SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2504.13151 · ICML · 2025

MIB: A Mechanistic Interpretability Benchmark

Jing Huang, Sarah Wiegreffe, Atticus Geiger, Yonatan Belinkov, David Bau, Alessandro Stolfo, Michael Hanna, Martin Tutek, Aaron Mueller, Nikhil Prakash, Tal Haklay, Rohan Gupta, and 12 more

arXiv · PDF · Open in the Atlas

Code that ran

3 of 3 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; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 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/2504.13151.json · badge: /api/ran/2504.13151.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
aaronmueller/mib canonical (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
FunctionStatusWhere it lives
get_task_module_and_pipeline Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
aaronmueller/mib/MIB-causal-variable-track/evaluate_submission.py
code served (permissive licence) · get_code("93c6048ec4539a63")
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) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 torch.nn as nn

torch.manual_seed(0)

# Mock the LMPipeline class since we can't import the real one
class MockTokenizer:
    def __init__(self):
        self.padding_side = 'left'
        self.pad_token_id = 0
        self.eos_token_id = 2

class MockLMPipeline:
    def __init__(self, model_path, max_new_tokens=1, device='cpu', dtype=torch.float32):
        self.model_path = model_path
        self.max_new_tokens = max_new_tokens
        self.device = device
        self.dtype = dtype
        self.tokenizer = MockTokenizer()

# Mock the task modules
class MockCausalModel:
    def __init__(self):
        self.name = "mock_causal_model"

def mock_get_causal_model():
    return MockCausalModel()

def mock_get_counterfactual_datasets():
    return []

# Patch the imports before calling the function
import sys
from unittest.mock import MagicMock

# Create mock modules
mock_pipeline_module = MagicMock()
mock_pipeline_module.LMPipeline = MockLMPipeline

mock_simple_mcqa = MagicMock()
mock_simple_mcqa.get_counterfactual_datasets = mock_get_counterfactual_datasets
mock_simple_mcqa.get_causal_model = mock_get_causal_model

mock_arc = MagicMock()
mock_arc.get_counterfactual_datasets = mock_get_counterfactual_datasets
mock_arc.get_causal_model = mock_get_causal_model

mock_arithmetic = MagicMock()
mock_arithmetic.get_counterfactual_datasets = mock_get_counterfactual_datasets
mock_arithmetic.get_causal_model = mock_get_causal_model

mock_ravel = MagicMock()
mock_ravel.get_counterfactual_datasets = mock_get_counterfactual_datasets
mock_ravel.get_causal_model = mock_get_causal_model

# Inject mocks into sys.modules
sys.modules['CausalAbstraction'] = MagicMock()
sys.modules['CausalAbstraction.neural'] = MagicMock()
sys.modules['CausalAbstraction.neural.pipeline'] = mock_pipeline_module
sys.modules['tasks'] = MagicMock()
sys.modules['tasks.simple_MCQA'] = MagicMock()
sys.modules['tasks.simple_MCQA.simple_MCQA'] = mock_simple_mcqa
sys.modules['tasks.ARC'] = MagicMock()
sys.modules['tasks.ARC.ARC'] = mock_arc
sys.modules['tasks.two_digit_addition_task'] = MagicMock()
sys.modules['tasks.two_digit_addition_task.arithmetic'] = mock_arithmetic
sys.modules['tasks.RAVEL'] = MagicMock()
sys.modules['tasks.RAVEL.ravel'] = mock_ravel

ARGS = {
    'task': '4_answer_MCQA',
    'model': 'GPT2LMHeadModel'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
CausalAbstraction, tasks, 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
import_custom_modules Ran
ran; our contract draft was wrong, not the code
aaronmueller/mib/MIB-causal-variable-track/evaluate_submission.py
code served (permissive licence) · get_code("004000eda7de2c91")
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) 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 os
import tempfile
import shutil

# Create a temporary directory to serve as submission_path
temp_dir = tempfile.mkdtemp()

# Create a simple featurizer.py file
featurizer_content = """
def featurize(x):
    return x * 2
"""

featurizer_path = os.path.join(temp_dir, 'featurizer.py')
with open(featurizer_path, 'w') as f:
    f.write(featurizer_content)

# Create a simple token_position.py file
token_position_content = """
def get_token_position(token):
    return 0
"""

token_position_path = os.path.join(temp_dir, 'token_position.py')
with open(token_position_path, 'w') as f:
    f.write(token_position_content)

ARGS = {
    'submission_path': temp_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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_submission_folder Ran
ran; our contract draft was wrong, not the code
aaronmueller/mib/MIB-causal-variable-track/evaluate_submission.py
code served (permissive licence) · get_code("2e86136938ddf889")
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) 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.

# Setup code for parse_submission_folder

ARGS = {
    'folder_name': '4_answer_MCQA_bert_lr_0.001'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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

Also present in other papers' code

Of 3 samples lifted from this paper's linked repositories, 3 carry a function identity and 0 do not. Of the 3 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 3 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

How can we know whether new mechanistic interpretability methods achieve real improvements? In pursuit of lasting evaluation standards, we propose MIB, a Mechanistic Interpretability Benchmark, with two tracks spanning four tasks and five models. MIB favors methods that precisely and concisely recover relevant causal pathways or causal variables in neural language models. The circuit localization track compares methods that locate the model components-and connections between them-most important for performing a task (e.g., attribution patching or information flow routes). The causal variable localization track compares methods that featurize a hidden vector, e.g., sparse autoencoders (SAEs) or distributed alignment search (DAS), and align those features to a task-relevant causal variable. Using MIB, we find that attribution and mask optimization methods perform best on circuit localization. For causal variable localization, we find that the supervised DAS method performs best, while SAE features are not better than neurons, i.e., non-featurized hidden vectors. These findings illustrate that MIB enables meaningful comparisons, and increases our confidence that there has been real progress in the field.

For agents

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

get_harvested_code_for_paper("2504.13151")
get_code_for_paper("2504.13151")
have("2504.13151")

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

curl https://syntology.ai/api/ran/2504.13151.json

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

[![Syntology run record](https://syntology.ai/api/ran/2504.13151.svg)](https://syntology.ai/paper/2504.13151)

Connect an agent — have() is free.