SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2104.04480 · CVPR · 2021

Improving the Efficiency and Robustness of Deepfakes Detection through Precise Geometric Features

arXiv · PDF · Open in the Atlas

Code that ran

11 of 16 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; 9 failures of our instrument; 5 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/2104.04480.json · badge: /api/ran/2104.04480.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
frederickszk/LRNet canonical (classifier label)
classifier basis: census
11 ran · 5 no recorded run
FunctionStatusWhere it lives
calculate_face_distance Ran
ran, honoured the contract we drafted
frederickszk/LRNet/demo/utils/landmark_utils.py
code served (permissive licence) · get_code("2051a96bbb72d3d5")
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 float64, ndim 0, 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 numpy as np

# Create sample face shape data
# shape_prev: coordinates of facial landmarks from previous frame
# Each landmark has (x, y) coordinates, so shape is (num_landmarks, 2)
shape_prev = np.array([
    [100.0, 150.0],
    [110.0, 160.0],
    [120.0, 155.0],
    [105.0, 165.0]
])

# shape_curr: coordinates of facial landmarks from current frame
# Similar structure, but with slightly different positions
shape_curr = np.array([
    [102.0, 152.0],
    [112.0, 162.0],
    [122.0, 157.0],
    [107.0, 167.0]
])

ARGS = {
    "shape_prev": shape_prev,
    "shape_curr": shape_curr
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
calculate_face_size Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
frederickszk/LRNet/calibrator/LC.py
code served (permissive licence) · get_code("afe7866194ad8ad3")
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 float64, ndim 0, 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 numpy as np

# The function expects 'shape' parameter which should be an array-like object
# with 2D coordinates (x, y). Looking at the function:
# - It calls np.min(shape, axis=0) to get minimum x, y values
# - It calls np.max(shape, axis=0) to get maximum x, y values
# - It calculates the difference between max and min for both dimensions
# This suggests 'shape' should be a 2D array where each row is a point (x, y)

# Create a small sample shape with a few 2D points
shape = np.array([
    [1.0, 2.0],
    [5.0, 3.0],
    [3.0, 6.0],
    [2.0, 4.0]
])

ARGS = {
    'shape': shape
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
check_and_merge Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
frederickszk/LRNet/calibrator/LC.py
code served (permissive licence) · get_code("fe7e329296d721d1")
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 numpy as np

ARGS = {
    "location": [
        np.array([[1.0, 2.0, 3.0, 4.0]] * 68),  # location[0]
        np.array([[1.1, 2.1, 3.1, 4.1]] * 68),  # location[1]
    ],
    "forward": [
        np.array([[1.0, 2.0, 3.0, 4.0]] * 68),  # forward[0] - forward_base
        np.array([[1.05, 2.05, 3.05, 4.05]] * 68),  # forward[1] - forward_predict
    ],
    "feedback": [
        np.array([[1.02, 2.02, 3.02, 4.02]] * 68),  # feedback[0] - feedback_predict
        np.array([[1.03, 2.03, 3.03, 4.03]] * 68),  # feedback[1] (not used in function)
    ],
    "P_predict": np.array([0.5] * 68),
    "status_fw": np.array([[1]] * 68),
    "status_fb": np.array([[1]] * 68),
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — 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
generate_weight Ran
ran, honoured the contract we drafted
frederickszk/LRNet/demo/utils/deprecated.py
code served (permissive licence) · get_code("57bd091b62ceeb21")
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 ndarray, ndim 3, dtype family float, shape (5, 5, 1) 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 numpy as np

# patch_size is an integer parameter
# The function uses it to create a weight matrix
# It should be a reasonable patch size for image processing
patch_size = 5

ARGS = {
    "patch_size": patch_size
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (5, 5, 1) — shape (5, 5, 1) 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
get_data Ran
ran; our contract draft was wrong, not the code
frederickszk/LRNet/training/data_utils.py
code served (permissive licence) · get_code("3d6516d06197ac78")
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 os
import numpy as np
import tempfile
import shutil

# Create a temporary directory with test landmark files
temp_dir = tempfile.mkdtemp()

# Create sample landmark files with shape (20, 4) - 20 frames, 4 landmarks dimensions
for file_idx in range(2):
    file_path = os.path.join(temp_dir, f"landmarks_{file_idx}.txt")
    data = np.random.randn(20, 4).astype(np.float64)
    np.savetxt(file_path, data)

ARGS = {
    "path": temp_dir,
    "fake": 1,
    "block": 5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tqdm — 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
get_data_for_test Ran
ran; our contract draft was wrong, not the code
frederickszk/LRNet/demo/classify.py
code served (permissive licence) · get_code("f967450a45f894d6")
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 numpy as np
import os
import tempfile
from os.path import join

# Create temporary directory with test data files
tmpdir = tempfile.mkdtemp()

# Create a few test files with vector data
# Each file contains vectors that will be processed
np.savetxt(join(tmpdir, 'video1.txt'), np.random.rand(20, 5))
np.savetxt(join(tmpdir, 'video2.txt'), np.random.rand(16, 5))

ARGS = {
    'path': tmpdir,
    'fake': 1,
    'block': 4
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tqdm — 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
get_data_for_test Ran
ran; our contract draft was wrong, not the code
frederickszk/LRNet/training/data_utils.py
code served (permissive licence) · get_code("ffa491400e748f12")
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 os
import tempfile
import numpy as np

# Create a temporary directory with test landmark files
temp_dir = tempfile.mkdtemp()

# Create sample landmark files with shape [num_frames, 136]
# Each file represents a video with landmarks data
for file_idx in range(2):
    filename = os.path.join(temp_dir, f"video_{file_idx}.txt")
    # Create landmarks data: 50 frames × 136 features
    data = np.random.randn(50, 136).astype(np.float32)
    np.savetxt(filename, data)

ARGS = {
    "path": temp_dir,
    "fake": 1,
    "block": 5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tqdm — 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
load_yaml Ran
ran; our contract draft was wrong, not the code
frederickszk/LRNet/training/configs/loader.py
code served (permissive licence) · get_code("3bffc6edf7b94a8f")
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 yaml
import os
import tempfile

# Create a temporary directory and a valid YAML file for testing
temp_dir = tempfile.mkdtemp()
yaml_content = """
test_key: test_value
nested:
  inner_key: inner_value
list_item:
  - item1
  - item2
"""

yaml_file_path = os.path.join(temp_dir, "test.yaml")
with open(yaml_file_path, 'w', encoding='utf-8') as f:
    f.write(yaml_content)

# Change to the temporary directory so the relative path works
original_cwd = os.getcwd()
os.chdir(temp_dir)

# Define the relative path from the current working directory
relative_path = "test.yaml"

ARGS = {
    "path": relative_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
yaml — 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
merge_video_prediction Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
frederickszk/LRNet/demo/classify.py
code served (permissive licence) · get_code("6f587a3468f0c200")
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

torch.manual_seed(0)

# Analyze the function:
# - mix_prediction: iterable of numeric values (predictions), zipped with s2v
# - s2v: iterable of labels/keys, same length as mix_prediction
# - vc: dict mapping labels to counts (used as vc[key] for division)

# Create sample data
mix_prediction = [0.3, 0.7, 0.6, 0.2, 0.8]  # 5 predictions
s2v = ['video_1', 'video_2', 'video_1', 'video_3', 'video_2']  # 5 labels, same length
vc = {
    'video_1': 2,  # divisor for video_1
    'video_2': 3,  # divisor for video_2
    'video_3': 1   # divisor for video_3
}

ARGS = {
    'mix_prediction': mix_prediction,
    's2v': s2v,
    'vc': vc
}
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
merge_video_prediction Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
frederickszk/LRNet/training/data_utils.py
code served (permissive licence) · get_code("4ca0cdd26ad809f0")
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

torch.manual_seed(0)

# Analyzing the function:
# - mix_prediction: iterable of numeric values (probabilities), compared with 0.5
# - s2v (sample_to_video): iterable parallel to mix_prediction, contains video labels (keys)
# - vc (video_count): dict mapping video labels to counts

# Create sample data
mix_prediction = [0.3, 0.7, 0.6, 0.2, 0.8]
s2v = [0, 0, 1, 1, 2]
vc = {0: 2, 1: 2, 2: 1}

ARGS = {
    "mix_prediction": mix_prediction,
    "s2v": s2v,
    "vc": vc
}
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
shape_to_face Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
frederickszk/LRNet/calibrator/LC.py
code served (permissive licence) · get_code("f1e34b9b56f48a35")
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 numpy as np

# Create a small landmark shape array (e.g., 5 landmarks with x,y coordinates)
shape = np.array([
    [10.0, 20.0],
    [50.0, 15.0],
    [30.0, 60.0],
    [20.0, 45.0],
    [40.0, 40.0]
], dtype=np.float32)

# Image dimensions
width = 640
height = 480

# Scale parameter
scale = 1.2

# Optional assigned face size
assign_face_size = None

# Threshold for center shift stability
shift_threshold = 2.0

# Optional previous center for jitter reduction
center_prev = None

ARGS = {
    "shape": shape,
    "width": width,
    "height": height,
    "scale": scale,
    "assign_face_size": assign_face_size,
    "shift_threshold": shift_threshold,
    "center_prev": center_prev
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — 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
lk_track Not yet run
unverified (harvested, no recorded run)
frederickszk/LRNet/demo/utils/deprecated.py
code served (permissive licence) · get_code("2b5c147200484bac")
How it ran

no invocation recorded.

load_data_test Not yet run
unverified (harvested, no recorded run)
frederickszk/LRNet/training/model_tf.py
code served (permissive licence) · get_code("2a53262ee3c69361")
How it ran

no invocation recorded.

load_data_train Not yet run
unverified (harvested, no recorded run)
frederickszk/LRNet/training/model_tf.py
code served (permissive licence) · get_code("4608835f4e2c1dc1")
How it ran

no invocation recorded.

predict Not yet run
unverified (harvested, no recorded run)
frederickszk/LRNet/demo/classify.py
code served (permissive licence) · get_code("ac93c18af42c2c35")
How it ran

no invocation recorded.

track_bidirectional Not yet run
unverified (harvested, no recorded run)
frederickszk/LRNet/demo/utils/deprecated.py
code served (permissive licence) · get_code("d0e3a6001a5c6f64")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 14 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 14 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

% &' " (&' " )*'+' " ,% $ % '-% ' ". /. 0 " 12'3 4 5 46786524 3!9% 9' :; <% 3 % <.!=' <='%-% '6>6 <? % @3 A % 4 (.!=' <='% .8=% ' / B A 4 % 4 4 67C3 4 % D9% 'EB 4 E E % < 9' :F4 3 *4 G63 A .H% $ % <*63 5'E? % @3 A % 4 (IH*?=='% J .K' <:6 <.8=% ' 0 L(M'N67CB' :OE 4 % P O6:'E1'4 'Q3 69A A % <.H*?P &LH?? % 4 :B 4 3 '4 % 6 'E86E E <.K' <:6 <.8=% ' RSTUVWXYZ [\]^_`^Ya b^cadeSaf`ghi`gjkl \][\j_e\`f[\\Ugj\mk ^]`^kejSgSaf`ghi`gjk n 9 39 oppqr stpuv uswx syz{|r}sṽ zv |u pz{yv pu {s x syu qsys sx pr szp | {p|x v v ys~|ypv yv p|u x pu v yv y u px v |uqx |wp}uu z{suv yr x v yp}py|rz|qx v { z|yr u v |y|rv yr |x }s v |y|xppyqwṽ zqsyv zx pv |upr r |x u r |xoppqr stpuv p|up pz v |y}sv yr|zu p|ysqqpsx syzprps x pu {v z{{spsxv u t|rwpv ywqsu u pwu| q{v u v zs p}syv qs v |ysũ |xpu v y{v {}|p~z|} qpv syupyu v v pypu u |y|v u ppu v pu {| |}v yp {p p}q|x s~r ps x pu|r}syv qs pv p|usypq|v {p}v u u v ~sy|qpypu v |ypqx |q|u psypr zv pysyx|wu r x s}p|x tys}ppr |xp pz v yoppqr stpuv p|u {x |{p}q|x s~}|pṽ y|yqx pzv u pp|}p x v zrps x pu y|p~zsṽ wx s v |y}|pvupv u p|py{syzp{pqx p zv u v |y|rp|}p x v zrps x pu }stv yv }|x pv u zx v }v ys v psys| u x ps}pzx x pypx s~p |x t v u z|yu x z pr|xu r zv pypq|v s v |y|r p}q|x s~r ps x pu |}qsx p |qx pv |u}p {|u |xqx |q|u p}p {|v u ṽ { px pv { psypsu v px |x sv y|x p|px |x}p {| {suu{|yx|wu ypu uvyp pz v y{v {z|}qx pu u p|x y|v u pz|x x q pv p|u x}|p~sz{v pp |y szp|x pyu v zu s su p psy{v pv {susx szpr ~p zṽ ypv yqpx r |x }syzp {pyrszpv {{v { z|}qx pu u pv p|u "

For agents

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

get_harvested_code_for_paper("2104.04480")
get_code_for_paper("2104.04480")
have("2104.04480")

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

curl https://syntology.ai/api/ran/2104.04480.json

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

[![Syntology run record](https://syntology.ai/api/ran/2104.04480.svg)](https://syntology.ai/paper/2104.04480)

Connect an agent — have() is free.