SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.07475 · ICLR · 2025

Partially Observed Trajectory Inference using Optimal Transport and a Dynamics Prior

Kristjan Greenewald, Anming Gu, Edward Chien

arXiv · PDF · Open in the Atlas

Code that ran

10 of 12 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (3 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 4 failures of our instrument; 2 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/2406.07475.json · badge: /api/ran/2406.07475.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
anminggu/partially-observed-traj-inference canonical (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
AnmingGu/partially-observed-traj-inference marked official in Papers with Code
Papers with Code archive
7 ran · 2 no recorded run
FunctionStatusWhere it lives
ConstrainLoss Ran
ran on a synthesized input · constructed an object; no result was computed
AnmingGu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("0247e14f424d4f21")
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, ConstrainLoss: 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)

# Construct small tensors for the parameters
# x: points in feature space, shape (batch_size, d)
x = torch.randn(2, 4)

# x_w: weights for x points, shape (batch_size,)
x_w = torch.ones(2)

# x_obs: observed points in feature space, shape (batch_obs, d)
x_obs = torch.randn(3, 4)

# x_obs_w: weights for observed points, shape (batch_obs,)
x_obs_w = torch.ones(3)

# sigma: standard deviation parameter (scalar)
sigma = 1.0

# scale_factor: scaling factor (scalar)
scale_factor = 0.5

# device: CPU device
device = torch.device('cpu')

ARGS = {
    'x': x,
    'x_w': x_w,
    'x_obs': x_obs,
    'x_obs_w': x_obs_w,
    'sigma': sigma,
    'scale_factor': scale_factor,
    'device': device,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ConstrainLoss — 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 ConstrainLoss 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
FitLoss Ran
ran on a synthesized input · constructed an object; no result was computed
AnmingGu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("7c829308e62f8ff4")
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, FitLoss: 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

torch.manual_seed(0)

# x: points in feature space, used in cdist
# Shape: (n_points, feature_dim)
x = torch.randn(2, 4)

# x_w: weights for x points, used in log and view(-1, 1)
# Shape: (n_points,)
x_w = torch.ones(2)

# y: target points in feature space, used in cdist
# Shape: (m_points, feature_dim)
y = torch.randn(3, 4)

# y_w: weights for y points, used in dot product
# Shape: (m_points,)
y_w = torch.ones(3)

# sigma: scalar parameter for scaling
sigma = 1.0

# scale_factor: scalar multiplier for sigma
scale_factor = 0.5

ARGS = {
    'x': x,
    'x_w': x_w,
    'y': y,
    'y_w': y_w,
    'sigma': sigma,
    'scale_factor': scale_factor,
}
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 FitLoss — 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 FitLoss 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
H Ran
ran; our contract draft was wrong, not the code
anminggu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("758720d8ba3137f9")
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 numpy as np

torch.manual_seed(0)

# The function H takes two parameters: p and q
# Both are used with torch.xlogy which expects tensors
# p is summed directly, and both p and q are used in xlogy operations
# xlogy(p, p) and xlogy(p, q) suggest p and q should be probability-like tensors

# Create small probability tensors
p = torch.tensor([[0.1, 0.2, 0.3, 0.4],
                  [0.25, 0.25, 0.25, 0.25]], dtype=torch.float32)

q = torch.tensor([[0.15, 0.15, 0.35, 0.35],
                  [0.1, 0.2, 0.3, 0.4]], dtype=torch.float32)

ARGS = {
    "p": p,
    "q": q
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, scipy, sklearn, 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
H_logdomain Ran
ran; our contract draft was wrong, not the code
anminggu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("e64e5799c6f5ba40")
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 numpy as np

torch.manual_seed(0)

# x should be a tensor that supports:
# - torch.xlogy(x, x) operation
# - .sum() method
# - element-wise multiplication with logK
x = torch.tensor([[0.1, 0.2, 0.3, 0.4],
                   [0.5, 0.6, 0.7, 0.8]], dtype=torch.float32)

# logK should be a tensor compatible with x for broadcasting and element-wise operations
logK = torch.tensor([[0.05, 0.1, 0.15, 0.2],
                      [0.25, 0.3, 0.35, 0.4]], dtype=torch.float32)

ARGS = {
    "x": x,
    "logK": logK
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, scipy, sklearn, 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
H_logdomain Ran
ran, honoured the contract we drafted
AnmingGu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("ab9d6403b7651b26")
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 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# x should be a tensor - looking at the function, it uses:
# - torch.xlogy(x, x) which requires x to be a numeric tensor
# - x.sum() operations
# - element-wise multiplication with logK
# Make it a small 2D tensor (batch=2, features=4)
x = torch.randn(2, 4)

# logK should be a tensor compatible with x for broadcasting and multiplication
# Same shape as x works
logK = torch.randn(2, 4)

ARGS = {
    "x": x,
    "logK": logK,
}
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 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
SinkhornLoss Ran
ran on a synthesized input · constructed an object; no result was computed
AnmingGu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("6fb57a4965a8add9")
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, SinkhornLoss: 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 math

torch.manual_seed(0)

# Create small tensors for the arguments
# x_w: weights for x samples, shape (2,)
x_w = torch.tensor([0.3, 0.7])

# x_spt: support points for x, shape (2, 4) - 2 samples, 4 dimensions
x_spt = torch.randn(2, 4)

# y_w: weights for y samples, shape (3,)
y_w = torch.tensor([0.2, 0.5, 0.3])

# y_spt: support points for y, shape (3, 4) - 3 samples, 4 dimensions
y_spt = torch.randn(3, 4)

# eps: regularization parameter
eps = 0.1

# n_iter: number of iterations for sinkhorn algorithm
n_iter = 10

# scale_factor: scaling factor for cost computation
scale_factor = 1.0

# lamda1: unbalanced OT parameter for x (None for balanced)
lamda1 = None

# lamda2: unbalanced OT parameter for y (None for balanced)
lamda2 = None

# warm_start: whether to warm start from previous solution
warm_start = False

# atol: absolute tolerance for convergence
atol = 0.0001

# rtol: relative tolerance for convergence
rtol = 0.0001

ARGS = {
    "x_w": x_w,
    "x_spt": x_spt,
    "y_w": y_w,
    "y_spt": y_spt,
    "eps": eps,
    "n_iter": n_iter,
    "scale_factor": scale_factor,
    "lamda1": lamda1,
    "lamda2": lamda2,
    "warm_start": warm_start,
    "atol": atol,
    "rtol": rtol,
}
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 SinkhornLoss — 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 SinkhornLoss 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
proxdiv_balanced Ran
ran, honoured the contract we drafted
AnmingGu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("bdf3301d00f12483")
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
import numpy as np

torch.manual_seed(0)

# The function simply divides p by s, so we need:
# p: a tensor or array-like
# s: a tensor or array-like (non-zero for valid division)
# u: some parameter (not used in function body)
# eps: some parameter (not used in function body)
# lamda: some parameter (not used in function body)

p = torch.randn(2, 4)
s = torch.ones(2, 4) + 0.1  # non-zero to avoid division issues
u = torch.randn(2, 4)
eps = 1e-6
lamda = 0.5

ARGS = {
    "p": p,
    "s": s,
    "u": u,
    "eps": eps,
    "lamda": lamda
}
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 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
proxdiv_unbalanced Ran
ran, honoured the contract we drafted
AnmingGu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("507016d797319fab")
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
import numpy as np

torch.manual_seed(0)

# Analyze the function:
# proxdiv_unbalanced(p, s, u, eps, lamda)
# - p, s, u appear to be tensors (division, exponentiation, exp() operations)
# - eps and lamda are scalars (used in arithmetic operations)

p = torch.randn(2, 4)
s = torch.randn(2, 4).abs() + 0.1  # ensure non-zero for division
u = torch.randn(2, 4)
eps = 0.1
lamda = 0.5

ARGS = {
    "p": p,
    "s": s,
    "u": u,
    "eps": eps,
    "lamda": lamda,
}
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 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
round_coupling Ran
ran; our contract draft was wrong, not the code
anminggu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("6cc2d823ffdf2166")
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 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
import numpy as np

torch.manual_seed(0)

# Create small test tensors
# x should be a 2D tensor (batch_size x num_features)
# p should be a 1D tensor (batch_size,)
# q should be a 1D tensor (num_features,)

batch_size = 2
num_features = 4

x = torch.rand(batch_size, num_features) + 0.1  # Add 0.1 to avoid near-zero values
p = torch.rand(batch_size) + 0.1
q = torch.rand(num_features) + 0.1

ARGS = {
    "x": x,
    "p": p,
    "q": q
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, scipy, sklearn, 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
round_coupling Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
AnmingGu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("8db2cbf5a8028bcc")
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

torch.manual_seed(0)

# Construct valid arguments for round_coupling
# The function expects:
# x: a 2D tensor (batch_size x features)
# p: a 1D or 2D tensor representing row marginals
# q: a 1D or 2D tensor representing column marginals

batch_size = 2
num_features = 4

# x should be a 2D tensor of shape (batch_size, num_features)
x = torch.rand(batch_size, num_features)

# p should have length batch_size (row marginals)
# Reshape to (-1, 1) in the function, so it's 1D or can be reshaped
p = torch.rand(batch_size)

# q should have length num_features (column marginals)
# Reshape to (1, -1) in the function, so it's 1D or can be reshaped
q = torch.rand(num_features)

ARGS = {
    "x": x,
    "p": p,
    "q": q,
}
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
PathsLoss Not yet run
unverified (harvested, no recorded run)
AnmingGu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("36fd2d25d764de6c")
How it ran

no invocation recorded.

TrajLoss Not yet run
unverified (harvested, no recorded run)
AnmingGu/partially-observed-traj-inference/models.py
code served (permissive licence) · get_code("51e66e3cde8768f7")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 10 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, 0 later) and 7 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
H 1 other paper
1 dated earlier
2205.07146 (2022-05)
Trajectory Inference via Mean-field Langevin in Path Space
H_logdomain
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2205.07146 (2022-05)
Trajectory Inference via Mean-field Langevin in Path Space
round_coupling
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2205.07146 (2022-05)
Trajectory Inference via Mean-field Langevin in Path Space

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

Trajectory inference seeks to recover the temporal dynamics of a population from snapshots of its (uncoupled) temporal marginals, i.e. where observed particles are not tracked over time. Prior works addressed this challenging problem under a stochastic differential equation (SDE) model with a gradient-driven drift in the observed space, introducing a minimum entropy estimator relative to the Wiener measure and a practical grid-free mean-field Langevin (MFL) algorithm using Schrödinger bridges. Motivated by the success of observable state space models in the traditional paired trajectory inference problem (e.g. target tracking), we extend the above framework to a class of latent SDEs in the form of observable state space models. In this setting, we use partial observations to infer trajectories in the latent space under a specified dynamics model (e.g. the constant velocity/acceleration models from target tracking). We introduce the PO-MFL algorithm to solve this latent trajectory inference problem and provide theoretical guarantees to the partially observed setting. Experiments validate the robustness of our method and the exponential convergence of the MFL dynamics, and demonstrate significant outperformance over the latent-free baseline in key scenarios.

For agents

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

get_harvested_code_for_paper("2406.07475")
get_code_for_paper("2406.07475")
have("2406.07475")

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

curl https://syntology.ai/api/ran/2406.07475.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.07475.svg)](https://syntology.ai/paper/2406.07475)

Connect an agent — have() is free.