SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.01692 · 2026

Reasoning as Gradient: Scaling MLE Agents Beyond Tree Search

Jian Wang, Xiao Yang, Weiqing Liu, Yuge Zhang, Jiang Bian, Shikai Fang, Yifei Zhang, Xu Yang, Qizheng Li, Jingyuan Li, Bowen Xian, Mingrui Xu

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 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; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 2 failures of our instrument; 0 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/2603.01692.json · badge: /api/ran/2603.01692.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
microsoft/RD-Agent canonical (classifier label)
classifier basis: census
4 ran · 0 no recorded run
FunctionStatusWhere it lives
find_context Ran
ran, honoured the contract we drafted
microsoft/RD-Agent/rdagent/utils/agent/apply_patch.py
code served (permissive licence) · get_code("5b08c8e830837a8b")
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 int; 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 find_context function

ARGS = {
    "lines": [
        "def hello():",
        "    print('world')",
        "    return 42",
        "",
        "def goodbye():",
        "    print('farewell')",
    ],
    "context": [
        "def hello():",
        "    print('world')",
    ],
    "start": 0,
    "eof": False,
}
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 int — 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
find_context_core Ran
ran, honoured the contract we drafted
microsoft/RD-Agent/rdagent/utils/agent/apply_patch.py
code served (permissive licence) · get_code("5aa0b311b6991142")
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 int; 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.

# Analysis of find_context_core function:
# - lines: list[str] - a list of strings
# - context: list[str] - a list of strings to search for
# - start: int - starting index for search
# The function searches for context within lines, starting from index 'start'
# It returns a tuple of (index, match_type)

# Create test data
lines = [
    "def hello():",
    "    print('world')",
    "    return 42",
    "def goodbye():",
    "    print('farewell')",
]

context = [
    "    print('world')",
    "    return 42",
]

start = 0

ARGS = {
    "lines": lines,
    "context": context,
    "start": start,
}
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 int — 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
get_caller_dir Ran
ran; our contract draft was wrong, not the code
microsoft/RD-Agent/rdagent/utils/agent/tpl.py
code served (permissive licence) · get_code("00287d3cf603df5a")
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 PosixPath. 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 inspect
from pathlib import Path

# The function uses inspect.stack() to get the caller's frame and module
# It doesn't actually use the parameters in a way that requires complex objects
# upshift is an int that controls how many levels up the stack to look

ARGS = {
    "upshift": 0
}
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 PosixPath — a PosixPath output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PosixPath 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
peek_next_section Ran
ran; our contract draft was wrong, not the code
microsoft/RD-Agent/rdagent/utils/agent/apply_patch.py
code served (permissive licence) · get_code("4758e9007243dd65")
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.

from dataclasses import dataclass, field

@dataclass
class Chunk:
    orig_index: int = -1
    del_lines: list[str] = field(default_factory=list)
    ins_lines: list[str] = field(default_factory=list)

# Test case 1: Simple addition
lines_1 = [
    " line1",
    " line2",
    "+added line",
    " line3",
    "@@"
]

# Test case 2: Deletion
lines_2 = [
    " keep1",
    "-delete1",
    "-delete2",
    " keep2",
    "*** End of File"
]

# Test case 3: Mixed operations with end marker
lines_3 = [
    " original1",
    "-old line",
    "+new line",
    " original2",
    "*** End of File"
]

# Test case 4: Just keep lines
lines_4 = [
    " line1",
    " line2",
    " line3",
    "@@"
]

# Test case 5: Complex scenario
lines_5 = [
    " context1",
    "-removed",
    "+inserted",
    " context2",
    "-another_removal",
    " context3",
    "*** End of File"
]

ARGS = {
    "lines": lines_5,
    "index": 0
}
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 — 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

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 4 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 4 the same month, 0 later) and 0 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
find_context 1 other paper
1 the same month
2603.01712 (2026-03)
FT-Dojo: Towards Autonomous LLM Fine-Tuning with Language Agents
find_context_core 1 other paper
1 the same month
2603.01712 (2026-03)
FT-Dojo: Towards Autonomous LLM Fine-Tuning with Language Agents
get_caller_dir 1 other paper
1 the same month
2603.01712 (2026-03)
FT-Dojo: Towards Autonomous LLM Fine-Tuning with Language Agents
peek_next_section 1 other paper
1 the same month
2603.01712 (2026-03)
FT-Dojo: Towards Autonomous LLM Fine-Tuning with Language Agents

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

LLM-based agents for machine learning engineering (MLE) predominantly rely on tree search, a form of gradient-free optimization that uses scalar validation scores to rank candidates. As LLM reasoning capabilities improve, exhaustive enumeration becomes increasingly inefficient compared to directed updates, analogous to how accurate gradients enable efficient descent over random search. We introduce Gome, an MLE agent that operationalizes gradient-based optimization. Gome maps structured diagnostic reasoning to gradient computation, success memory to momentum, and multi-trace execution to distributed optimization. Under a closed-world protocol that isolates architectural effects from external knowledge, Gome achieves a state-of-the-art 35.1% any-medal rate on MLE-Bench with a restricted 12-hour budget on a single V100 GPU. Scaling experiments across 10 models reveal a critical crossover: with weaker models, tree search retains advantages by compensating for unreliable reasoning through exhaustive exploration; as reasoning capability strengthens, gradient-based optimization progressively outperforms, with the gap widening at frontier-tier models. Given the rapid advancement of reasoning-oriented LLMs, this positions gradientbased optimization as an increasingly favorable paradigm. We release our codebase and GPT-5 traces at: https://github.com/microsoft/RD-Agent.

For agents

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

get_harvested_code_for_paper("2603.01692")
get_code_for_paper("2603.01692")
have("2603.01692")

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

curl https://syntology.ai/api/ran/2603.01692.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.01692.svg)](https://syntology.ai/paper/2603.01692)

Connect an agent — have() is free.