SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2203.08553 · NeurIPS · 2022

PMIC: Improving Multi-Agent Reinforcement Learning with Progressive Mutual Information Collaboration

Tianpei Yang, Hongyao Tang, Jianye Hao, Xiaotian Hao, Yan Zheng, Pengyi Li, Zhen Wang, Tong Sang, Matthew Taylor, Wenyuan Tao, Fazl Barez

arXiv · PDF · Open in the Atlas

Code that ran

8 of 9 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 5 have no contract verdict; 3 failures of our instrument; 1 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/2203.08553.json · badge: /api/ran/2203.08553.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
yeshenpy/pmic community
Papers with Code archive
8 ran · 1 no recorded run
FunctionStatusWhere it lives
Actor Ran
ran on a synthesized input
yeshenpy/pmic/algorithms/mpe_new_maxminMADDPG.py
code served (permissive licence) · get_code("02d8b297accf97de")
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. 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

INIT_ARGS = {
    'state_dim': 4,
    'action_dim': 3,
    'max_action': 2.0,
    'layer_sizes': [4, 8, 8]
}

ARGS = {
    'x': torch.randn(2, 4)
}

if __name__ == '__main__':
    from collections import OrderedDict
    import torch.nn.functional as F
    
    class Actor(nn.Module):

        def __init__(self, state_dim, action_dim, max_action, layer_sizes=None):
            super(Actor, self).__init__()
            if layer_sizes is None:
                layer_sizes = [state_dim, 64, 64]
            self.nonlinearity = F.relu
            self.num_layers = len(layer_sizes)
            for i in range(1, self.num_layers):
                self.add_module('layer{0}'.format(i), nn.Linear(layer_sizes[i - 1], layer_sizes[i]))
            self.a_layer = nn.Linear(layer_sizes[-1], action_dim)
            self.max_action = max_action

        def forward(self, x, params=None):
            if params is None:
                params = OrderedDict(self.named_parameters())
            output = x
            output_list = []
            for i in range(1, self.num_layers):
                output = F.linear(output, weight=params['layer{0}.weight'.format(i)], bias=params['layer{0}.bias'.format(i)])
                output = self.nonlinearity(output, inplace=True)
                output_list.append(output)
            obs_embedding = output_list[0]
            policy_embedding = output_list[1]
            output = F.linear(output, weight=params['a_layer.weight'], bias=params['a_layer.bias'])
            output = self.max_action * torch.tanh(output)
            return (output, obs_embedding, policy_embedding)

    actor = Actor(**INIT_ARGS)
    result = actor(**ARGS)
    print("Output shape:", result[0].shape)
    print("Obs embedding shape:", result[1].shape)
    print("Policy embedding shape:", result[2].shape)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
CLUB Ran
ran on a synthesized input
yeshenpy/pmic/algorithms/mpe_new_maxminMADDPG.py
code served (permissive licence) · get_code("93f26c6f16a8c0ae")
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 Tensor, ndim 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

class CLUB(nn.Module):
    """
        This class provides the CLUB estimation to I(X,Y)
        Method:
            forward() :      provides the estimation with input samples
            loglikeli() :   provides the log-likelihood of the approximation q(Y|X) with input samples
        Arguments:
            x_dim, y_dim :         the dimensions of samples from X, Y respectively
            hidden_size :          the dimension of the hidden layer of the approximation network q(Y|X)
            x_samples, y_samples : samples from X and Y, having shape [sample_size, x_dim/y_dim]
    """

    def __init__(self, x_dim, y_dim, hidden_size):
        super(CLUB, self).__init__()
        self.p_mu = nn.Sequential(nn.Linear(x_dim, hidden_size // 2), nn.ReLU(inplace=True), nn.Linear(hidden_size // 2, y_dim))
        self.p_logvar = nn.Sequential(nn.Linear(x_dim, hidden_size // 2), nn.ReLU(inplace=True), nn.Linear(hidden_size // 2, y_dim), nn.Tanh())

    def get_mu_logvar(self, x_samples):
        mu = self.p_mu(x_samples)
        logvar = self.p_logvar(x_samples)
        return (mu, logvar)

    def forward(self, x_samples, y_samples):
        mu, logvar = self.get_mu_logvar(x_samples)
        positive = -(mu - y_samples) ** 2 / 2.0 / logvar.exp()
        prediction_1 = mu.unsqueeze(1)
        y_samples_1 = y_samples.unsqueeze(0)
        negative = -((y_samples_1 - prediction_1) ** 2).mean(dim=1) / 2.0 / logvar.exp()
        return positive.sum(dim=-1) - negative.sum(dim=-1)

    def loglikeli(self, x_samples, y_samples):
        mu, logvar = self.get_mu_logvar(x_samples)
        return (-(mu - y_samples) ** 2 / logvar.exp() - logvar).sum(dim=1).mean(dim=0)

    def learning_loss(self, x_samples, y_samples):
        return -self.loglikeli(x_samples, y_samples)


INIT_ARGS = {
    'x_dim': 4,
    'y_dim': 3,
    'hidden_size': 8,
}

ARGS = {
    'x_samples': torch.randn(2, 4),
    'y_samples': torch.randn(2, 3),
}

if __name__ == '__main__':
    club = CLUB(**INIT_ARGS)
    output = club(**ARGS)
    print("Forward output shape:", output.shape)
    print("Forward output:", output)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 1, dtype family float, shape (2,) — shape (2,) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Critic Ran
ran on a synthesized input
yeshenpy/pmic/algorithms/mpe_new_maxminMADDPG.py
code served (permissive licence) · get_code("0b5742fd1596a2a9")
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 Tensor, ndim 2, dtype family float, shape (2, 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 torch
import torch.nn as nn
import torch.nn.functional as F

class Critic(nn.Module):

    def __init__(self, state_dim, action_dim):
        super(Critic, self).__init__()
        self.l1 = nn.Linear(state_dim + action_dim, 64)
        self.l2 = nn.Linear(64, 64)
        self.l3 = nn.Linear(64, 1)

    def forward(self, x, u):
        xu = torch.cat([x, u], 1)
        x = F.relu(self.l1(xu), inplace=True)
        x = F.relu(self.l2(x), inplace=True)
        x = self.l3(x)
        return x


INIT_ARGS = {
    'state_dim': 4,
    'action_dim': 3,
}

ARGS = {
    'x': torch.randn(2, 4),
    'u': torch.randn(2, 3),
}

if __name__ == '__main__':
    critic = Critic(**INIT_ARGS)
    output = critic(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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, 1) — shape (2, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
NEW_MINE Ran
ran on a synthesized input
yeshenpy/pmic/algorithms/mpe_new_maxminMADDPG.py
code served (permissive licence) · get_code("fe832e6e436cb7d8")
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. 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
import torch.nn.functional as F
import math

def fenchel_dual_loss(l, m, measure=None):
    """Computes the f-divergence distance between positive and negative joint distributions.
    Note that vectors should be sent as 1x1.
    Divergences supported are Jensen-Shannon `JSD`, `GAN` (equivalent to JSD),
    Squared Hellinger `H2`, Chi-squeared `X2`, `KL`, and reverse KL `RKL`.
    Args:
        l: Local feature map.
        m: Multiple globals feature map.
        measure: f-divergence measure.
    Returns:
        torch.Tensor: Loss.
    """
    N, units = l.size()
    u = torch.mm(m, l.t())
    mask = torch.eye(N).to(l.device)
    n_mask = 1 - mask
    E_pos = get_positive_expectation(u, measure, average=False)
    E_neg = get_negative_expectation(u, measure, average=False)
    MI = (E_pos * mask).sum(1)
    E_pos_term = (E_pos * mask).sum(1)
    E_neg_term = (E_neg * n_mask).sum(1) / (N - 1)
    loss = E_neg_term - E_pos_term
    return (loss, MI)

def get_negative_expectation(q_samples, measure, average=True):
    log_2 = math.log(2.0)
    if measure == 'GAN':
        Eq = F.softplus(-q_samples) + q_samples
    elif measure == 'JSD':
        Eq = F.softplus(-q_samples) + q_samples - log_2
    elif measure == 'X2':
        Eq = -0.5 * (torch.sqrt(q_samples ** 2) + 1.0) ** 2
    elif measure == 'KL':
        q_samples = torch.clamp(q_samples, -1000000.0, 9.5)
        Eq = torch.exp(q_samples - 1.0)
    elif measure == 'RKL':
        Eq = q_samples - 1.0
    elif measure == 'H2':
        Eq = torch.exp(q_samples) - 1.0
    elif measure == 'W1':
        Eq = q_samples
    else:
        assert 1 == 2
    if average:
        return Eq.mean()
    else:
        return Eq

def get_positive_expectation(p_samples, measure, average=True):
    log_2 = math.log(2.0)
    if measure == 'GAN':
        Ep = -F.softplus(-p_samples)
    elif measure == 'JSD':
        Ep = log_2 - F.softplus(-p_samples)
    elif measure == 'X2':
        Ep = p_samples ** 2
    elif measure == 'KL':
        Ep = p_samples
    elif measure == 'RKL':
        Ep = -torch.exp(-p_samples)
    elif measure == 'DV':
        Ep = p_samples
    elif measure == 'H2':
        Ep = 1.0 - torch.exp(-p_samples)
    elif measure == 'W1':
        Ep = p_samples
    else:
        assert 1 == 2
    if average:
        return Ep.mean()
    else:
        return Ep

class NEW_MINE(nn.Module):

    def __init__(self, state_size, com_a_size, measure='JSD'):
        super(NEW_MINE, self).__init__()
        self.measure = measure
        self.com_a_size = com_a_size
        self.state_size = state_size
        self.nonlinearity = F.leaky_relu
        self.l1 = nn.Linear(self.state_size, 32)
        self.l2 = nn.Linear(self.com_a_size, 32)

    def forward(self, state, joint_action, params=None):
        em_1 = self.nonlinearity(self.l1(state), inplace=True)
        em_2 = self.nonlinearity(self.l2(joint_action), inplace=True)
        two_agent_embedding = [em_1, em_2]
        loss, MI = fenchel_dual_loss(two_agent_embedding[0], two_agent_embedding[1], measure=self.measure)
        return (loss, MI)

INIT_ARGS = {
    'state_size': 5,
    'com_a_size': 7,
    'measure': 'JSD'
}

ARGS = {
    'state': torch.randn(3, 5),
    'joint_action': torch.randn(3, 7),
    'params': None
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
OUNoise Ran
ran on a synthesized input
yeshenpy/pmic/algorithms/mpe_new_maxminMADDPG.py
code served (permissive licence) · get_code("5a6b141fd404c7c4")
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 ndarray, ndim 1, dtype family float, shape (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 numpy as np

class OUNoise:

    def __init__(self, action_dimension, scale=0.1, mu=0, theta=0.15, sigma=0.2):
        self.action_dimension = action_dimension
        self.scale = scale
        self.mu = mu
        self.theta = theta
        self.sigma = sigma
        self.state = np.ones(self.action_dimension) * self.mu
        self.reset()

    def reset(self):
        self.state = np.ones(self.action_dimension) * self.mu

    def noise(self):
        x = self.state
        dx = self.theta * (self.mu - x) + self.sigma * np.random.randn(len(x))
        self.state = x + dx
        return self.state * self.scale


INIT_ARGS = {
    'action_dimension': 4,
    'scale': 0.1,
    'mu': 0,
    'theta': 0.15,
    'sigma': 0.2
}

ARGS = {}

METHOD = 'noise'

ou_noise = OUNoise(**INIT_ARGS)
result = ou_noise.noise()
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 float, shape (4,) — shape (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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
fenchel_dual_loss Ran
ran; our contract draft was wrong, not the code
yeshenpy/pmic/algorithms/mpe_new_maxminMADDPG.py
code served (permissive licence) · get_code("dc127827c3f6691e")
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 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 torch
import math

torch.manual_seed(0)

# Create small tensors for the function
# l: Local feature map - shape (N, units)
# m: Multiple globals feature map - shape (units, N) or similar
# The function does: u = torch.mm(m, l.t())
# So if l is (N, units), l.t() is (units, N)
# m should be (N, units) to get u as (N, N)

N = 2
units = 4

l = torch.randn(N, units)
m = torch.randn(N, units)

measure = 'JSD'

ARGS = {
    'l': l,
    'm': m,
    'measure': measure
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 2, 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
get_negative_expectation Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
yeshenpy/pmic/algorithms/mpe_new_maxminMADDPG.py
code served (permissive licence) · get_code("640aa7bb79c50862")
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 Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import math

torch.manual_seed(0)

# Create a small tensor for q_samples
q_samples = torch.randn(2, 4)

# Test with different measures
measure = 'GAN'

# average can be True or False
average = True

ARGS = {
    'q_samples': q_samples,
    'measure': measure,
    'average': average
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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
get_positive_expectation Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
yeshenpy/pmic/algorithms/mpe_new_maxminMADDPG.py
code served (permissive licence) · get_code("0edef8661a209cf5")
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 Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import math

torch.manual_seed(0)

# Create a small tensor of positive and negative values for p_samples
p_samples = torch.randn(2, 4)

# Test with different measures
measures = ['GAN', 'JSD', 'X2', 'KL', 'RKL', 'DV', 'H2', 'W1']

# Create ARGS for a basic call with one measure
ARGS = {
    'p_samples': p_samples,
    'measure': 'GAN',
    'average': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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
MA_T_DDPG Not yet run
unverified (harvested, no recorded run)
yeshenpy/pmic/algorithms/mpe_new_maxminMADDPG.py
code served (permissive licence) · get_code("96cedade50014de1")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 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
OUNoise
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
1706.02275 (2017-06)
Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments

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

Learning to collaborate is critical in Multi-Agent Reinforcement Learning (MARL). Previous works promote collaboration by maximizing the correlation of agents' behaviors, which is typically characterized by Mutual Information (MI) in different forms. However, we reveal sub-optimal collaborative behaviors also emerge with strong correlations, and simply maximizing the MI can, surprisingly, hinder the learning towards better collaboration. To address this issue, we propose a novel MARL framework, called Progressive Mutual Information Collaboration (PMIC), for more effective MI-driven collaboration. PMIC uses a new collaboration criterion measured by the MI between global states and joint actions. Based on this criterion, the key idea of PMIC is maximizing the MI associated with superior collaborative behaviors and minimizing the MI associated with inferior ones. The two MI objectives play complementary roles by facilitating better collaborations while avoiding falling into sub-optimal ones. Experiments on a wide range of MARL benchmarks show the superior performance of PMIC compared with other algorithms. Our code is open-source and available at https://github.com/yeshenpy/PMIC.

For agents

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

get_harvested_code_for_paper("2203.08553")
get_code_for_paper("2203.08553")
have("2203.08553")

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

curl https://syntology.ai/api/ran/2203.08553.json

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

[![Syntology run record](https://syntology.ai/api/ran/2203.08553.svg)](https://syntology.ai/paper/2203.08553)

Connect an agent — have() is free.