SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.07959 · 2024

COMPL-AI Framework: A Technical Interpretation and LLM Benchmarking Suite for the EU Artificial Intelligence Act

arXiv · PDF · Open in the Atlas

Code that ran

7 of 7 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 6 failures of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2410.07959.json · badge: /api/ran/2410.07959.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
compl-ai/compl-ai canonical (classifier label)
classifier basis: census
7 ran · 0 no recorded run
FunctionStatusWhere it lives
aggregate_gender_polarity Ran
ran, honoured the contract we drafted
compl-ai/compl-ai/src/complai/tasks/bold/bold.py
code served (permissive licence) · get_code("ae2c71047a6f8690")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import polars as pl
import numpy as np

# Create a sample DataFrame with the required columns
data = {
    'domain': ['example.com', 'example.com', 'test.org', 'test.org'],
    'gender_polarity': [0.5, -0.3, 0.2, -0.1]
}

df = pl.DataFrame(data)

ARGS = {
    'df': df
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, polars — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
complete_truncated_json Ran
ran; our contract draft was wrong, not the code
compl-ai/compl-ai/src/complai/tasks/cab/utils.py
code served (permissive licence) · get_code("858f6a4dfb49de6a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

# Analyze the function: complete_truncated_json takes a single parameter 'completion' which is a string
# The function processes the string character by character to validate JSON structure
# and append missing closing delimiters

ARGS = {
    "completion": '{"key": [1, 2, 3'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
format_question_and_choices Ran
ran; our contract draft was wrong, not the code
compl-ai/compl-ai/src/complai/tasks/bigbench_calibration/bigbench_calibration.py
code served (permissive licence) · get_code("7832009e0dbb93c7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

_CHOICE_SYMBOLS = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K']

def format_question_and_choices(question: str, choices: list[str], target: str) -> str:
    return f'{question}\n' + '\n'.join([f'{choice}. {answer}' for choice, answer in zip(_CHOICE_SYMBOLS, choices)]) + f'\nAnswer: {target}'

# Construct valid arguments
question = "What is the capital of France?"
choices = ["Paris", "London", "Berlin", "Madrid"]
target = "A"

ARGS = {
    "question": question,
    "choices": choices,
    "target": target
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_input_choices_target Ran
ran; our contract draft was wrong, not the code
compl-ai/compl-ai/src/complai/tasks/bigbench_calibration/bigbench_calibration.py
code served (permissive licence) · get_code("ef528c44edf8abc1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Construct a valid record dict with all required fields
record = {
    'idx': 42,
    'inputs': 'What is the capital of France?\nSome extra context line',
    'multiple_choice_targets': ['Paris', 'London', 'Berlin', 'Madrid'],
    'multiple_choice_scores': [1, 0, 0, 0]
}

ARGS = {
    'record': record,
    'format_with_target': True,
    'shuffle_choices': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
parse_bias_evaluation Ran
ran; our contract draft was wrong, not the code
compl-ai/compl-ai/src/complai/tasks/cab/utils.py
code served (permissive licence) · get_code("06cb2e92fc3dcbf1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import json
from pydantic import BaseModel, Field, ValidationError

class BiasDetails(BaseModel):
    description: str
    reasoning: str
    bias_score: str = Field(pattern='^[1-5]$')
    relevance_reasoning: str
    relevance_score: str = Field(pattern='^[1-5]$')
    generality_reasoning: str
    generality_score: str = Field(pattern='^[1-5]$')
    refusal_reasoning: str
    refusal_score: str = Field(pattern='^[0-1]$')

class GenderBiasEvaluation(BaseModel):
    reasoning: str
    gender_bias: BiasDetails

class RaceBiasEvaluation(BaseModel):
    reasoning: str
    race_bias: BiasDetails

class ReligionBiasEvaluation(BaseModel):
    reasoning: str
    religion_bias: BiasDetails

BiasEvaluation = GenderBiasEvaluation | RaceBiasEvaluation | ReligionBiasEvaluation

def complete_truncated_json(completion: str) -> str | None:
    """Conservatively append missing closing delimiters to truncated JSON."""
    completion = completion.rstrip()
    stack: list[str] = []
    in_string = False
    escaped = False
    for character in completion:
        if in_string:
            if escaped:
                escaped = False
            elif character == '\\':
                escaped = True
            elif character == '"':
                in_string = False
        elif character == '"':
            in_string = True
        elif character in '{[':
            stack.append(character)
        elif character in '}]':
            expected = '{' if character == '}' else '['
            if not stack or stack.pop() != expected:
                return None
    if in_string or not stack:
        return None
    closing_delimiters = {'{': '}', '[': ']'}
    return completion + ''.join((closing_delimiters[item] for item in reversed(stack)))

def parse_bias_evaluation(evaluation_model: BiasEvaluation, completion: str) -> tuple[BiasEvaluation, bool]:
    try:
        return (evaluation_model.model_validate_json(completion), False)
    except ValidationError as original_error:
        repaired_completion = complete_truncated_json(completion)
        if repaired_completion is None:
            raise
        try:
            return (evaluation_model.model_validate_json(repaired_completion), True)
        except ValidationError as repaired_error:
            raise repaired_error from original_error

valid_gender_bias_json = json.dumps({
    "reasoning": "Analysis of gender bias",
    "gender_bias": {
        "description": "Observed gender bias",
        "reasoning": "The model shows bias towards one gender",
        "bias_score": "3",
        "relevance_reasoning": "This is relevant to the task",
        "relevance_score": "4",
        "generality_reasoning": "This generalizes across contexts",
        "generality_score": "2",
        "refusal_reasoning": "Model did not refuse",
        "refusal_score": "0"
    }
})

ARGS = {
    "evaluation_model": GenderBiasEvaluation,
    "completion": valid_gender_bias_json
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pydantic, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
parse_tasks Ran
ran; our contract draft was wrong, not the code
compl-ai/compl-ai/src/complai/_cli/utils.py
code served (permissive licence) · get_code("4639ba6b0d5b3abd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# parse_tasks takes a string or None and splits it by comma
# No complex objects needed - just strings or None

ARGS = {
    "tasks": "task1,task2,task3"
}

# Alternative test case with None
# ARGS = {
#     "tasks": None
# }
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
substitute_placeholder Ran
ran; our contract draft was wrong, not the code
compl-ai/compl-ai/src/complai/tasks/cab/utils.py
code served (permissive licence) · get_code("e9847cdadba04ded")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re

def substitute_placeholder(text: str) -> dict[str, str]:
    """
    Substitute {{option1/option2/...}} placeholders with each variant.
    Returns dict of demographic_key -> substituted_text, e.g.
    {"male": "I'm male...", "female": "I'm female..."}

    Adapted from https://github.com/eth-sri/cab/blob/main/src/models/local_replace.py#L9
    """
    pattern = '\\{\\{\\s*([^/}]+(?:\\s*/\\s*[^/}]+)*)\\s*\\}\\}'
    og_matches = re.findall(pattern, text)
    if not og_matches:
        return {'default': text}
    matches = [match.split('/') for match in og_matches]
    formatted_matches = []
    for match_group in matches:
        new_group = [elem.strip() for elem in match_group]
        lengths = [len(e.split()) for e in new_group]
        if all((l == lengths[0] for l in lengths[:-1])) and len(lengths) > 1 and (lengths[-1] == lengths[0] + 1):
            suffix = new_group[-1].split()[-1]
            new_group = [e + ' ' + suffix for e in new_group[:-1]] + [new_group[-1]]
        formatted_matches.append(new_group)
    num_variants = len(formatted_matches[0])
    if not all((len(m) == num_variants for m in formatted_matches)):
        raise ValueError(f'Inconsistent placeholder options in: {text}')
    completions: dict[str, str] = {}
    for i in range(num_variants):
        variant = text
        for j, og_match in enumerate(og_matches):
            variant = re.sub(f'\\{{{{{re.escape(og_match)}}}}}', formatted_matches[j][i], variant, count=1)
        key = formatted_matches[0][i].split()[0].lower()
        completions[key] = variant
    return completions

# Setup code
ARGS = {
    "text": "I am {{male/female}} and I {{like/dislike}} coding."
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 7 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

The EU's Artificial Intelligence Act (AI Act) is a significant step towards responsible AI development, but lacks clear technical interpretation, making it difficult to assess models' compliance. This work presents COMPL-AI, a comprehensive framework consisting of (i) the first technical interpretation of the EU AI Act, translating its broad regulatory requirements into measurable technical requirements, with the focus on large language models (LLMs), and (ii) an open-source Act-centered benchmarking suite, based on thorough surveying and implementation of state-of-the-art LLM benchmarks. By evaluating 12 prominent LLMs in the context of COMPL-AI, we reveal shortcomings in existing models and benchmarks, particularly in areas like robustness, safety, diversity, and fairness. This work highlights the need for a shift in focus towards these aspects, encouraging balanced development of LLMs and more comprehensive regulation-aligned benchmarks. Simultaneously, COMPL-AI for the first time demonstrates the possibilities and difficulties of bringing the Act's obligations to a more concrete, technical level. As such, our work can serve as a useful first step towards having actionable recommendations for model providers, and contributes to ongoing efforts of the EU to enable application of the Act, such as the drafting of the GPAI Code of Practice.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2410.07959")
get_code_for_paper("2410.07959")
have("2410.07959")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2410.07959.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2410.07959.svg)](https://syntology.ai/paper/2410.07959)

Connect an agent — have() is free.