SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.05764 · 2026

TML-bench: Benchmark for Data Science Agents on Tabular ML Tasks

Mykola Pinchuk

arXiv · PDF · Open in the Atlas

Code that ran

6 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; 5 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/2603.05764.json · badge: /api/ran/2603.05764.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
MykolaPinchuk/TML-bench canonical (classifier label)
classifier basis: census
6 ran · 1 no recorded run
FunctionStatusWhere it lives
fetch_baselines Ran
ran, honoured the contract we drafted
MykolaPinchuk/TML-bench/orchestrator/db.py
code served (permissive licence) · get_code("d90a9ca84ee94456")
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 tempfile
from pathlib import Path

# Create a temporary directory for the database
temp_dir = tempfile.mkdtemp()
db_path = Path(temp_dir) / "test.db"

ARGS = {
    "db_path": db_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 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
fetch_runs Ran
ran; our contract draft was wrong, not the code
MykolaPinchuk/TML-bench/orchestrator/db.py
code served (permissive licence) · get_code("73014115df7ec4bb")
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 sqlite3
import tempfile
from pathlib import Path

# Create a temporary SQLite database with a runs table
temp_db = tempfile.NamedTemporaryFile(delete=False, suffix='.db')
db_path = temp_db.name
temp_db.close()

# Initialize the database with a runs table
con = sqlite3.connect(db_path)
con.execute('''
    CREATE TABLE runs (
        id INTEGER PRIMARY KEY,
        competition_id TEXT,
        created_at TIMESTAMP,
        name TEXT
    )
''')

# Insert some test data
con.execute(
    "INSERT INTO runs (competition_id, created_at, name) VALUES (?, ?, ?)",
    ('comp_1', '2024-01-02 10:00:00', 'run_2')
)
con.execute(
    "INSERT INTO runs (competition_id, created_at, name) VALUES (?, ?, ?)",
    ('comp_1', '2024-01-01 10:00:00', 'run_1')
)
con.execute(
    "INSERT INTO runs (competition_id, created_at, name) VALUES (?, ?, ?)",
    ('comp_2', '2024-01-03 10:00:00', 'run_3')
)
con.commit()
con.close()

ARGS = {
    'db_path': db_path,
    'competition_id': 'comp_1'
}
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
sha256_bytes Ran
ran; our contract draft was wrong, not the code
MykolaPinchuk/TML-bench/orchestrator/provenance.py
code served (permissive licence) · get_code("7f5ffed1f98ca2b1")
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 hashlib

# The function sha256_bytes takes a bytes object and returns its SHA256 hash as a hex string
# We need to provide a bytes argument

test_data = b"Hello, World!"

ARGS = {
    "data": test_data
}
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
sha256_file Ran
ran; our contract draft was wrong, not the code
MykolaPinchuk/TML-bench/orchestrator/provenance.py
code served (permissive licence) · get_code("8d12fe7f3b8fd20d")
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 hashlib
from pathlib import Path
import tempfile

# Create a temporary file with some content
with tempfile.NamedTemporaryFile(mode='wb', delete=False, suffix='.txt') as tmp:
    tmp.write(b'This is test content for sha256 hashing.\n' * 100)
    tmp_path = tmp.name

# Create a Path object pointing to the temporary file
test_path = Path(tmp_path)

ARGS = {
    'path': test_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
sha256_text Ran
ran; our contract draft was wrong, not the code
MykolaPinchuk/TML-bench/orchestrator/provenance.py
code served (permissive licence) · get_code("e2eaf67441c33629")
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 hashlib

def sha256_text(text: str) -> str:
    return hashlib.sha256(text.encode('utf-8')).hexdigest()

ARGS = {
    "text": "hello world"
}
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
strip_ansi Ran
ran; our contract draft was wrong, not the code
MykolaPinchuk/TML-bench/orchestrator/kilo_cli.py
code served (permissive licence) · get_code("322e82910c43c741")
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 re

_ANSI_RE = re.compile('\\x1b\\[[0-?]*[ -/]*[@-~]')

def strip_ansi(s: str) -> str:
    return _ANSI_RE.sub('', s)


ARGS = {
    's': '\x1b[31mHello\x1b[0m World'
}
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
sha256_file Not yet run
unverified (harvested, no recorded run)
MykolaPinchuk/TML-bench/orchestrator/hash_utils.py
code served (permissive licence) · get_code("d106646067c3cd11")
How it ran

no invocation recorded.

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, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 6 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
sha256_bytes 1 other paper
1 dated earlier
2505.14683 (2025-05)
Emerging Properties in Unified Multimodal Pretraining

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

Autonomous coding agents can produce strong tabular baselines quickly on Kaggle-style tasks. Practical value depends on end-to-end correctness and reliability under time limits. This paper introduces TML-bench, a tabular benchmark for data science agents on Kaggle-style tasks. This paper evaluates 10 OSS LLMs on four Kaggle competitions and three time budgets (240s, 600s, and 1200s). Each model is run five times per task and budget. A run is successful if it produces a valid submission and a private-holdout score on hidden labels that are not accessible to the agent. This paper reports median performance, success rates, and run-torun variability. MiniMax-M2.1-TEE achieves the best aggregate performance score on all four competitions under the paper's primary aggregation. Average performance improves with larger time budgets. Scaling is noisy for some individual models at the current run count. Code and materials are available at https://github.com/MykolaPinchuk/TML-bench/tree/master. Tabular machine learning remains a common practical workload. The workflow includes data loading, feature preparation, model training, evaluation, iteration, and producing a correctly formatted submission artifact. Benchmarks that only test isolated coding tasks miss important failure modes and trade-offs. This paper evaluates data science agents on a strict tabular benchmark with private-holdout scoring. The target is an auditable leaderboard built from repeatable runs. The benchmark considers score distributions across runs rather than a single best attempt. For practitioners, this framing matters for two reasons. First, a good agent must be reliable. Capability on a single lucky run is insufficient. Second, comparisons should remain meaningful when tasks use different metrics and when runs are time-bounded. This paper makes the following contributions: • This paper introduces a strict benchmark protocol for Kaggle-style tabular tasks with deterministic preparation, strict submission validation, and private-holdout scoring. • This paper uses a repeatable reporting policy with fixed agent instructions, a fixed suite, and median-of-five aggregation with explicit coverage requirements.

For agents

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

get_harvested_code_for_paper("2603.05764")
get_code_for_paper("2603.05764")
have("2603.05764")

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

curl https://syntology.ai/api/ran/2603.05764.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.05764.svg)](https://syntology.ai/paper/2603.05764)

Connect an agent — have() is free.