLORS['highlight']}]-1[/{COLORS['highlight']}] to evaluate all"
    )
  File "C:\Users\Jaime\Code\github\railtracks-visualizer\.venv\Lib\site-packages\rich\prompt.py", line 141, in ask
    return _prompt(default=default, stream=stream)
  File "C:\Users\Jaime\Code\github\railtracks-visualizer\.venv\Lib\site-packages\rich\prompt.py", line 274, in __call__
    value = self.get_input(self.console, prompt, self.password, stream=stream)
  File "C:\Users\Jaime\Code\github\railtracks-visualizer\.venv\Lib\site-packages\rich\prompt.py", line 203, in get_input
    return console.input(prompt, password=password, stream=stream)
           ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Jaime\Code\github\railtracks-visualizer\.venv\Lib\site-packages\rich\console.py", line 2123, in input
    result = input()
KeyboardInterrupt

Reproducing Steps (include code examples if applicable)

use the following eval.py

import railtracks as rt
from railtracks import evaluations as evals
from railtownai import upload_agent_evaluation

# Set EVALUATIONS_API_TOKEN in .env (from Conductr Agent Evaluations onboarding)

agent_data = evals.extract_agent_data_points(".railtracks/data/sessions/")

judge = evals.JudgeEvaluator(
    llm=rt.llm.OpenAILLM(model_name="gpt-5.2"),
    metrics=[
        evals.metrics.Categorical(
            name="Helpfullness",
            categories=["Not Helpful", "Somewhat Helpful", "Very Helpful"],
        ),
        evals.metrics.Categorical(
            name="Efficiency",
            description=(
                "How efficiently did the agent complete the task? "
                "Consider factors such as speed, resource usage, and overall effectiveness."
            ),
            categories=["Not Efficient", "Somewhat Efficient", "Very Efficient"],
        ),
    ],
    reasoning=True,
    timeout=1000,
)

results = evals.evaluate(
    data=agent_data,
    # agents=["Job Header Extraction Agent"],
    evaluators=[evals.LLMInferenceEvaluator(), evals.ToolUseEvaluator(), judge],
    name="Small Dataset Evaluation",
    payload_callback=upload_agent_evaluation,
)

Additional Context

No response

Operating System

No response

railtracks Version

No response

Python Version

No response