SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2405.00823 · 2024

WorkBench: a Benchmark Dataset for Agents in a Realistic Workplace Setting

arXiv · PDF · Open in the Atlas

Code that ran

12 of 13 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, 1 violated it and 0 have no contract verdict; 11 failures of our instrument; 1 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/2405.00823.json · badge: /api/ran/2405.00823.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
olly-styles/workbench canonical (classifier label)
classifier basis: census
12 ran · 1 no recorded run
FunctionStatusWhere it lives
accept_either_chart_end_date Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/evals/evaluation.py
code served (permissive licence) · get_code("0a9a03a51369c346")
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.

ARGS = {
    "actions": [
        'analytics.create_plot(time_max="2023-11-30")',
        'some_other_action',
        'analytics.create_plot(time_max="2023-11-29")',
        'another_action_with_2023-11-30_elsewhere',
        'analytics.create_plot(time_max="2023-11-30", other_param="value")',
    ]
}
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 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
convert_intermediate_step_to_function_call Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/evals/actions.py
code served (permissive licence) · get_code("82d39fede4039848")
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.

# Analyze the function to understand what arguments it needs
# convert_intermediate_step_to_function_call takes:
# - tool_name: str
# - tool_input: dict[str, str]

# The function iterates over tool_input items and formats them as key="value" pairs
# It uses _escape_arg_value which handles string escaping (backslashes, quotes, newlines)

# Construct valid arguments
ARGS = {
    "tool_name": "calculator",
    "tool_input": {
        "operation": "add",
        "x": "5",
        "y": "3"
    }
}
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 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
end_date_minor_error Ran
ran, violated the contract we drafted
olly-styles/workbench/src/evals/evaluation.py
code served (permissive licence) · get_code("d9798b807ed518df")
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 bool, 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 torch
import numpy as np

BENCHMARK_END_DATE = '2023-11-29'
BENCHMARK_END_DATE_OFF_BY_ONE = '2023-11-30'

def end_date_minor_error(ground_truth: list[str], prediction: list[str]) -> bool:
    """Function to check if the end date is off by one day in the prediction

    Parameters
    ----------
    ground_truth : list
        List of ground truth actions as strings.
    prediction : list
        List of predicted actions as strings.

    Returns
    -------
    bool
        True if the end date is off by one day in the prediction.
    """
    if not ground_truth:
        return False
    return all((BENCHMARK_END_DATE in func and func.replace(BENCHMARK_END_DATE, BENCHMARK_END_DATE_OFF_BY_ONE) in prediction for func in ground_truth))


ARGS = {
    "ground_truth": [
        "schedule meeting on 2023-11-29",
        "book appointment 2023-11-29"
    ],
    "prediction": [
        "schedule meeting on 2023-11-30",
        "book appointment 2023-11-30"
    ]
}
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 bool, 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
get_first_free_slot Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/data_generation/data_generation_utils.py
code served (permissive licence) · get_code("dea9fe8105e2dc28")
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 Timestamp. 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 pandas as pd
from datetime import datetime, timedelta

# Create a sample DataFrame with events
events_data = {
    'event_start': [
        pd.Timestamp('2024-01-15 09:30:00'),
        pd.Timestamp('2024-01-15 11:00:00'),
        pd.Timestamp('2024-01-15 14:00:00'),
    ],
    'duration': [30, 60, 45],
}
original_events_on_date = pd.DataFrame(events_data)

# Set up arguments
ARGS = {
    'date': '2024-01-15',
    'original_events_on_date': original_events_on_date,
    'duration_minutes': 60,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Timestamp — a Timestamp output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Timestamp 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_function_name Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/evals/evaluation.py
code served (permissive licence) · get_code("26778cb3a7db5e73")
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

torch.manual_seed(0)

action = "module.submodule.function_name(arg1, arg2)"

ARGS = {
    "action": action
}
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
get_random_future_date Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/data_generation/data_generation_utils.py
code served (permissive licence) · get_code("555b4c7cfe86c224")
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 random
import pandas as pd

HARDCODED_CURRENT_TIME = pd.to_datetime('2023-11-30T00:00:00')

def get_random_future_date(dates: list[str]) -> str:
    date = random.choice(dates)
    while date < str(HARDCODED_CURRENT_TIME).split(' ')[0]:
        date = random.choice(dates)
    return date

# Set seed for reproducibility
random.seed(0)

# Create a list of date strings, some before and some after the hardcoded current time
dates = [
    '2023-11-20',
    '2023-11-25',
    '2023-12-01',
    '2023-12-15',
    '2024-01-10',
    '2024-02-28'
]

ARGS = {
    'dates': dates
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — 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_random_future_datetime Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/data_generation/data_generation_utils.py
code served (permissive licence) · get_code("7c6964aeaeaaa3e5")
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 Timestamp. 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 random
import numpy as np
import pandas as pd

# Set seed for reproducibility
random.seed(0)
np.random.seed(0)

# Create a list of future dates (all after the hardcoded current time of 2023-11-30)
future_dates = [
    '2023-12-01',
    '2023-12-05',
    '2023-12-10',
    '2023-12-15',
    '2023-12-20',
    '2024-01-01',
    '2024-01-15',
    '2024-02-01',
    '2024-03-01',
]

ARGS = {
    'dates': future_dates
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Timestamp — a Timestamp output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Timestamp 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
ground_truth_version_for_results Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/evals/metrics.py
code served (permissive licence) · get_code("47b624708c307a0c")
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 os
import tempfile

# Create a temporary directory for test files
temp_dir = tempfile.mkdtemp()

# Create a results file with .csv extension
results_file_path = os.path.join(temp_dir, 'test_results.csv')
with open(results_file_path, 'w') as f:
    f.write('dummy,results\n1,2\n')

# Create a corresponding _meta.json sidecar with ground_truth_version
meta_file_path = os.path.join(temp_dir, 'test_results_meta.json')
meta_data = {
    'ground_truth_version': 'v2',
    'other_field': 'some_value'
}
with open(meta_file_path, 'w') as f:
    json.dump(meta_data, f)

# Define ARGS with the results_path parameter
ARGS = {
    'results_path': results_file_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 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
meta_path_for_results Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/evals/metrics.py
code served (permissive licence) · get_code("eb3f0fbaf1a421dc")
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

# Construct valid arguments for meta_path_for_results
ARGS = {
    'results_path': 'experiment_results.csv'
}
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
money Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/retro/figs/plot_cost_vs_accuracy.py
code served (permissive licence) · get_code("1ef3e6bd912b699a")
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

torch.manual_seed(0)

ARGS = {
    'value': 42.5678,
    '_': object()
}
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
parse_action Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/evals/agent.py
code served (permissive licence) · get_code("a3db7ab2f37601d6")
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 json
import re

PARSE_ERROR = '__parse_error__'
FINAL_ANSWER = 'Final Answer'

def parse_action(text: str) -> tuple[str, dict[str, str] | str]:
    match = re.search('```(?:json)?\\s*\\n?(.*?)\\n?\\s*```', text, re.DOTALL)
    if not match:
        json_match = re.search('\\{\\s*"action"\\s*:', text, re.DOTALL)
        if json_match:
            start = json_match.start()
            brace_count = 0
            end = start
            for i in range(start, len(text)):
                if text[i] == '{':
                    brace_count += 1
                elif text[i] == '}':
                    brace_count -= 1
                    if brace_count == 0:
                        end = i + 1
                        break
            raw = text[start:end]
        else:
            return (FINAL_ANSWER, text.strip())
    else:
        raw = match.group(1).strip()
    try:
        parsed = json.loads(raw)
    except json.JSONDecodeError as e:
        return (PARSE_ERROR, f'Could not parse your action as JSON ({e}). Respond with a single JSON object with "action" and "action_input" keys.')
    if not isinstance(parsed, dict):
        return (PARSE_ERROR, 'Your action must be a JSON object with "action" and "action_input" keys.')
    action = parsed.get('action', FINAL_ANSWER)
    action_input = parsed.get('action_input', '')
    return (action, action_input)


ARGS = {
    'text': '```json\n{"action": "search", "action_input": "what is python"}\n```'
}
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
strip_results_suffix Ran
ran; our contract draft was wrong, not the code
olly-styles/workbench/src/evals/metrics.py
code served (permissive licence) · get_code("6871aad32f38dd25")
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.

# Setup code for strip_results_suffix function

ARGS = {
    'filename': 'results.csv'
}
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 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
resolve_route Not yet run
unverified (harvested, no recorded run)
olly-styles/workbench/src/evals/agent.py
code served (permissive licence) · get_code("c12489c977a3c127")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 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 13 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

We introduce WorkBench: a benchmark dataset for evaluating agents' ability to execute tasks in a workplace setting. WorkBench contains a sandbox environment with five databases, 26 tools, and 690 tasks. These tasks represent common business activities, such as sending emails and scheduling meetings. The tasks in WorkBench are challenging as they require planning, tool selection, and often multiple actions. If a task has been successfully executed, one (or more) of the database values may change. The correct outcome for each task is unique and unambiguous, which allows for robust, automated evaluation. We call this key contribution outcome-centric evaluation. We evaluate five existing ReAct agents on WorkBench, finding they successfully complete as few as 3% of tasks (Llama2-70B), and just 43% for the best-performing (GPT-4). We further find that agents' errors can result in the wrong action being taken, such as an email being sent to the wrong person. WorkBench reveals weaknesses in agents' ability to undertake common business activities, raising questions about their use in high-stakes workplace settings. WorkBench is publicly available as a free resource at https://github.com/olly-styles/WorkBench.

For agents

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

get_harvested_code_for_paper("2405.00823")
get_code_for_paper("2405.00823")
have("2405.00823")

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

curl https://syntology.ai/api/ran/2405.00823.json

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

[![Syntology run record](https://syntology.ai/api/ran/2405.00823.svg)](https://syntology.ai/paper/2405.00823)

Connect an agent — have() is free.