SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1606.03476 · 2016

Generative Adversarial Imitation Learning

Stefano Ermon, Jonathan Ho

arXiv · PDF · Open in the Atlas

Code that ran

24 of 36 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (6 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 1 violated it and 17 have no contract verdict; 4 failures of our instrument; 12 no recorded run), measured 2026-11-21; 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/1606.03476.json · badge: /api/ran/1606.03476.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
Techget/gail-tf-sc2 community (classifier label)
classifier basis not recorded on the link
9 ran · 3 no recorded run
Kaixhin/imitation-learning community
Papers with Code archive
5 ran · 2 no recorded run
bukysun/gail-traj-eb community (classifier label)
classifier basis not recorded on the link
4 ran · 0 no recorded run
emunaran/stochastic-human-driving-policies-drl community
Papers with Code archive
3 ran · 1 no recorded run
morikatron/GAIL_PPO community
Papers with Code archive
1 ran · 0 no recorded run
twni2016/f-IRL community
Papers with Code archive
1 ran · 0 no recorded run
Div99/IQ-Learn community
Papers with Code archive
1 ran · 0 no recorded run
KAIST-AILab/deeprl_practice_colab community
Papers with Code archive
0 ran · 5 no recorded run
nav74neet/gail-tf-gym community
Papers with Code archive
0 ran · 1 no recorded run
FunctionStatusWhere it lives
ActorCritic Ran
ran on a synthesized input · constructed an object; no result was computed
emunaran/stochastic-human-driving-policies-drl/model/algorithm/gail.py
pointer only (licence: NONE) · get_code("cb3f4da86ee1b680")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, emunaran/stochastic-human-driving-policies-drl when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Discriminator Ran
ran on a synthesized input · constructed an object; no result was computed
emunaran/stochastic-human-driving-policies-drl/model/algorithm/gail.py
pointer only (licence: NONE) · get_code("86a2d590f55ea39d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, emunaran/stochastic-human-driving-policies-drl when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Discriminator Ran
ran on a synthesized input
morikatron/GAIL_PPO/algo/discriminator.py
code served (permissive licence) · get_code("c30fb27c55c474d2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) 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 tensorflow as tf
import numpy as np

EPS = 1e-08

class Discriminator(tf.Module):
    """
    ニューラルネットワークが絡む推論、学習の2つの計算を行うクラスです。
    @tf.functionデコレータは実行時に関数をグラフにコンパイルしてくれるもので、学習の高速化・GPUでの実行を可能にします。
    """

    def __init__(self, num_obs, num_actions, config):
        super(Discriminator, self).__init__(name='discriminator')
        self.num_actions = num_actions
        self.input_shape = (num_obs + num_actions,)
        self.batch_size = config.batch_size
        self.config = config
        self.network = self._build_model()
        self.optimizer = tf.keras.optimizers.Adam(learning_rate=config.learning_rate)
        self.gradient_clip = config.gradient_clip

    @tf.function
    def inference(self, obs, action):
        one_hot_action = tf.one_hot(action, self.num_actions, dtype=tf.float32)
        agent_input = tf.concat([obs, one_hot_action], axis=1)
        p = self.network(agent_input)
        reward = p
        return reward

    @tf.function
    def train(self, demo_obs, demo_act, agent_obs, agent_act):
        with tf.GradientTape() as tape:
            agent_one_hot_act = tf.one_hot(agent_act, self.num_actions)
            demo_one_hot_act = tf.one_hot(demo_act, self.num_actions)
            agent_input = tf.concat([agent_obs, agent_one_hot_act], axis=1)
            demo_input = tf.concat([demo_obs, demo_one_hot_act], axis=1)
            agent_p = self.network(agent_input)
            demo_p = self.network(demo_input)
            agent_loss = -tf.reduce_mean(tf.math.log(1.0 - agent_p + EPS))
            demo_loss = -tf.reduce_mean(tf.math.log(demo_p + EPS))
            loss = agent_loss + demo_loss
        grads = tape.gradient(loss, self.network.trainable_variables)
        if self.gradient_clip is not None:
            clipped_grads = []
            for grad in grads:
                clipped_grads.append(tf.clip_by_norm(grad, self.gradient_clip))
            grads = clipped_grads
        agent_acc = tf.reduce_mean(tf.cast(agent_p < 0.5, dtype=tf.float32))
        demo_acc = tf.reduce_mean(tf.cast(demo_p > 0.5, dtype=tf.float32))
        grads_and_vars = zip(grads, self.network.trainable_variables)
        self.optimizer.apply_gradients(grads_and_vars)
        return (loss, agent_loss, demo_loss, agent_acc, demo_acc)

    def _build_model(self):
        input_x = tf.keras.Input(shape=self.input_shape)
        h1 = tf.keras.layers.Dense(units=self.config.num_units, activation='relu')(input_x)
        h2 = tf.keras.layers.Dense(units=self.config.num_units, activation='relu')(h1)
        p = tf.keras.layers.Dense(1, activation='sigmoid')(h2)
        model = tf.keras.Model(inputs=input_x, outputs=p)
        return model


class Config:
    def __init__(self, batch_size, learning_rate, num_units, gradient_clip):
        self.batch_size = batch_size
        self.learning_rate = learning_rate
        self.num_units = num_units
        self.gradient_clip = gradient_clip


config = Config(batch_size=2, learning_rate=0.001, num_units=8, gradient_clip=1.0)

INIT_ARGS = {
    'num_obs': 4,
    'num_actions': 3,
    'config': config
}

ARGS = {
    'demo_obs': tf.constant(np.random.randn(2, 4).astype(np.float32)),
    'demo_act': tf.constant(np.array([0, 1], dtype=np.int32)),
    'agent_obs': tf.constant(np.random.randn(2, 4).astype(np.float32)),
    'agent_act': tf.constant(np.array([2, 0], dtype=np.int32))
}

METHOD = 'train'

discriminator = Discriminator(**INIT_ARGS)
result = getattr(discriminator, METHOD)(**ARGS)
print("Training result:", result)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, tensorflow — 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
GAILDiscriminator Ran
ran on a synthesized input · constructed an object; no result was computed
Kaixhin/imitation-learning/training.py
code served (permissive licence) · get_code("58c499108eff7753")
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 an instance of its own class, GAILDiscriminator: the run constructed the object and did not call it, so no result was computed. 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
from omegaconf import DictConfig

torch.manual_seed(0)

# Create a minimal DictConfig for imitation_cfg
imitation_cfg = DictConfig({
    'state_only': False,
    'spectral_norm': False,
    'discriminator': {
        'hidden_size': 8,
        'depth': 2,
        'activation': 'relu',
        'reward_shaping': True,
        'subtract_log_policy': False,
        'reward_function': 'GAIL'
    }
})

state_size = 4
action_size = 3
discount = 0.99

ARGS = {
    'state_size': state_size,
    'action_size': action_size,
    'imitation_cfg': imitation_cfg,
    'discount': discount
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
omegaconf, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type GAILDiscriminator — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a GAILDiscriminator 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
QuarterSizeSigmoid Ran
ran on a synthesized input · constructed an object; no result was computed
emunaran/stochastic-human-driving-policies-drl/model/algorithm/gail.py
pointer only (licence: NONE) · get_code("be4ed3c2e906962c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, emunaran/stochastic-human-driving-policies-drl when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ResNetAIRLDisc Ran
ran on a synthesized input · constructed an object; no result was computed
twni2016/f-IRL/firl/models/discrim.py
code served (permissive licence) · get_code("93ab130e938732f0")
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 an instance of its own class, ResNetAIRLDisc: the run constructed the object and did not call it, so no result was computed. 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)

ARGS = {
    'input_dim': 4,
    'num_layer_blocks': 2,
    'hid_dim': 100,
    'hid_act': 'relu',
    'use_bn': True,
    'clamp_magnitude': 10.0,
}
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 ResNetAIRLDisc — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ResNetAIRLDisc 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
SoftActor Ran
ran on a synthesized input · constructed an object; no result was computed
Kaixhin/imitation-learning/training.py
code served (permissive licence) · get_code("d0f9a51097534c31")
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 an instance of its own class, SoftActor: the run constructed the object and did not call it, so no result was computed. 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
from omegaconf import DictConfig

torch.manual_seed(0)

# Create a minimal DictConfig for model_cfg
model_cfg = DictConfig({
    'hidden_size': 8,
    'depth': 2,
    'activation': 'relu',
    'input_dropout': 0.1,
    'dropout': 0.05
})

# Define ARGS with all parameters for SoftActor
ARGS = {
    'state_size': 4,
    'action_size': 3,
    'model_cfg': model_cfg
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
omegaconf, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type SoftActor — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a SoftActor 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
_create_fcnn Ran
ran; our contract draft was wrong, not the code
Kaixhin/imitation-learning/training.py
code served (permissive licence) · get_code("ee98cbfd5f8b7c6d")
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 Sequential. 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
from torch.nn.utils import parametrizations

torch.manual_seed(0)

ARGS = {
    'input_size': 4,
    'hidden_size': 8,
    'depth': 2,
    'output_size': 3,
    'activation_function': 'relu',
    'input_dropout': 0.1,
    'dropout': 0.1,
    'final_gain': 1.0,
    'spectral_norm': False
}
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 Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential output the recorded ndim 1 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
_join_state_action Ran
ran, honoured the contract we drafted
Kaixhin/imitation-learning/training.py
code served (permissive licence) · get_code("e374e2ee3dbb3f45")
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 Tensor, ndim 2, dtype family float, shape (2, 7) 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
from torch import Tensor

torch.manual_seed(0)

# Create small tensors for state and action
# state: batch_size=2, state_dim=4
state = torch.randn(2, 4)

# action: batch_size=2, action_dim=3
action = torch.randn(2, 3)

ARGS = {
    "state": state,
    "action": action,
}
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 Tensor, ndim 2, dtype family float, shape (2, 7) — shape (2, 7) 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
_mix_vars Ran
ran, violated the contract we drafted
Kaixhin/imitation-learning/training.py
code served (permissive licence) · get_code("df69a6d635ac78ad")
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 Tensor, ndim 2, dtype family float, shape (2, 4) 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
from torch import Tensor

torch.manual_seed(0)

# Construct tensors for the function
# x_1 and x_2 should be compatible for broadcasting with eps
# The function checks if x_1.ndim == 2 to decide unsqueeze behavior
# Let's create 2D tensors (batch_size=2, features=4)
x_1 = torch.randn(2, 4)
x_2 = torch.randn(2, 4)

# eps should be a 1D tensor of shape (batch_size,) or (batch_size, 1)
# The function unsqueezes it to (batch_size, 1) if x_1.ndim == 2
eps = torch.rand(2)

ARGS = {
    "x_1": x_1,
    "x_2": x_2,
    "eps": eps,
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) 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
find_task_for_env_id_in_any_benchmark Ran
ran; our contract draft was wrong, not the code
Techget/gail-tf-sc2/gailtf/baselines/bench/benchmarks.py
pointer only (licence: MIT); the stored copy get_code serves, from seungyulhan/disc, is pointer only (licence: MIT) · get_code("71bdb8bae1ed81f7")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from seungyulhan/disc, and MIT, as GitHub detects it on that repository, is in the permissive set, but a licence decision keeps the body a pointer: its licence file names an outside party, so it may cover vendored code rather than the repository owner's files (recorded as pointer:licence_names_third_party_corporation). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

flatten_lists Ran
ran, honoured the contract we drafted
bukysun/gail-traj-eb/gailtf/algo/trpo_mpi.py
pointer only (licence: MIT); the stored copy get_code serves, from Jekyll1021/RARL, is pointer only (licence: NONE) · get_code("0173e4280d9a9884")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from Jekyll1021/RARL, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

flatten_param Ran
ran on a synthesized input
Techget/gail-tf-sc2/gailtf/algo/trpo_mpi.py
code served (permissive licence) · get_code("c45e54fc5355bc2d")
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 (12,) 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

# The function flatten_param takes a param argument and:
# 1. Converts it to a numpy array
# 2. Flattens it
# 3. Returns the flattened array

# We need to provide a value for 'param' that can be converted to a numpy array
# and flattened. We'll use a multi-dimensional array to make the flattening meaningful.

param_value = np.array([[1, 2, 3, 4],
                        [5, 6, 7, 8],
                        [9, 10, 11, 12]])

ARGS = {
    "param": param_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — 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 (12,) — shape (12,) 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_cur_ep_rewards Ran
ran on a synthesized input
bukysun/gail-traj-eb/gailtf/algo/trpo_traj_mpi.py
code served (permissive licence) · get_code("8bb63d18cbda13c3")
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 tensorflow as tf

class MockDiscriminator:
    def get_reward(self, traj, traj_len):
        """Returns rewards for the trajectory"""
        # Return rewards with shape (traj_len,) or (traj_len, 1)
        return np.random.randn(traj_len, 1)

# Create small test data
obs_dim = 4
acs_dim = 3
traj_len = 2
seq_length = 5

# obs: (traj_len, obs_dim)
obs = np.random.randn(traj_len, obs_dim)

# acs: (traj_len, acs_dim)
acs = np.random.randn(traj_len, acs_dim)

# Create discriminator
discriminator = MockDiscriminator()

ARGS = {
    "discriminator": discriminator,
    "obs": obs,
    "acs": acs,
    "seq_length": seq_length,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — 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_monitor_files Ran
ran on a synthesized input
Techget/gail-tf-sc2/gailtf/baselines/bench/monitor.py
code served (permissive licence) · get_code("af3d493bdc17ad51")
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 tempfile
import os
import json

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

# Create some dummy monitor files for testing
monitor_file_1 = os.path.join(temp_dir, 'run1.monitor.json')
monitor_file_2 = os.path.join(temp_dir, 'run2.monitor.json')

# Write some dummy JSON lines to the monitor files
with open(monitor_file_1, 'w') as f:
    f.write(json.dumps({'t_start': 0, 'gym_version': '0.17.0', 'env_id': 'CartPole-v0'}) + '\n')
    f.write(json.dumps({'r': 10.0, 'l': 5, 't': 1.5}) + '\n')

with open(monitor_file_2, 'w') as f:
    f.write(json.dumps({'t_start': 0, 'gym_version': '0.17.0', 'env_id': 'CartPole-v0'}) + '\n')
    f.write(json.dumps({'r': 15.0, 'l': 8, 't': 2.0}) + '\n')

ARGS = {
    'dir': temp_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
gym — 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
get_task Ran
ran; our contract draft was wrong, not the code
Techget/gail-tf-sc2/gailtf/baselines/bench/benchmarks.py
pointer only (licence: MIT); the stored copy get_code serves, from seungyulhan/disc, is pointer only (licence: MIT) · get_code("3e82b9de30b1625f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from seungyulhan/disc, and MIT, as GitHub detects it on that repository, is in the permissive set, but a licence decision keeps the body a pointer: its licence file names an outside party, so it may cover vendored code rather than the repository owner's files (recorded as pointer:licence_names_third_party_corporation). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_task_name Ran
ran on a synthesized input
bukysun/gail-traj-eb/main_traj_eb.py
code served (permissive licence) · get_code("7d3ce86ca78ceba8")
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 numpy as np

class Args:
    def __init__(self):
        self.algo = 'bc'
        self.traj_limitation = np.inf
        self.env_id = 'Walker2d-v2'
        self.pretrained = False
        self.ret_threshold = 0
        self.g_step = 3
        self.d_step = 1
        self.policy_entcoeff = 0.001
        self.adversary_entcoeff = 0.001

ARGS = {
    'args': Args()
}
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 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
iq_loss Ran
ran; our contract draft was wrong, not the code
Div99/IQ-Learn/iq_learn/iq.py
pointer only (licence: NOASSERTION) · get_code("cf7c0b7b5de1e98b")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, Div99/IQ-Learn (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

logit_bernoulli_entropy Ran
ran on a synthesized input
Techget/gail-tf-sc2/gailtf/common/tf_util.py
code served (permissive licence) · get_code("2e4f14f2d4a888ff")
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 EagerTensor, ndim 2, dtype family float, shape (2, 4) 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 tensorflow as tf
import numpy as np

def logsigmoid(a):
    """Equivalent to tf.log(tf.sigmoid(a))"""
    return -tf.nn.softplus(-a)

def logit_bernoulli_entropy(logits):
    ent = (1.0 - tf.nn.sigmoid(logits)) * logits - logsigmoid(logits)
    return ent

# Create a small tensor of logits with shape (2, 4)
logits_np = np.random.randn(2, 4).astype(np.float32)
logits = tf.convert_to_tensor(logits_np)

ARGS = {
    "logits": logits
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) 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
logsigmoid Ran
ran on a synthesized input
Techget/gail-tf-sc2/gailtf/common/tf_util.py
code served (permissive licence) · get_code("3032d5dbb694f8ba")
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 EagerTensor, ndim 2, dtype family float, shape (2, 4) 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 tensorflow as tf
import numpy as np

def logsigmoid(a):
    """Equivalent to tf.log(tf.sigmoid(a))"""
    return -tf.nn.softplus(-a)

# Create a small TensorFlow tensor for the parameter 'a'
# Using shape (2, 4) for a tiny batch
a_value = tf.constant([
    [0.5, -0.3, 1.2, -0.8],
    [0.1, 0.9, -0.5, 0.2]
], dtype=tf.float32)

ARGS = {
    "a": a_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) 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
make_output_format Ran
ran on a synthesized input
Techget/gail-tf-sc2/gailtf/baselines/logger.py
code served (permissive licence) · get_code("071e746ec3f03d74")
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 HumanOutputFormat. 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 sys
import tempfile
import os

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

ARGS = {
    'format': 'stdout',
    'ev_dir': temp_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type HumanOutputFormat — a HumanOutputFormat output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a HumanOutputFormat 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
process_coordinates_param_nn_output Ran
ran on a synthesized input
Techget/gail-tf-sc2/gailtf/algo/trpo_mpi.py
code served (permissive licence) · get_code("a9fed3a61f50d4f0")
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 float, shape (8,) 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 = {
    "coordinate": np.array([[45.7, 32.2, 10.5, 55.3], [5.1, 20.9, 60.2, 15.8]])
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — 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 float, shape (8,) — shape (8,) 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
traj2trans Ran
ran on a synthesized input
bukysun/gail-traj-eb/gailtf/algo/trpo_traj_mpi.py
code served (permissive licence) · get_code("ae6c5ba444e45a52")
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 numpy as np

# Construct small test trajectories
# Each trajectory is a 2D array where columns are [observations..., actions...]
# We'll use num_ob=3, so first 3 columns are observations, remaining are actions

# Trajectory 1: shape (5, 6) - 5 timesteps, 3 obs + 3 actions
traj1 = np.random.randn(5, 6).astype(np.float32)

# Trajectory 2: shape (4, 6) - 4 timesteps, 3 obs + 3 actions
traj2 = np.random.randn(4, 6).astype(np.float32)

trajs = [traj1, traj2]
traj_lens = [5, 4]
num_ob = 3

ARGS = {
    "trajs": trajs,
    "traj_lens": traj_lens,
    "num_ob": num_ob
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — 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
wrap_deepmind Ran
ran on a synthesized input
Techget/gail-tf-sc2/gailtf/baselines/common/atari_wrappers.py
code served (permissive licence) · get_code("6ad2196300b77d83")
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 ClipRewardEnv. 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 gym
from gym import spaces
import numpy as np

# Create a mock Atari environment
class MockAtariEnv(gym.Env):
    def __init__(self):
        self.spec = type('obj', (object,), {'id': 'PongNoFrameskip-v4'})()
        self.observation_space = spaces.Box(low=0, high=255, shape=(210, 160, 3), dtype=np.uint8)
        self.action_space = spaces.Discrete(6)
        self._step_count = 0
        self.ale = type('obj', (object,), {'lives': lambda: 5})()
    
    def reset(self):
        self._step_count = 0
        return np.random.randint(0, 256, (210, 160, 3), dtype=np.uint8)
    
    def step(self, action):
        self._step_count += 1
        obs = np.random.randint(0, 256, (210, 160, 3), dtype=np.uint8)
        reward = 1.0
        done = self._step_count > 100
        info = {}
        return obs, reward, done, info
    
    def get_action_meanings(self):
        return ['NOOP', 'FIRE', 'UP', 'DOWN', 'LEFT', 'RIGHT']
    
    def render(self, mode='human'):
        pass
    
    @property
    def unwrapped(self):
        return self

# Create the environment
env = MockAtariEnv()

ARGS = {
    'env': env,
    'episode_life': True,
    'clip_rewards': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, gym, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ClipRewardEnv — a ClipRewardEnv output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ClipRewardEnv 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
Discriminator Not yet run
unverified (harvested, no recorded run)
nav74neet/gail-tf-gym/gail-ppo-tf-gym/network_models/discriminator.py
pointer only (licence: NONE) · get_code("40b17d8ecf42cfe6")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, nav74neet/gail-tf-gym when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

GAIL Not yet run
unverified (harvested, no recorded run)
emunaran/stochastic-human-driving-policies-drl/model/algorithm/gail.py
pointer only (licence: NONE) · get_code("654066a3d6f0d55b")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, emunaran/stochastic-human-driving-policies-drl when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

RaggedArray Not yet run
unverified (harvested, no recorded run)
KAIST-AILab/deeprl_practice_colab/day3/gail/policyopt/rl.py
pointer only (licence: NONE) · get_code("0654ba2a385cf2c4")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, KAIST-AILab/deeprl_practice_colab when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

SamplingPolicyOptimizer Not yet run
unverified (harvested, no recorded run)
KAIST-AILab/deeprl_practice_colab/day3/gail/policyopt/rl.py
pointer only (licence: NONE) · get_code("4117e20071c70a04")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, KAIST-AILab/deeprl_practice_colab when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

adversarial_imitation_update Not yet run
unverified (harvested, no recorded run)
Kaixhin/imitation-learning/training.py
code served (permissive licence) · get_code("75e9f2f5049574cb")
How it ran

no invocation recorded.

compute_advantage Not yet run
unverified (harvested, no recorded run)
KAIST-AILab/deeprl_practice_colab/day3/gail/policyopt/rl.py
pointer only (licence: NONE) · get_code("12a65e5d7d9e912d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, KAIST-AILab/deeprl_practice_colab when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

compute_qvals Not yet run
unverified (harvested, no recorded run)
KAIST-AILab/deeprl_practice_colab/day3/gail/policyopt/rl.py
pointer only (licence: NONE) · get_code("5ce9c8b1227054f5")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, KAIST-AILab/deeprl_practice_colab when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

get_benchmark Not yet run
unverified (harvested, no recorded run)
Techget/gail-tf-sc2/gailtf/baselines/bench/benchmarks.py
pointer only (licence: MIT); the stored copy get_code serves, from seungyulhan/disc, is pointer only (licence: MIT) · get_code("7bd934aa47da390a")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from seungyulhan/disc, and MIT, as GitHub detects it on that repository, is in the permissive set, but a licence decision keeps the body a pointer: its licence file names an outside party, so it may cover vendored code rather than the repository owner's files (recorded as pointer:licence_names_third_party_corporation). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

load_results Not yet run
unverified (harvested, no recorded run)
Techget/gail-tf-sc2/gailtf/baselines/bench/monitor.py
code served (permissive licence) · get_code("0c09148efe18bf06")
How it ran

no invocation recorded.

make_gail_input Not yet run
unverified (harvested, no recorded run)
Kaixhin/imitation-learning/training.py
code served (permissive licence) · get_code("6782b3c2166c0b4f")
How it ran

no invocation recorded.

raggedstack Not yet run
unverified (harvested, no recorded run)
KAIST-AILab/deeprl_practice_colab/day3/gail/policyopt/rl.py
pointer only (licence: NONE) · get_code("343ea6d8bbc7b6dc")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, KAIST-AILab/deeprl_practice_colab when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

wrap_dqn Not yet run
unverified (harvested, no recorded run)
Techget/gail-tf-sc2/gailtf/baselines/common/atari_wrappers_deprecated.py
code served (permissive licence) · get_code("7e4f579c2aabd9b5")
How it ran

no invocation recorded.

Also present in other papers' code

Of 36 samples lifted from this paper's linked repositories, 36 carry a function identity and 0 do not. Of the 36 distinct functions with an identity, 9 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 9 later) and 27 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
flatten_lists 8 other papers
8 later
1703.02702 (2017-03)
Robust Adversarial Reinforcement Learning
1801.01432 (2018-01)
Jointly Learning to Construct and Control Agents using Deep Reinforcement Learning
1905.09191 (2019-05)
Learning Robust Options by Conditional Value at Risk Optimization
and 5 more not named here
find_task_for_env_id_in_any_benchmark 5 other papers
1 dated earlier, 4 later
1604.06778 (2016-04)
Benchmarking Deep Reinforcement Learning for Continuous Control
1905.02363 (2019-05)
Dimension-Wise Importance Sampling Weight Clipping for Sample-Efficient Reinforcement Learning
2001.03415 (2020-01)
Multi-Agent Interactions Modeling with Correlated Policies
and 2 more not named here
get_benchmark 5 other papers
1 dated earlier, 4 later
1604.06778 (2016-04)
Benchmarking Deep Reinforcement Learning for Continuous Control
1905.02363 (2019-05)
Dimension-Wise Importance Sampling Weight Clipping for Sample-Efficient Reinforcement Learning
2001.03415 (2020-01)
Multi-Agent Interactions Modeling with Correlated Policies
and 2 more not named here
get_task 5 other papers
1 dated earlier, 4 later
1604.06778 (2016-04)
Benchmarking Deep Reinforcement Learning for Continuous Control
1905.02363 (2019-05)
Dimension-Wise Importance Sampling Weight Clipping for Sample-Efficient Reinforcement Learning
2001.03415 (2020-01)
Multi-Agent Interactions Modeling with Correlated Policies
and 2 more not named here
make_output_format 2 other papers
2 later
1710.11198 (2017-10)
Action-depedent Control Variates for Policy Optimization via Stein's Identity
1710.11417 (2017-10)
TreeQN and ATreeC: Differentiable Tree-Structured Models for Deep Reinforcement Learning
wrap_dqn 2 other papers
2 later
1710.00814 (2017-10)
Detecting Adversarial Attacks on Neural Network Policies with Visual Foresight
2105.06022 (2021-05)
Principled Exploration via Optimistic Bootstrapping and Backward Induction
SoftActor
brick of this paper's assembly: helper
1 other paper
1 later
1801.01290 (2018-01)
Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
_create_fcnn
brick of this paper's assembly: helper
1 other paper
1 later
1801.01290 (2018-01)
Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
iq_loss
brick of this paper's assembly: entry
1 other paper
1 later
2602.10793 (2026-02)
Semi-Supervised Cross-Domain Imitation Learning

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 2026-11-21.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Consider learning a policy from example expert behavior, without interaction with the expert or access to reinforcement signal. One approach is to recover the expert's cost function with inverse reinforcement learning, then extract a policy from that cost function with reinforcement learning. This approach is indirect and can be slow. We propose a new general framework for directly extracting a policy from data, as if it were obtained by reinforcement learning following inverse reinforcement learning. We show that a certain instantiation of our framework draws an analogy between imitation learning and generative adversarial networks, from which we derive a model-free imitation learning algorithm that obtains significant performance gains over existing model-free methods in imitating complex behaviors in large, high-dimensional environments.

For agents

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

get_harvested_code_for_paper("1606.03476")
get_code_for_paper("1606.03476")
have("1606.03476")

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

curl https://syntology.ai/api/ran/1606.03476.json

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

[![Syntology run record](https://syntology.ai/api/ran/1606.03476.svg)](https://syntology.ai/paper/1606.03476)

Connect an agent — have() is free.