SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2607.28008 · 2026

RepBench: A Benchmark for Representation Engineering

Long Zhang, Xueru Bai, Yanshi Li, Shuman Liu

arXiv · PDF · Open in the Atlas

Code that ran

3 of 3 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, 0 violated it and 0 have no contract verdict; 3 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/2607.28008.json · badge: /api/ran/2607.28008.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
xlands/RepBench canonical (classifier label)
classifier basis: census
3 ran · 0 no recorded run
FunctionStatusWhere it lives
compact Ran
ran; our contract draft was wrong, not the code
xlands/RepBench/src/representation_cluster/representation_fit_pipeline.py
code served (permissive licence) · get_code("c85bac0ce6fb18e5")
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
import torch

torch.manual_seed(0)

# The function 'compact' takes a 'value' parameter and an optional 'limit' parameter
# It converts value to string, removes extra whitespace, and truncates if needed

# Test with various value types to ensure robustness
test_value = "This is a sample text with   multiple    spaces that should be compacted"
test_limit = 50

ARGS = {
    "value": test_value,
    "limit": test_limit
}
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
read_jsonl Ran
ran; our contract draft was wrong, not the code
xlands/RepBench/src/representation_cluster/representation_fit_pipeline.py
code served (permissive licence) · get_code("7902fbf13b57a305")
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 json
import tempfile
from pathlib import Path

# Create a temporary JSONL file with test data
temp_dir = tempfile.TemporaryDirectory()
temp_path = Path(temp_dir.name) / "test.jsonl"

# Write some test data to the JSONL file
with open(temp_path, 'w', encoding='utf-8') as f:
    f.write(json.dumps({"id": 1, "text": "hello"}) + "\n")
    f.write(json.dumps({"id": 2, "text": "world"}) + "\n")
    f.write(json.dumps({"id": 3, "value": 42}) + "\n")

ARGS = {
    "path": str(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 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
top_list Ran
ran; our contract draft was wrong, not the code
xlands/RepBench/src/representation_cluster/representation_fit_pipeline.py
code served (permissive licence) · get_code("b44048f7db6af12c")
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 torch
import numpy as np

# Analyze the function:
# - items: expects an iterable of tuples (name, count) or None
# - n: integer, default 10
# - function slices items[:n] and creates dicts with 'name' and 'count' keys

# Create sample data: list of (name, count) tuples
items = [
    ('apple', 5),
    ('banana', 3),
    ('cherry', 8),
    ('date', 2),
    ('elderberry', 12),
    ('fig', 1),
    ('grape', 9),
    ('honeydew', 4),
]

ARGS = {
    'items': items,
    'n': 5
}
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 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

Also present in other papers' code

Of 3 samples lifted from this paper's linked repositories, 3 carry a function identity and 0 do not. Of the 3 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 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.

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

Representation engineering reads and steers directions in language-model hidden states, yet most methods are evaluated on paper-specific synthetic or hand-built data. RepBench turns existing public benchmarks into reusable supervision for representation engineering. It compiles 13,427 benchmark papers into a taxonomy of 182 representation categories and 337 public datasets into 44,387 audited texts covering 94 textprobeable representation targets. We first test whether these data can replace bespoke supervision in two published protocols while holding the model, estimator, data budget, intervention, and external evaluation fixed. RepBench nearly matches the original-data refusal readout (maximum AUROC difference 0.0004), produces the target behavior on 82-91% of prompts across five activation interventions, and retains 78% of CAA's probability effect with the same bidirectional monotonic response on 817 untouched TruthfulQA questions. RepBench is built around independent benchmark sources rather than within-dataset splits. On 16 well-supported representation targets, increasing the benchmarks pooled on each side from one to three raises 16-way hidden-state retrieval top-1 from 0.206 to 0.339 and MRR from 0.402 to 0.550, with monotonic gains on all 12 models. Representation geometry and a 12-model, four-readout LOBO grid further characterize cross-benchmark structure and readout performance. RepBench thus provides scalable, externally useful benchmark supervision as a common foundation for future probing and steering methods. Code and data are available at https://github.com/xlands/RepBench.

For agents

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

get_harvested_code_for_paper("2607.28008")
get_code_for_paper("2607.28008")
have("2607.28008")

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

curl https://syntology.ai/api/ran/2607.28008.json

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

[![Syntology run record](https://syntology.ai/api/ran/2607.28008.svg)](https://syntology.ai/paper/2607.28008)

Connect an agent — have() is free.