SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2312.02405 · NeurIPS · 2023

BEDD: The MineRL BASALT Evaluation and Demonstrations Dataset for Training and Benchmarking Agents that Solve Fuzzy Tasks

Stephanie Milani, Brandon Houghton, Anssi Kanervisto, Karolis Ramanauskas, Sander Schulhoff

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 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, 2 violated it and 0 have no contract verdict; 8 failures of our instrument; 1 no recorded run), measured 2026-11-22; 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/2312.02405.json · badge: /api/ran/2312.02405.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
minerllabs/basalt-benchmark canonical (classifier label)
classifier basis: census
10 ran · 1 no recorded run
FunctionStatusWhere it lives
create_json_entry_dict Ran
ran; our contract draft was wrong, not the code
minerllabs/basalt-benchmark/basalt/rollout_model.py
code served (permissive licence) · get_code("4459e00c3d091105")
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 json
import numpy as np

# Create observation dictionary with all required keys
obs = {
    'equipped_items': {'item_1': np.array(5), 'item_2': np.array(0)},
    'life_stats': {'health': np.array(20), 'hunger': np.array(0)},
    'location_stats': {'x': np.array(100.5), 'y': np.array(64.0), 'z': np.array(200.3)},
    'use_item': {'count': np.array(3)},
    'drop': {'item_dropped': np.array(1), 'unused': np.array(0)},
    'pickup': {'item_picked': np.array(2)},
    'break_item': {'durability_lost': np.array(0)},
    'craft_item': {'crafted_count': np.array(4)},
    'mine_block': {'blocks_mined': np.array(7)},
    'damage_dealt': {'damage': np.array(5.5)},
    'entity_killed_by': {'entity_type': 'zombie'},
    'kill_entity': {'killed_count': np.array(2)},
    'full_stats': {'total_time': np.array(1500), 'level': np.array(0)}
}

# Create action dictionary
action = {
    'action_type': np.array(1),
    'action_value': np.array([10, 20, 30, 40]),
    'unused_field': np.array(0)
}

ARGS = {
    'obs': obs,
    'action': action
}
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 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_band_diagonal_mask Ran
ran, violated the contract we drafted
minerllabs/basalt-benchmark/basalt/vpt_lib/masked_attention.py
code served (permissive licence) · get_code("2bcf842598e9b062")
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 Tensor, ndim 3, dtype family float, shape (2, 3, 6) 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 torch as th

th.manual_seed(0)

ARGS = {
    "t": 3,
    "T": 6,
    "maxlen": 3,
    "batchsize": 2,
    "device": th.device("cpu"),
}
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 Tensor, ndim 3, dtype family float, shape (2, 3, 6) — shape (2, 3, 6) 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
get_mask Ran
ran; our contract draft was wrong, not the code
minerllabs/basalt-benchmark/basalt/vpt_lib/masked_attention.py
code served (permissive licence) · get_code("eb197d6a76f63639")
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 torch as th
import torch.nn as nn

th.manual_seed(0)

# Define small distinct dimensions for clarity
batchsize = 2
t = 3  # rows (frames receiving gradient)
T = 5  # cols (t + past context)
maxlen = 4  # maximum frames any frame can attend to
heads = 2  # number of attention heads
device = th.device("cpu")

# Create first_b11: boolean tensor [batchsize, 1, 1]
# Indicates if the first timestep for each batch element had first=True
first_b11 = th.tensor([[[True]], [[False]]], dtype=th.bool)

# Create state_mask: [batchsize, 1, T - t]
# Mask tensor for the past T - t frames
state_mask = th.ones((batchsize, 1, T - t), dtype=th.bool, device=device)

ARGS = {
    "first_b11": first_b11,
    "state_mask": state_mask,
    "t": t,
    "T": T,
    "maxlen": maxlen,
    "heads": heads,
    "device": device,
}
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 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
intprod Ran
ran, violated the contract we drafted
minerllabs/basalt-benchmark/basalt/vpt_lib/misc.py
code served (permissive licence) · get_code("5234c0dfb707f813")
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 int, ndim 0, dtype family int, 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 torch

torch.manual_seed(0)

# intprod takes a sequence of integers and returns their product
# xs: a sequence (list, tuple, etc.) of integers

ARGS = {
    "xs": [2, 3, 4, 5]
}
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 int, ndim 0, dtype family int, 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
load_model_parameters Ran
ran; our contract draft was wrong, not the code
minerllabs/basalt-benchmark/basalt/embed_trajectories.py
code served (permissive licence) · get_code("b5f9bf2f6cbd78b7")
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 pickle
import tempfile
import os

# Create a temporary file with pickled model parameters
agent_parameters = {
    'model': {
        'args': {
            'net': {
                'args': {
                    'hidden_dim': 128,
                    'num_layers': 2,
                    'activation': 'relu'
                }
            },
            'pi_head_opts': {
                'temperature': '0.5',
                'output_dim': 64
            }
        }
    }
}

# Write to a temporary pickle file
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.pkl')
temp_path = temp_file.name
temp_file.close()

with open(temp_path, 'wb') as f:
    pickle.dump(agent_parameters, f)

ARGS = {
    'path_to_model_file': temp_path
}
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 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
remove_numpyness_and_remove_zeros Ran
ran; our contract draft was wrong, not the code
minerllabs/basalt-benchmark/basalt/rollout_model.py
code served (permissive licence) · get_code("abb7428785776805")
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 numpy as np

# Construct test data with nested dicts containing numpy arrays and scalar values
dict_with_numpy_arrays = {
    "layer1": {
        "weights": np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]),
        "bias": np.array([0.5]),
        "zero_value": np.array([0]),
    },
    "layer2": {
        "params": np.array([7.0, 8.0, 9.0, 10.0]),
        "scale": np.array([2.0]),
    },
    "metadata": {
        "name": "test_model",
        "count": np.array([5]),
        "empty": np.array([0.0]),
    },
}

ARGS = {
    "dict_with_numpy_arrays": dict_with_numpy_arrays,
}
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 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
resize_image Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
minerllabs/basalt-benchmark/basalt/vpt_lib/agent.py
code served (permissive licence) · get_code("69a190b789185068")
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 ndarray, ndim 3, dtype family int, shape (48, 64, 3) 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 cv2
import numpy as np

# Create a small test image (height=100, width=80, channels=3 for RGB)
img = np.random.randint(0, 256, (100, 80, 3), dtype=np.uint8)

# Define target resolution as (width, height) tuple for cv2.resize
target_resolution = (64, 48)

ARGS = {
    "img": img,
    "target_resolution": target_resolution
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (48, 64, 3) — shape (48, 64, 3) 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
store_args Ran
ran; our contract draft was wrong, not the code
minerllabs/basalt-benchmark/basalt/vpt_lib/minecraft_util.py
pointer only (licence: MIT); the stored copy get_code serves, from craftjarvis/mc-planner, is pointer only (licence: NONE) · get_code("e9f218ee05e29fad")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from craftjarvis/mc-planner, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

transpose Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
minerllabs/basalt-benchmark/basalt/vpt_lib/misc.py
code served (permissive licence) · get_code("6e2674f0120e0c29")
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 Tensor, ndim 3, dtype family float, shape (3, 4, 2) 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 torch

torch.manual_seed(0)

# The function transposes a tensor by permuting dimensions according to string specifications
# x: a tensor with ndim dimensions
# before: string specifying current dimension order
# after: string specifying desired dimension order

# Create a small tensor with distinct dimension sizes for probing
# Using shape (2, 3, 4) for dimensions a, b, c respectively
x = torch.randn(2, 3, 4)

# Define the dimension ordering strings
before = 'abc'
after = 'bca'

ARGS = {
    'x': x,
    'before': before,
    'after': after
}
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 Tensor, ndim 3, dtype family float, shape (3, 4, 2) — shape (3, 4, 2) 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
transpose_undo Ran
ran; our contract draft was wrong, not the code
minerllabs/basalt-benchmark/basalt/vpt_lib/misc.py
code served (permissive licence) · get_code("5760c363cbb2cc44")
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 torch
import torch.nn as nn

torch.manual_seed(0)

# Create a small tensor with distinct dimension sizes for probing
# Shape: (2, 3, 4) for dimensions a, b, c
x = torch.randn(2, 3, 4)

# Parameters for transpose_undo
before = 'abc'
after = 'bca'
undo = None  # Can be None as shown in the docstring example

ARGS = {
    'x': x,
    'before': before,
    'after': after,
    'undo': undo
}
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 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
get_norm_entropy_from_cat_head Not yet run
unverified (harvested, no recorded run)
minerllabs/basalt-benchmark/basalt/vpt_lib/minecraft_util.py
code served (permissive licence) · get_code("46d393af0ba227d6")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 8 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 0 the same month, 6 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.

FunctionAlso present inEarliest-datedOthers named
store_args 4 other papers
3 dated earlier, 1 later
2005.03420 (2020-05)
Curious Hierarchical Actor-Critic Reinforcement Learning
2301.10034 (2023-01)
Open-World Multi-Task Control Through Goal-Aware Representation Learning and Adaptive Horizon Prediction
2302.05103 (2023-02)
Controllability-Aware Unsupervised Skill Discovery
and 1 more not named here
intprod 2 other papers
1 dated earlier, 1 later
2308.00398 (2023-08)
DriveAdapter: Breaking the Coupling Barrier of Perception and Planning in End-to-End Autonomous Driving
2403.12037 (2024-03)
MineDreamer: Learning to Follow Instructions via Chain-of-Imagination for Simulated-World Control
load_model_parameters 2 other papers
2 dated earlier
2206.11795 (2022-06)
Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos
2303.13512 (2023-03)
Towards Solving Fuzzy Tasks with Human Feedback: A Retrospective of the MineRL BASALT 2022 Competition
get_band_diagonal_mask 1 other paper
1 later
2403.12037 (2024-03)
MineDreamer: Learning to Follow Instructions via Chain-of-Imagination for Simulated-World Control
get_norm_entropy_from_cat_head 1 other paper
1 later
2403.12037 (2024-03)
MineDreamer: Learning to Follow Instructions via Chain-of-Imagination for Simulated-World Control
resize_image 1 other paper
1 dated earlier
2303.13512 (2023-03)
Towards Solving Fuzzy Tasks with Human Feedback: A Retrospective of the MineRL BASALT 2022 Competition
transpose 1 other paper
1 later
2403.12037 (2024-03)
MineDreamer: Learning to Follow Instructions via Chain-of-Imagination for Simulated-World Control
transpose_undo 1 other paper
1 later
2403.12037 (2024-03)
MineDreamer: Learning to Follow Instructions via Chain-of-Imagination for Simulated-World Control

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 MineRL BASALT competition has served to catalyze advances in learning from human feedback through four hard-to-specify tasks in Minecraft, such as create and photograph a waterfall. Given the completion of two years of BASALT competitions, we offer to the community a formalized benchmark through the BASALT Evaluation and Demonstrations Dataset (BEDD), which serves as a resource for algorithm development and performance assessment. BEDD consists of a collection of 26 million image-action pairs from nearly 14,000 videos of human players completing the BASALT tasks in Minecraft. It also includes over 3,000 dense pairwise human evaluations of human and algorithmic agents. These comparisons serve as a fixed, preliminary leaderboard for evaluating newly-developed algorithms. To enable this comparison, we present a streamlined codebase for benchmarking new algorithms against the leaderboard. In addition to presenting these datasets, we conduct a detailed analysis of the data from both datasets to guide algorithm development and evaluation. The released code and data are available at https://github.com/minerllabs/basalt-benchmark.

For agents

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

get_harvested_code_for_paper("2312.02405")
get_code_for_paper("2312.02405")
have("2312.02405")

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

curl https://syntology.ai/api/ran/2312.02405.json

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

[![Syntology run record](https://syntology.ai/api/ran/2312.02405.svg)](https://syntology.ai/paper/2312.02405)

Connect an agent — have() is free.