SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.01421 · 2026

SciDER: Scientific Data-centric End-to-end Researcher

Preslav Nakov, Ke Lin, Yilin Lu, Yiyang Luo, Xuehang Guo, Owais Aijaz, Scider Pipeline

arXiv · PDF · Open in the Atlas

Code that ran

7 of 8 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; 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/2603.01421.json · badge: /api/ran/2603.01421.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
leonardodalinky/SciDER canonical (classifier label)
classifier basis: census
7 ran · 1 no recorded run
FunctionStatusWhere it lives
allow Ran
ran; our contract draft was wrong, not the code
leonardodalinky/SciDER/scider/core/permissions.py
code served (permissive licence) · get_code("1e261be740641d9e")
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 PermissionResult. 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 enum
from dataclasses import dataclass

class PermissionBehavior(str, enum.Enum):
    ALLOW = 'allow'
    DENY = 'deny'

@dataclass
class PermissionResult:
    behavior: PermissionBehavior
    reason: str | None = None
    tool_name: str | None = None

    @property
    def allowed(self) -> bool:
        return self.behavior == PermissionBehavior.ALLOW

def allow(reason: str | None=None) -> PermissionResult:
    return PermissionResult(behavior=PermissionBehavior.ALLOW, reason=reason)

ARGS = {
    "reason": "User has sufficient permissions to access this resource"
}
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 PermissionResult — a PermissionResult output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PermissionResult 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
build_system_reminder Ran
ran; our contract draft was wrong, not the code
leonardodalinky/SciDER/scider/core/agent_utils.py
code served (permissive licence) · get_code("15c0c89f7b36eef7")
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 = {
    "parts": [
        "You are a helpful assistant.",
        "Please respond concisely.",
        "Avoid harmful content."
    ],
    "preamble": "System Instructions:"
}
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
deny Ran
ran; our contract draft was wrong, not the code
leonardodalinky/SciDER/scider/core/permissions.py
code served (permissive licence) · get_code("bee7fac42a31bd37")
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 PermissionResult. 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 enum
from dataclasses import dataclass

class PermissionBehavior(str, enum.Enum):
    ALLOW = 'allow'
    DENY = 'deny'

@dataclass
class PermissionResult:
    behavior: PermissionBehavior
    reason: str | None = None
    tool_name: str | None = None

    @property
    def allowed(self) -> bool:
        return self.behavior == PermissionBehavior.ALLOW

def deny(reason: str) -> PermissionResult:
    return PermissionResult(behavior=PermissionBehavior.DENY, reason=reason)

ARGS = {
    "reason": "User does not have permission to access this resource"
}
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 PermissionResult — a PermissionResult output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PermissionResult 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_spec_for_review Ran
ran; our contract draft was wrong, not the code
leonardodalinky/SciDER/scider/core/hypo_data_gen.py
code served (permissive licence) · get_code("5ffe78ea579152ab")
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
from typing import Any, Literal
from pydantic import BaseModel, field_validator

class FeatureSpec(BaseModel):
    name: str
    dtype: Literal['numeric', 'categorical', 'integer']
    distribution: Literal['gaussian', 'uniform', 'categorical', 'integer_range', 'formula']
    params: dict[str, Any]
    description: str = ''

    @field_validator('name')
    @classmethod
    def name_must_be_identifier(cls, v: str) -> str:
        v = v.strip().replace(' ', '_')
        if not v.isidentifier():
            raise ValueError(f'Feature name must be a valid identifier, got: {v!r}')
        return v

class DataGenSpec(BaseModel):
    dataset_name: str = 'synthetic_data'
    features: list[FeatureSpec]

def format_spec_for_review(spec: DataGenSpec) -> str:
    """Format a ``DataGenSpec`` as a readable table for user approval."""
    lines = [f'**Dataset**: {spec.dataset_name}', '']
    lines.append('| Feature | Type | Distribution | Parameters / Formula |')
    lines.append('|---------|------|--------------|----------------------|')
    for f in spec.features:
        if f.distribution == 'formula':
            params_str = f"`{f.params.get('expression', '')}`"
        else:
            params_str = ', '.join((f'{k}={v}' for k, v in f.params.items()))
        lines.append(f'| {f.name} | {f.dtype} | {f.distribution} | {params_str} |')
        if f.description:
            lines.append(f'|  | | | _{f.description}_ |')
    return '\n'.join(lines)

# Construct valid arguments for format_spec_for_review
feature1 = FeatureSpec(
    name='age',
    dtype='numeric',
    distribution='gaussian',
    params={'mean': 30, 'std': 10},
    description='Age of the subject'
)

feature2 = FeatureSpec(
    name='category',
    dtype='categorical',
    distribution='categorical',
    params={'categories': ['A', 'B', 'C'], 'probabilities': [0.5, 0.3, 0.2]},
    description='Categorical feature'
)

feature3 = FeatureSpec(
    name='count',
    dtype='integer',
    distribution='integer_range',
    params={'min': 0, 'max': 100},
    description=''
)

feature4 = FeatureSpec(
    name='derived',
    dtype='numeric',
    distribution='formula',
    params={'expression': 'age * 2 + count'},
    description='Formula-based feature'
)

spec = DataGenSpec(
    dataset_name='my_synthetic_dataset',
    features=[feature1, feature2, feature3, feature4]
)

ARGS = {
    'spec': spec
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pydantic, 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
is_dangerous_path Ran
ran, violated the contract we drafted
leonardodalinky/SciDER/scider/core/permissions.py
code served (permissive licence) · get_code("42fdc7d61373ee11")
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
from pathlib import Path

# Test case 1: A regular safe file path
safe_path = "/home/user/documents/file.txt"

# Test case 2: A dangerous path with .git in it
dangerous_path = "/home/user/.git/config"

# Test case 3: A dangerous path with .env
env_path = "/app/.env"

# Test case 4: A dangerous path with .bashrc
bashrc_path = "/root/.bashrc"

# Test case 5: A safe nested path
safe_nested = "/var/lib/application/data/output.json"

# We'll use one of these for the actual ARGS
ARGS = {
    "file_path": dangerous_path
}
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
sprint_chained_exception Ran
ran; our contract draft was wrong, not the code
leonardodalinky/SciDER/scider/core/errors.py
code served (permissive licence) · get_code("04efce5e517d1662")
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

# Create a chain of exceptions
try:
    try:
        try:
            raise ValueError("Inner exception")
        except ValueError as e1:
            raise RuntimeError("Middle exception") from e1
    except RuntimeError as e2:
        raise TypeError("Outer exception") from e2
except TypeError as e3:
    chained_exception = e3

ARGS = {
    "e": chained_exception
}
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
str_to_bool Ran
ran, violated the contract we drafted
leonardodalinky/SciDER/scider/core/constant.py
code served (permissive licence) · get_code("4547b076015e1d60")
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

torch.manual_seed(0)

ARGS = {
    "s": "true"
}
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 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
function_to_json_schema Not yet run
unverified (harvested, no recorded run)
leonardodalinky/SciDER/scider/core/llms.py
code served (permissive licence) · get_code("208257adcdf5fffc")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 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 8 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

While large language models accelerate scientific discovery, existing agents face severe limitations in adaptability, domain generalization, and multimodal scalability, often struggling to autonomously process raw, domain-specific experimental data. To overcome these barriers, we introduce SciDER, a multi-agent system designed to flexibly automate the entire research lifecycle. This framework employs a novel data-centric approach and integrates a dynamic multimodal skill system across four specialized sub-agents. Specifically, an ideation agent generates novel hypotheses via Evolutionary Idea Search, a data analysis agent systematically structures raw data, an experimentation agent synthesizes executable code grounded in dataset characteristics, and a critic agent drives iterative selfrefinement. To democratize open-source scientific discovery, we release OpenSciDER-SFT-8K, a high-quality execution trajectory dataset, alongside the OpenSciDER-27B fine-tuned model. Across six benchmarks, SciDER and OpenSciDER obtain competitive or leading results, with especially strong gains on data-centric analysis, end-to-end research execution, and multimodal scientific visualization. By integrating data analysis with experimental execution, SciDER bridges the gap between abstract scientific reasoning and reproducible experimentation synthesis.

For agents

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

get_harvested_code_for_paper("2603.01421")
get_code_for_paper("2603.01421")
have("2603.01421")

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

curl https://syntology.ai/api/ran/2603.01421.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.01421.svg)](https://syntology.ai/paper/2603.01421)

Connect an agent — have() is free.