appmerit 0.1.3

AI Testing Framework

Merit

Merit is a Python testing framework for AI projects. It follows pytest syntax and culture while introducing components essential for testing AI software: metrics, typed datasets, semantic predicates (LLM-as-a-Judge), and OTEL traces.

---

Installation

uv add appmerit

---

Merit 101

Follow pytest habits...

• Create 'merit_*.py' files
• Write 'def merit_*' functions
• Use 'merit.resource' instead of 'pytest.fixture'
• Add 'assert' expressions within the functions
• Run 'uv run merit test'

...while leveraging Merit APIs.

• Use 'with metrics()' context to turn failed assertions into quality metrics
• Use 'has_facts()' and other semantic predicates for asserting natural language
• Access OTEL span data and assert it with 'follows_policy()' predicate
• Parse datasets into clearly typed and validated data objects

---

Example

import merit
from merit import Case, Metric, metrics
from merit.predicates import has_unsupported_facts, follows_policy

from pydantic import BaseModel

@merit.sut
def store_chatbot(prompt: str) -> str:
    return call_llm(prompt)

@merit.metric
def accuracy():
    metric = Metric()
    yield metric

    assert metric.mean > 0.8
    yield metric.mean

class Refs(BaseModel):
    kb: str
    expected_tool: str | None = None

cases = [
    Case(sut_input_values={"prompt": "When are you open?"}, references=Refs(kb="Store hours: 9 AM - 6 PM, Monday-Saturday. Closed Sundays.")),
    Case(sut_input_values={"prompt": "Return policy?"}, references=Refs(kb="30-day returns with receipt.")),
    Case(sut_input_values={"prompt": "How much for the Nike Air Max?"}, references=Refs(kb="Nike Air Max: $129.99", expected_tool="offer_product")),
]

@merit.iter_cases(cases)
@merit.repeat(3)
async def merit_chatbot_no_hallucinations(
    case: Case[Refs],
    store_chatbot,
    accuracy: Metric,
    trace_context):
    """AI agent relies on knowledge base and tool calls for transactional questions"""
    response = store_chatbot(**case.sut_input_values)

    # Verify the answer don't have any unsupported facts
    with metrics(accuracy):
        assert not await has_unsupported_facts(response, case.references.kb)

    # Verify tool was called when expected
    if expected_tool := case.references.expected_tool:
        sut_spans = trace_context.get_sut_spans(name="store_chatbot")
        tool_names = [
            s.attributes.get("llm.request.functions.0.name")
            for s in trace_context.get_llm_calls()
            if s.attributes
        ]
        assert expected_tool in tool_names

Run it:

merit test --trace

Use a custom run UUID when you need stable correlation IDs:

merit test --trace --run-id 3f5f5e9a-1c2d-4b5f-9c2b-7f6d8a9b0c1d

Output:

Merit Test Runner
=================

Collected 1 test

test_example.py::merit_chatbot_responds ✓

==================== 1 passed in 0.08s ====================

Documentation

Full documentation: docs.appmerit.com

Getting Started:

• Quick Start - Get up and running in 5 minutes

Usage:

• Writing Merits - How to define a proper merit suite
• Running Merits - How to execute suits and merits

Concepts:

• Merit - Like test but better
• Resource - Like fixtures but better
• Case - Container for parsed dataset entities
• Metric - Aggregating assertions
• Semantic Predicates - Asserting language and logs
• SUT (System Under Test) - Collecting and accesing traces

API Reference:

• Merit Definitions APIs - Tune discovery and execution
• Merit Predicates APIs - Build your own semantic predicates
• Merit Metric APIs - Build complex metric systems
• Merit Tracing APIs - OpenTelemetry integration

---

Contributing

We welcome contributions! To get started:

1. Fork the repository
2. Clone your fork: git clone https://github.com/YOUR_USERNAME/merit.git
3. Create a branch: git checkout -b your-feature-name
4. Install dependencies: uv sync
5. Make your changes
6. Run tests: uv run merit test
7. Run lints: uv run ruff check .
8. Submit a pull request

For more details, see CONTRIBUTING.md.

Development Setup:

# Clone the repository
git clone https://github.com/appMerit/merit.git
cd merit

# Install dependencies
uv sync

# Run tests
uv run merit test

# Run lints
uv run ruff check .
uv run mypy .

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Support

• Documentation: docs.appmerit.com
• GitHub Issues: github.com/appMerit/merit/issues
• Email: [e-mail]