SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2607.26648 · 2026

The Sparsity Ceiling: Where Spiking Networks Canand Cannot-Trade Activity for Energy

Zeyu Wang

arXiv · PDF

Code that ran

15 of 19 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; 4 honoured the contract we drafted, 3 violated it and 0 have no contract verdict; 8 failures of our instrument; 4 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/2607.26648.json · badge: /api/ran/2607.26648.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
zeyuyuyu/sparsity-ceiling canonical (classifier label)
classifier basis: census
15 ran · 4 no recorded run
FunctionStatusWhere it lives
Hb Ran
ran, honoured the contract we drafted
zeyuyuyu/sparsity-ceiling/agg_shrinkH.py
code served (permissive licence) · get_code("c16e1f218499e541")
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 float, 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 math

def Hb(p):
    if p <= 0 or p >= 1:
        return 0.0
    return -p * math.log2(p) - (1 - p) * math.log2(1 - p)

# The function Hb computes binary entropy for a probability value p
# It takes a single parameter p which should be a float between 0 and 1 (exclusive)

ARGS = {
    "p": 0.5
}
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 float, 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
batch Ran
ran, honoured the contract we drafted
zeyuyuyu/sparsity-ceiling/exp1_2d.py
code served (permissive licence) · get_code("8ff194927721d885")
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 3, 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 torch.nn as nn

torch.manual_seed(0)

# Analyze the function signature and body:
# - N: sequence length (used in tensor shapes)
# - S: vocabulary size (used in randint and tensor creation)
# - bs: batch size (used in tensor shapes)
# - dev: device (used in tensor creation)

# The function creates tensors with shapes like (bs, N), (bs, 1), (bs, N+1)
# It uses torch.randint(1, S+1, ...) so S must be an integer
# All parameters are used for tensor construction

ARGS = {
    "N": 4,      # sequence length
    "S": 7,      # vocabulary size
    "bs": 2,     # batch size
    "dev": "cpu" # device
}
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 tuple, ndim 3, 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
cbatch Ran
ran, honoured the contract we drafted
zeyuyuyu/sparsity-ceiling/exp2_dissoc.py
code served (permissive licence) · get_code("ca1a5e34cf44e557")
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 3, 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 torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "N": 4,
    "S": 3,
    "bs": 2,
    "dev": "cpu"
}
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 tuple, ndim 3, 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
floor_for Ran
ran, honoured the contract we drafted
zeyuyuyu/sparsity-ceiling/agg_shrinkH.py
code served (permissive licence) · get_code("accf3f47443d66ea")
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 float, 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 math

DEMAND = 16 * math.log2(16)

def Hb(p):
    if p <= 0 or p >= 1:
        return 0.0
    return -p * math.log2(p) - (1 - p) * math.log2(1 - p)

def floor_for(H):
    t = DEMAND / H
    if t >= 1:
        return 0.5
    lo, hi = (0.0, 0.5)
    for _ in range(60):
        mid = (lo + hi) / 2
        if Hb(mid) < t:
            lo = mid
        else:
            hi = mid
    return (lo + hi) / 2

# The function floor_for takes one parameter: H (a float)
# H is used in division: t = DEMAND / H
# Then H is compared indirectly through Hb calculations
# H should be a positive float value

ARGS = {
    "H": 2.5
}
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 float, 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
gated Ran
ran, violated the contract we drafted
zeyuyuyu/sparsity-ceiling/energy_shape.py
code served (permissive licence) · get_code("cb289e2be9818ade")
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)

# The function gated() takes a single parameter 'd' and calls .get() on it
# with string keys 'out_theta' and 'out_prand'. This means 'd' should be
# a dict-like object. The function returns a boolean based on whether either
# of these keys has a truthy value.

# Test case 1: d with out_theta set (truthy)
d1 = {'out_theta': 0.5, 'out_prand': None}

# Test case 2: d with out_prand set (truthy)
d2 = {'out_theta': None, 'out_prand': 0.3}

# Test case 3: d with both set (truthy)
d3 = {'out_theta': 0.1, 'out_prand': 0.2}

# Test case 4: d with neither set (falsy)
d4 = {'out_theta': None, 'out_prand': None}

# We'll use the first test case as the main argument
ARGS = {
    'd': d1
}
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
label Ran
ran; our contract draft was wrong, not the code
zeyuyuyu/sparsity-ceiling/agg_stream.py
code served (permissive licence) · get_code("ec060dce8d9f07c6")
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)

# The function takes a tuple k with 5 elements: v, th, dn, it, ot
# v: a string (used with f-string concatenation)
# th: optional value that can be None or something convertible to string
# dn: a number (compared with > 0)
# it: a number (compared with > 0)
# ot: a number (compared with > 0)

k = (
    "model_v1",      # v: string
    0.5,             # th: threshold value (not None)
    0.02,            # dn: dropout noise amount
    0.1,             # it: input threshold
    0.15             # ot: output threshold
)

ARGS = {
    'k': k
}
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
label Ran
ran; our contract draft was wrong, not the code
zeyuyuyu/sparsity-ceiling/energy_datapath.py
code served (permissive licence) · get_code("ae8400477d5a6dce")
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

torch.manual_seed(0)

# Test case 1: analog variant
config_analog = {
    'variant': 'analog',
    'analog': {
        'theta': 0.5
    }
}

# Test case 2: digital variant with regularization
config_digital_reg = {
    'variant': 'digital',
    'dig_reg': {
        'dig_noise': 0.01,
        'dig_bits': 8,
        'wd': 0.0001
    }
}

# Test case 3: digital variant without regularization
config_digital_plain = {
    'variant': 'digital'
}

# Test case 4: digital variant with partial regularization
config_digital_partial = {
    'variant': 'digital',
    'dig_reg': {
        'dig_bits': 16
    }
}

# Use the analog config as the primary test case
ARGS = {
    'name': 'test_model',
    'd': config_analog
}
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
load_nogate Ran
ran; our contract draft was wrong, not the code
zeyuyuyu/sparsity-ceiling/energy_shape.py
code served (permissive licence) · get_code("fff786372b2be601")
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 dict. 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 os
import json
import tempfile
import shutil

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

# Create test JSON files that match the function's expectations
test_files = [
    {
        'task': 'charlm',
        'variant': 'analog',
        'analog': {'theta': 0.5},
        'out_theta': None,
        'out_prand': None,
        'dig_reg': {}
    },
    {
        'task': 'charlm',
        'variant': 'digital',
        'out_theta': None,
        'out_prand': None,
        'dig_reg': {
            'dig_noise': 0.1,
            'dig_bits': 8,
            'wd': 0.001
        }
    },
    {
        'task': 'charlm',
        'variant': 'digital',
        'out_theta': None,
        'out_prand': None,
        'dig_reg': {
            'dig_noise': 0.05
        }
    },
    {
        'task': 'charlm',
        'variant': 'vanilla',
        'out_theta': None,
        'out_prand': None,
        'dig_reg': {}
    },
    {
        'task': 'other_task',
        'variant': 'analog',
        'analog': {'theta': 0.3},
        'out_theta': None,
        'out_prand': None,
        'dig_reg': {}
    },
    {
        'task': 'charlm',
        'variant': 'analog',
        'analog': {'theta': 0.7},
        'out_theta': 0.5,
        'out_prand': None,
        'dig_reg': {}
    }
]

# Write test files to temp directory
for i, test_data in enumerate(test_files):
    filename = os.path.join(temp_dir, f'charlm_test_{i}.json')
    with open(filename, 'w') as f:
        json.dump(test_data, f)

ARGS = {
    'runs': temp_dir
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
make_data Ran
ran; our contract draft was wrong, not the code
zeyuyuyu/sparsity-ceiling/lm_poc.py
code served (permissive licence) · get_code("a2ee87a04024c071")
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.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Construct arguments for make_data(txt, L, dev)
txt = "hello world"  # Small text string
L = 3  # Sequence length
dev = "cpu"  # Device (CPU only)

ARGS = {
    "txt": txt,
    "L": L,
    "dev": dev
}
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 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
mean Ran
ran, violated the contract we drafted
zeyuyuyu/sparsity-ceiling/agg_copy.py
code served (permissive licence) · get_code("0479fb5f08e12e14")
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 float, 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

torch.manual_seed(0)

# The function mean(x) expects:
# - x: an iterable that supports sum() and len()
# This could be a list of numbers, tensors, or other summable objects

# Create a simple list of numbers
x = [1.0, 2.0, 3.0, 4.0, 5.0]

ARGS = {
    "x": x
}
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 float, 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
mean_sd Ran
ran, violated the contract we drafted
zeyuyuyu/sparsity-ceiling/energy_shape.py
code served (permissive licence) · get_code("4ca655af1f9299e3")
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 torch
import numpy as np

torch.manual_seed(0)

# Analyze the function:
# - mean_sd takes parameter 'v'
# - v must support len() to get n
# - v must support sum() and iteration
# - Elements must support arithmetic (+, -, **)
# - The function computes mean and standard deviation

# Create a simple list of numbers for v
v = [1.0, 2.0, 3.0, 4.0, 5.0]

ARGS = {
    "v": v
}
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 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
ms Ran
ran; our contract draft was wrong, not the code
zeyuyuyu/sparsity-ceiling/agg_charlm.py
code served (permissive licence) · get_code("04805ecc69f243fa")
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 statistics as st

def ms(v):
    return (st.mean(v), st.stdev(v) if len(v) > 1 else 0.0)

# The function ms takes a single parameter v, which should be an iterable of numbers
# It computes the mean and standard deviation of v
# We need to provide a list/sequence of numeric values

ARGS = {
    "v": [1.0, 2.0, 3.0, 4.0, 5.0]
}
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, 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
ms Ran
ran; our contract draft was wrong, not the code
zeyuyuyu/sparsity-ceiling/agg_copy.py
code served (permissive licence) · get_code("204405eee59d00b5")
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 math

torch_manual_seed_0 = __import__('torch').manual_seed(0) if 'torch' in dir() else None

# Analyze the function ms(x, p=4):
# - x: must be a list/sequence of numbers (used with len(x), sum(x), iteration)
# - p: integer parameter with default value 4 (used in string formatting)

# Construct a simple list of numeric values
x = [1.5, 2.3, 1.8, 2.1, 1.9]

# p is optional with default=4, but we'll provide it explicitly
p = 4

ARGS = {
    'x': x,
    'p': p,
}
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
regtag Ran
ran; our contract draft was wrong, not the code
zeyuyuyu/sparsity-ceiling/agg_charlm.py
code served (permissive licence) · get_code("07c9cd4fa0af9a1d")
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)

# Test case 1: unregularized (all zeros/None)
d1 = {
    'dig_reg': {
        'dig_noise': 0,
        'dig_bits': 0,
        'wd': 0
    }
}

# Test case 2: with regularization parameters
d2 = {
    'dig_reg': {
        'dig_noise': 0.1,
        'dig_bits': 8,
        'wd': 0.0001
    }
}

ARGS = {
    'd': d2
}
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
sd Ran
ran; our contract draft was wrong, not the code
zeyuyuyu/sparsity-ceiling/agg_copy.py
code served (permissive licence) · get_code("26b753f97c09e079")
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 float, 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 math

# The function sd() takes parameter 'x' and computes standard deviation
# It calls: len(x), sum(x), mean(x) on x
# It iterates through x with 'for v in x'
# So x should be a sequence of numbers

# Create a simple list of numbers with at least 2 elements
x = [1.0, 2.0, 3.0, 4.0, 5.0]

ARGS = {
    'x': x
}
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 float, 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
full_event Not yet run
unverified (harvested, no recorded run)
zeyuyuyu/sparsity-ceiling/agg_eventro.py
code served (permissive licence) · get_code("97be458480768712")
How it ran

no invocation recorded.

load Not yet run
unverified (harvested, no recorded run)
zeyuyuyu/sparsity-ceiling/agg_seed12.py
code served (permissive licence) · get_code("95c543f8ad34ebd8")
How it ran

no invocation recorded.

load_text Not yet run
unverified (harvested, no recorded run)
zeyuyuyu/sparsity-ceiling/lm_poc.py
code served (permissive licence) · get_code("767d205d9a2a72df")
How it ran

no invocation recorded.

terms Not yet run
unverified (harvested, no recorded run)
zeyuyuyu/sparsity-ceiling/energy_datapath.py
code served (permissive licence) · get_code("80106af23d3c137f")
How it ran

no invocation recorded.

Also present in other papers' code

Of 19 samples lifted from this paper's linked repositories, 19 carry a function identity and 0 do not. Of the 19 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 18 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
mean 2 other papers
2 dated earlier
2402.05099 (2024-02)
Hydragen: High-Throughput LLM Inference with Shared Prefixes
2407.14985 (2024-07)
Generalization v.s. Memorization: Tracing Language Models' Capabilities Back to Pretraining Data

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

Spiking neural networks (SNNs) are promoted as an energy-efficient substrate because sparse, event-driven activity replaces dense multiply-accumulates (MACs) with cheap accumulates (ACs). We argue that the energy dividend of sparsity is not a property of SNNs but of the task. Using a controlled protocol-architecture held fixed, only the hidden unit swapped (continuous vs. leaky-integrateand-fire), plus a two-sided target-firing-rate probe-we measure how far activity can be pushed down before quality breaks. On feed-forward perception, hidden firing is naturally low and drives precisely to a 5% target with no accuracy loss. On recurrent character-level language modeling the same probe, targeting 10%, cannot move firing below ∼ 50%: the recurrent state must stay active to carry information. A spiking Transformer, by contrast, sparsifies freely to 2% firing at no extra quality cost (3 seeds) -so the ceiling is a property of recurrent compression, not of sequence modeling. Attention escapes the floor only by storing the full key-value cache, trading a firing floor for a memory wall: neither architecture escapes cost on neuromorphic hardware, they pay on different axes. We formalize the ceiling with an informationtheoretic firing-floor bound ρ ≥ H -1 b (log 2 M/H) and confirm its predictions on synthetic tasks: the floor rises with memory load and falls with state width, and-refuting a naive memory-only reading-also rises with task difficulty, exactly as a load-based bound requires. A layerwise accounting exposes an input floor that further caps op reduction under dense (frame-replayed) input, isolating native event-driven input as the setting where neuromorphic hardware wins. The results recast "SNNs save energy" as a task-conditional claim with a concrete, measurable rule for where neuromorphic computation pays off.

For agents

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

get_harvested_code_for_paper("2607.26648")
get_code_for_paper("2607.26648")
have("2607.26648")

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

curl https://syntology.ai/api/ran/2607.26648.json

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

[![Syntology run record](https://syntology.ai/api/ran/2607.26648.svg)](https://syntology.ai/paper/2607.26648)

Connect an agent — have() is free.