rubric 1.2.8

rubric

Rubric: A Python library for LLM-based evaluation using weighted rubrics.

---

Installation

uv add rubric

Usage

1. Set up environment variables:

export OPENAI_API_KEY=your_api_key_here
# Or any other model API key used in your `generate_fn`

2. Run the example below

import asyncio
import os
from openai import AsyncOpenAI
from rubric import Rubric
from rubric.autograders import PerCriterionGrader

# Declare custom generate function with any model and inference provider
async def generate_with_openai(system_prompt: str, user_prompt: str) -> str:
    client = AsyncOpenAI(api_key=os.getenv("OPENAI_API_KEY"))
    response = await client.chat.completions.create(
        model="gpt-5-mini",
        messages=[
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_prompt},
        ],
        max_tokens=400,
        temperature=0.0,
    )
    return response.choices[0].message.content or ""

async def main():
    # Build rubric
    rubric = Rubric.from_dict([
        {"weight": 10.0, "requirement": "States Q4 2023 base margin as 17.2%"},
        {"weight": 8.0, "requirement": "Explicitly uses Shapley attribution for decomposition"},
        {"weight": -15.0, "requirement": "Uses total deliveries instead of cash-only deliveries"}
    ])

    # Select autograder strategy
    grader = PerCriterionGrader(
        generate_fn=generate_with_openai,
        system_prompt="This overrides the default grader system prompt",
    )

    # Grade output
    result = await rubric.grade(
        query="Input query...",
        to_grade="Output to evaluate...",
        autograder=grader
    )

    print(f"Score: {result.score:.2f}")  # Score is 0.0-1.0
    for criterion in result.report:
        print(f"  [{criterion.verdict}] {criterion.requirement}")
        print(f"    → {criterion.reason}")

asyncio.run(main())

Autograder Strategies

PerCriterionGrader

Evaluates each criterion in parallel inference calls.

Scoring Formula:

For each criterion $i$:

• If verdict = MET, contribution = $w_i$
• If verdict = UNMET, contribution = 0

Final score:

$$ \text{score} = \max\left(0, \min\left(1, \frac{\sum_{i=1}^{n} \mathbb{1}[\text{verdict}i = \text{MET}] \cdot w_i}{\sum{i=1}^{n} \max(0, w_i)}\right)\right) $$

Where:

• $w_i$ = weight of criterion $i$
• $\mathbb{1}[\text{verdict}_i = \text{MET}]$ = 1 if criterion is MET, 0 otherwise
• Denominator = $\sum_{i=1}^{n} \max(0, w_i)$ (positive weights only)
• Numerator = sum of weights for MET criteria
• Result clamped to [0, 1]

PerCriterionOneShotGrader

PerCriterionOneShotGrader makes 1 inference call that evaluates all criteria together and returns a structured output, unlike PerCriterionGrader which makes $n$ inference calls.

Scoring Formula:

Same as PerCriterionGrader:

$$ \text{score} = \max\left(0, \min\left(1, \frac{\sum_{i=1}^{n} \mathbb{1}[\text{verdict}i = \text{MET}] \cdot w_i}{\sum{i=1}^{n} \max(0, w_i)}\right)\right) $$

RubricAsJudgeGrader

Holistic evaluation where the model returns a final score directly.

Scoring Formula:

The model is instructed to mentally evaluate all criteria and return a score from 0-100:

$$ \text{score} = \frac{\text{LLM-judged score}}{100} $$

Clamped to [0, 1]. The model is guided to use the same weighted scoring logic, but computes the result in-context rather than aggregating score post-hoc.

Default System Prompts

Each autograder uses a specialized system prompt optimized for its evaluation approach:

PerCriterionGrader - Detailed criterion-by-criterion evaluation with strict JSON formatting requirements. The prompt instructs the LLM to evaluate each criterion independently, handling both positive and negative criteria with specific response formats.

PerCriterionOneShotGrader - Streamlined prompt for evaluating all criteria in a single response. Focuses on providing verdicts (MET/UNMET) and explanations for each criterion in a structured JSON format.

RubricAsJudgeGrader - Holistic evaluation prompt that asks the LLM to consider the output as a whole and provide a single overall score from 0-100, taking into account the weights of all criteria.

You can view the complete default prompts in the source files:

• per_criterion_grader.py
• per_criterion_one_shot_grader.py
• rubric_as_judge_grader.py

Customizing System Prompts: You can override the default system prompt by passing a system_prompt parameter to any autograder:

grader = PerCriterionGrader(
    generate_fn=your_function,
    system_prompt="Your custom system prompt here"
)

Customization

You can customize grading at multiple levels:

1. Custom generate_fn (most common) Pass any function that takes (system_prompt, user_prompt) and returns a string. Use any LLM provider (OpenAI, Anthropic, local models, etc.):

grader = PerCriterionGrader(generate_fn=your_custom_function)

2. Override specific methods Subclass any autograder and override:

• judge() - Orchestrates LLM calls to evaluate criteria and parse responses into structured results
• generate() - Wraps your generate_fn to customize how prompts are sent to the LLM
• aggregate() - Transforms individual criterion results into a final score and optional report

3. Full control Override the entire grade() method for complete end-to-end control over the grading process.

Loading Rubrics

# Direct construction
rubric = Rubric([
    Criterion(weight=10.0, requirement="States Q4 2023 base margin as 17.2%"),
    Criterion(weight=8.0, requirement="Explicitly uses Shapley attribution for decomposition"),
    Criterion(weight=-15.0, requirement="Uses total deliveries instead of cash-only deliveries")
])

# From list of dictionaries
rubric = Rubric.from_dict([
    {"weight": 10.0, "requirement": "States Q4 2023 base margin as 17.2%"},
    {"weight": 8.0, "requirement": "Explicitly uses Shapley attribution for decomposition"},
    {"weight": -15.0, "requirement": "Uses total deliveries instead of cash-only deliveries"}
])

# From JSON string
rubric = Rubric.from_json('[{"weight": 10.0, "requirement": "Example requirement"}]')

# From YAML string
yaml_data = '''
- weight: 10.0
  requirement: "Example requirement"
'''
rubric = Rubric.from_yaml(yaml_data)

# From files
rubric = Rubric.from_file('rubric.json')
rubric = Rubric.from_file('rubric.yaml')

JSON Format

[
  {
    "weight": 10.0,
    "requirement": "States Q4 2023 base margin as 17.2%"
  },
  {
    "weight": 8.0,
    "requirement": "Explicitly uses Shapley attribution for decomposition"
  },
  {
    "weight": -15.0,
    "requirement": "Uses total deliveries instead of cash-only deliveries"
  }
]

YAML Format

- weight: 10.0
  requirement: "States Q4 2023 base margin as 17.2%"
- weight: 8.0
  requirement: "Explicitly uses Shapley attribution for decomposition"
- weight: -15.0
  requirement: "Uses total deliveries instead of cash-only deliveries"

Requirements

• Python 3.10+
• An LLM API (e.g., OpenAI, Anthropic, OpenRouter) - set appropriate API keys as environment variables

License

MIT License - see LICENSE file for details.