SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2106.06351 · CVPR · 2021

Part-aware Panoptic Segmentation

Chenyang Lu, Gijs Dubbelman, Daan De Geus, Panagiotis Meletis, Xiaoxiao Wen

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 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 7 have no contract verdict; 1 failure 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/2106.06351.json · badge: /api/ran/2106.06351.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
tue-mps/panoptic_parts canonical (classifier label)
classifier basis: census
10 ran · 1 no recorded run
FunctionStatusWhere it lives
color_map Ran
ran, honoured the contract we drafted
tue-mps/panoptic_parts/panoptic_parts/utils/utils.py
code served (permissive licence) · get_code("3219c4a8628db650")
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 2, dtype family int, shape (256, 3) 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

ARGS = {
    'N': 256,
    'normalized': False
}
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 2, dtype family int, shape (256, 3) — shape (256, 3) 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
encode_ids Ran
ran on a synthesized input
tue-mps/panoptic_parts/panoptic_parts/utils/format.py
code served (permissive licence) · get_code("4eb66c979842c5d8")
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 1, dtype family int, shape (3,) 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

ARGS = {
    "sids": np.array([5, 10, 15], dtype=np.int32),
    "iids": np.array([50, 100, 200], dtype=np.int32),
    "pids": np.array([3, 7, 12], dtype=np.int32),
}
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 1, dtype family int, shape (3,) — shape (3,) 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
filepaths_pairs_fn Ran
ran on a synthesized input
tue-mps/panoptic_parts/panoptic_parts/evaluation/experimental_eval_PartIOU.py
code served (permissive licence) · get_code("dcfcd463fb656dde")
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 2, 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 glob
import os.path as op
import tempfile
import os

# Create a temporary directory structure for testing
temp_dir = tempfile.mkdtemp()
gt_dir = op.join(temp_dir, 'gt')
pred_dir = op.join(temp_dir, 'pred')
os.makedirs(gt_dir, exist_ok=True)
os.makedirs(pred_dir, exist_ok=True)

# Create some dummy ground truth files
for i in range(3):
    gt_file = op.join(gt_dir, f'image_{i}.txt')
    with open(gt_file, 'w') as f:
        f.write(f'dummy gt content {i}')

# Create corresponding prediction files
for i in range(3):
    pred_file = op.join(pred_dir, f'image_{i}.tif')
    with open(pred_file, 'w') as f:
        f.write(f'dummy pred content {i}')

# Set up the function arguments
filepath_pattern_gt = op.join(gt_dir, '*.txt')
basepath_pred = pred_dir

ARGS = {
    'filepath_pattern_gt': filepath_pattern_gt,
    'basepath_pred': basepath_pred,
}
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 2, 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
filepaths_pairs_fn_OLD Ran
ran; our contract draft was wrong, not the code
tue-mps/panoptic_parts/panoptic_parts/evaluation/eval_PartPQ.py
code served (permissive licence) · get_code("c6704fa8b1518821")
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) and returned list, ndim 2, 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 os
import tempfile
import glob

# Create temporary directories and files for testing
tmpdir_gt = tempfile.mkdtemp()
tmpdir_pred = tempfile.mkdtemp()

# Set up for Cityscapes Panoptic Parts dataset structure
# Create subdirectories and files matching the expected pattern
cityscapes_subdir = os.path.join(tmpdir_gt, 'frankfurt')
os.makedirs(cityscapes_subdir, exist_ok=True)

# Create a ground truth file with the expected naming pattern
# Pattern: *_gtFine_panopticIds.tif (23 characters at end)
gt_filename_cityscapes = 'frankfurt_000000_000019_gtFine_panopticIds.tif'
gt_filepath_cityscapes = os.path.join(cityscapes_subdir, gt_filename_cityscapes)
with open(gt_filepath_cityscapes, 'w') as f:
    f.write('dummy')

# Create corresponding prediction file
pred_filename_cityscapes = 'frankfurt_000000_000019_gtFine_leftImg8bit.png'
pred_filepath_cityscapes = os.path.join(tmpdir_pred, pred_filename_cityscapes)
with open(pred_filepath_cityscapes, 'w') as f:
    f.write('dummy')

# Set up for PASCAL Panoptic Parts dataset structure
# Create ground truth file with .tif extension
gt_filename_pascal = '2010_000002_gtFine_panopticIds.tif'
gt_filepath_pascal = os.path.join(tmpdir_gt, gt_filename_pascal)
with open(gt_filepath_pascal, 'w') as f:
    f.write('dummy')

# Create corresponding prediction file
pred_filename_pascal = '2010_000002.png'
pred_filepath_pascal = os.path.join(tmpdir_pred, pred_filename_pascal)
with open(pred_filepath_pascal, 'w') as f:
    f.write('dummy')

# Test with Cityscapes
ARGS = {
    'dst_name': 'Cityscapes Panoptic Parts',
    'basepath_gt': tmpdir_gt,
    'basepath_pred': tmpdir_pred,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 2, 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
parse_dataset_sid_pid2eval_sid_pid Ran
ran on a synthesized input
tue-mps/panoptic_parts/panoptic_parts/utils/evaluation_PartPQ.py
code served (permissive licence) · get_code("cb4beed10cfa8fcf")
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 torch
import numpy as np

# Analyze the function:
# - It takes dataset_sid_pid2eval_sid_pid which is a dict
# - It calls .get(k, k) on it, so it needs to support dict.get()
# - The dict maps integers (sid_pid) to either integers or the string 'IGNORED'
# - It iterates k from 0 to 9999
# - It expects all values in the output to be in range(10000)

# Create a sparse mapping that covers some cases
dataset_sid_pid2eval_sid_pid = {
    0: 5,           # maps sid_pid 0 to 5
    10: 20,         # maps sid_pid 10 to 20
    100: 'IGNORED', # maps sid_pid 100 to 'IGNORED' (should become 100)
    5000: 9999,     # maps sid_pid 5000 to 9999
}

ARGS = {
    "dataset_sid_pid2eval_sid_pid": dataset_sid_pid2eval_sid_pid
}
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 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
parse_sid_pid2eval_id Ran
ran on a synthesized input
tue-mps/panoptic_parts/panoptic_parts/utils/experimental_evaluation_IOU.py
code served (permissive licence) · get_code("8dc6717ccc50757a")
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 torch
import numpy as np

torch.manual_seed(0)

# Analyzing the function parse_sid_pid2eval_id:
# - Takes sid_pid2eval_id: Dict[int, int] - a mapping dictionary
# - Takes max_sid: int - maximum session id
# - The function populates and transforms the dictionary
# - Final assertion: set of keys should equal range((max_sid + 1) * 100)
# - This means we need (max_sid + 1) * 100 keys total in the final dict

# Let's construct a minimal valid input:
# max_sid = 1 means we need (1 + 1) * 100 = 200 keys in the final dict
# We'll start with a simple mapping that has -1 as a wildcard

max_sid = 1

# Start with a dict that has -1 as a catch-all eval_id
# and some specific mappings
sid_pid2eval_id = {
    -1: 5,  # catch-all evaluation id
    0: 10,  # specific eval id for session 0
}

ARGS = {
    "sid_pid2eval_id": sid_pid2eval_id,
    "max_sid": max_sid,
}
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 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
pred_reader_fn Ran
ran, honoured the contract we drafted
tue-mps/panoptic_parts/panoptic_parts/evaluation/eval_PartPQ.py
code served (permissive licence) · get_code("8e9c1d96b586dcc6")
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) and returned tuple, ndim 3, 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.

import numpy as np
from PIL import Image
import tempfile
import os

# Create a temporary 3-channel image file in the format expected by pred_reader_fn
# The function expects a 3-channel image where:
# - Channel 0: pan_classes
# - Channel 1: pan_inst_ids
# - Channel 2: parts_output

# Create a small 4x3 image with 3 channels
height, width = 4, 3
pan_classes_data = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12]], dtype=np.int32)
pan_inst_ids_data = np.array([[10, 20, 30], [40, 50, 60], [70, 80, 90], [100, 110, 120]], dtype=np.int32)
parts_output_data = np.array([[100, 101, 102], [103, 104, 105], [106, 107, 108], [109, 110, 111]], dtype=np.int32)

# Stack into a 3-channel image
image_array = np.stack([pan_classes_data, pan_inst_ids_data, parts_output_data], axis=-1)

# Convert to uint8 for PIL (will be read back as int32 by the function)
image_array_uint8 = image_array.astype(np.uint8)

# Create a temporary file
temp_dir = tempfile.gettempdir()
temp_file = os.path.join(temp_dir, "test_pred_image.png")

# Save the image
img = Image.fromarray(image_array_uint8, mode='RGB')
img.save(temp_file)

# Define ARGS with the filepath
ARGS = {
    "fp_pred": temp_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
PIL, numpy — 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 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
prediction_parsing Ran
ran on a synthesized input
tue-mps/panoptic_parts/panoptic_parts/utils/evaluation_PartPQ.py
code served (permissive licence) · get_code("9cae17e3342feb5e")
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 numpy as np

# Create small test maps (4x3 dimensions)
height, width = 4, 3

# sem_map: semantic class map with values corresponding to semantic classes
sem_map = np.array([
    [24, 24, 25],
    [24, 25, 25],
    [26, 26, 27],
    [26, 27, 28]
], dtype=np.int32)

# inst_map: instance ID map 
inst_map = np.array([
    [1, 1, 2],
    [1, 2, 2],
    [3, 3, 4],
    [3, 4, 4]
], dtype=np.int32)

# part_map: part segmentation map with part class IDs
part_map = np.array([
    [1, 1, 2],
    [1, 2, 2],
    [1, 1, 3],
    [1, 3, 3]
], dtype=np.int32)

# cat_definition: category definition with semantic classes and parts
cat_definition = {
    "num_cats": 2,
    "cat_def": [
        {
            "sem_cls": [24, 25],
            "parts_cls": [1, 2, 3, 4]
        },
        {
            "sem_cls": [26, 27, 28],
            "parts_cls": [1, 2, 3, 4, 5]
        }
    ]
}

ARGS = {
    "cat_definition": cat_definition,
    "sem_map": sem_map,
    "inst_map": inst_map,
    "part_map": part_map
}
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 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
random_colors Ran
ran on a synthesized input
tue-mps/panoptic_parts/panoptic_parts/utils/visualization.py
code served (permissive licence) · get_code("edb4c38e39147282")
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 2, 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.

import numpy as np

ARGS = {
    "num": 5
}
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 list, ndim 2, dtype family int — 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
safe_write Ran
ran on a synthesized input
tue-mps/panoptic_parts/panoptic_parts/utils/utils.py
code served (permissive licence) · get_code("d29f58cb43eb1a91")
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 numpy as np
import tempfile
import os

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

# Create a small valid numpy image (RGB format)
# PIL.Image.fromarray expects a numpy array
# For RGB image: shape should be (height, width, 3) with uint8 dtype
image = np.random.randint(0, 256, size=(4, 3, 3), dtype=np.uint8)

# Create a path that doesn't exist
path = os.path.join(temp_dir, "subdir", "test_image.png")

ARGS = {
    "path": path,
    "image": image,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — 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
uids_lids2uids_cids Not yet run
unverified (harvested, no recorded run)
tue-mps/panoptic_parts/panoptic_parts/utils/utils.py
code served (permissive licence) · get_code("918c7ea59b75cb7a")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 3 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 1 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.

FunctionAlso present inEarliest-datedOthers named
color_map 9 other papers
6 dated earlier, 3 later
1703.01780 (2017-03)
Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results
1802.07934 (2018-02)
Adversarial Learning for Semi-Supervised Semantic Segmentation
1908.05724 (2019-08)
Semi-Supervised Semantic Segmentation with High- and Low-level Consistency
and 6 more not named here
filepaths_pairs_fn_OLD 1 other paper
1 dated earlier
2004.07944 (2020-04)
Cityscapes-Panoptic-Parts and PASCAL-Panoptic-Parts datasets for Scene Understanding
pred_reader_fn 1 other paper
1 dated earlier
2004.07944 (2020-04)
Cityscapes-Panoptic-Parts and PASCAL-Panoptic-Parts datasets for Scene Understanding

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

In this work, we introduce the new scene understanding task of Part-aware Panoptic Segmentation (PPS), which aims to understand a scene at multiple levels of abstraction, and unifies the tasks of scene parsing and part parsing. For this novel task, we provide consistent annotations on two commonly used datasets: Cityscapes and Pascal VOC. Moreover, we present a single metric to evaluate PPS, called Part-aware Panoptic Quality (PartPQ). For this new task, using the metric and annotations, we set multiple baselines by merging results of existing state-of-the-art methods for panoptic segmentation and part segmentation. Finally, we conduct several experiments that evaluate the importance of the different levels of abstraction in this single task.

For agents

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

get_harvested_code_for_paper("2106.06351")
get_code_for_paper("2106.06351")
have("2106.06351")

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

curl https://syntology.ai/api/ran/2106.06351.json

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

[![Syntology run record](https://syntology.ai/api/ran/2106.06351.svg)](https://syntology.ai/paper/2106.06351)

Connect an agent — have() is free.