"""
Test script to reproduce axis label clipping issue #11312.
This script generates synthetic LLM traces with long model names to reproduce
the axis label clipping/overlap issues in the Metrics charts:
1. "Top models by cost" - Y-axis model names clipped
2. "Top models by tokens" - Y-axis model names clipped
3. "Cost" time series - Y-axis label overlaps tick values
Usage:
uv run scripts/test_axis_label_clipping.py
Prerequisites:
- Phoenix running locally at http://localhost:6006
"""
from random import choice, randint
from secrets import token_hex
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk.resources import Resource
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
# Configuration
PHOENIX_ENDPOINT = "http://localhost:6006/v1/traces"
PROJECT_NAME = "axis-label-test"
NUM_TRACES = 25
# Models with varying name lengths to test axis label rendering
# Include both short and long names to demonstrate the clipping issue
MODELS_WITH_PROVIDERS = [
# Short names (should display fine)
("gpt-4", "openai"),
("gpt-4o", "openai"),
("o1-mini", "openai"),
# Medium names
("gpt-4o-mini", "openai"),
("claude-3-opus", "anthropic"),
("gemini-1.5-pro", "google"),
# Long names that cause clipping (from the issue screenshot)
("claude-sonnet-4-5", "anthropic"), # 17 chars - gets clipped
("gpt-5.1-2025-11-13", "openai"), # 18 chars - gets clipped
("claude-opus-4-20250514", "anthropic"), # 22 chars
("gpt-4o-mini-2024-07-18", "openai"), # 22 chars
# Very long names (AWS Bedrock style)
("anthropic.claude-opus-4-20250514-v1:0", "aws"), # 40 chars
("us.anthropic.claude-sonnet-4-20250514-v1:0", "aws"), # 43 chars
("meta.llama4-maverick-17b-instruct-v1:0", "aws"), # 40 chars
]
# Setup tracer
tracer_provider = TracerProvider(resource=Resource({"openinference.project.name": PROJECT_NAME}))
tracer_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(PHOENIX_ENDPOINT)))
tracer = tracer_provider.get_tracer(__name__)
def create_llm_span(model_name: str, provider: str) -> None:
"""Create an LLM span with the given model name and random token counts."""
with tracer.start_as_current_span(f"llm_call_{token_hex(4)}") as span:
span.set_attribute("openinference.span.kind", "LLM")
span.set_attribute("llm.model_name", model_name)
span.set_attribute("llm.provider", provider)
# Random token counts - varying scale to create interesting cost distribution
scale = 10 ** randint(2, 5) # 100 to 100,000
prompt_tokens = randint(10, scale)
completion_tokens = randint(10, scale)
span.set_attribute("llm.token_count.prompt", prompt_tokens)
span.set_attribute("llm.token_count.completion", completion_tokens)
span.set_attribute("llm.token_count.total", prompt_tokens + completion_tokens)
# Add some input/output for realism
span.set_attribute("input.value", "What is machine learning?")
span.set_attribute("output.value", "Machine learning is a subset of AI...")
def main() -> None:
print(f"Generating {NUM_TRACES} traces for project '{PROJECT_NAME}'...")
print(f"Sending to: {PHOENIX_ENDPOINT}")
print()
# Bias toward longer model names to demonstrate the issue
# Weight longer names more heavily
weighted_models = (
MODELS_WITH_PROVIDERS[:3] * 1 # short names: 1x weight
+ MODELS_WITH_PROVIDERS[3:6] * 2 # medium names: 2x weight
+ MODELS_WITH_PROVIDERS[6:10] * 4 # long names: 4x weight
+ MODELS_WITH_PROVIDERS[10:] * 3 # very long names: 3x weight
)
for i in range(NUM_TRACES):
with tracer.start_as_current_span(f"trace_{token_hex(4)}") as root_span:
root_span.set_attribute("session.id", f"session_{randint(1, 5)}")
# Each trace has 1-4 LLM calls
num_llm_calls = randint(1, 4)
for _ in range(num_llm_calls):
model_name, provider = choice(weighted_models)
create_llm_span(model_name, provider)
if (i + 1) % 5 == 0:
print(f" Created {i + 1}/{NUM_TRACES} traces...")
tracer_provider.force_flush()
print()
print("Done! Open Phoenix and navigate to:")
print(" http://localhost:6006/projects/axis-label-test/metrics")
print()
print("Expected issues to observe:")
print(" 1. 'Top models by cost' - long model names clipped on left")
print(" 2. 'Top models by tokens' - long model names clipped on left")
print(" 3. 'Cost' chart - 'Cost (USD)' label may overlap tick values")
if __name__ == "__main__":
main()
