SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2005.06409 · ACL · 2020

Dense-Caption Matching and Frame-Selection Gating for Temporal Localization in VideoQA

Mohit Bansal, Zineng Tang, Hyounghun Kim

arXiv · PDF · Open in the Atlas

Code that ran

9 of 12 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, 0 violated it and 6 have no contract verdict; 1 failure of our instrument; 3 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/2005.06409.json · badge: /api/ran/2005.06409.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
hyounghk/VideoQADenseCapFrameGate-ACL2020 marked official in Papers with Code
Papers with Code archive
8 ran · 3 no recorded run
copy not recorded community
not in the paper's declared repositories
1 ran · 0 no recorded run
FunctionStatusWhere it lives
DepthwiseSeparableConv Ran
ran on a synthesized input · constructed an object; no result was computed
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("eabc5a84173598c5")
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, DepthwiseSeparableConv: 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)

# Define ARGS dict with all parameters for DepthwiseSeparableConv.__init__
ARGS = {
    'in_ch': 4,
    'out_ch': 8,
    'k': 3,
    'dim': 1,
    'relu': True
}
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 DepthwiseSeparableConv — 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 DepthwiseSeparableConv 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
LinearWrapper Ran
ran on a synthesized input · constructed an object; no result was computed
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("92c3decd59ca89ec")
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, LinearWrapper: 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 = {
    'in_hsz': 4,
    'out_hsz': 3,
    'layer_norm': True,
    'dropout': 0.1,
    'relu': True
}
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 LinearWrapper — 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 LinearWrapper 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
MultiHeadedAttention Ran
ran on a synthesized input · constructed an object; no result was computed
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("27eddc30c65e3b95")
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, MultiHeadedAttention: 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
import math
import copy

torch.manual_seed(0)

def clones(module, n):
    return nn.ModuleList([copy.deepcopy(module) for _ in range(n)])

class MultiHeadedAttention(nn.Module):

    def __init__(self, nh, d_model, dropout=0.1):
        super(MultiHeadedAttention, self).__init__()
        assert d_model % nh == 0
        self.d_k = d_model // nh
        self.nh = nh
        self.linears = clones(nn.Linear(d_model, d_model), 4)
        self.dropout = nn.Dropout(p=dropout)

    def forward(self, x, mask=None):
        bsz = x.size(0)
        if mask is not None:
            mask = mask.view(bsz, 1, -1, 1)
        query, key, value = [l(x).view(bsz, -1, self.nh, self.d_k).transpose(1, 2) for l, x in zip(self.linears, (x, x, x))]
        x, attn = self.attention(query, key, value, mask=mask, dropout=self.dropout)
        x = x.transpose(1, 2).contiguous().view(bsz, -1, self.nh * self.d_k)
        return self.linears[-1](x)

    def attention(self, query, key, value, mask=None, dropout=None):
        scores = torch.matmul(query, key.transpose(-2, -1)) / math.sqrt(self.d_k)
        if mask is not None:
            scores = scores.masked_fill(mask == 0, -1000000000.0)
        p_attn = torch.softmax(scores, dim=-1)
        if dropout is not None:
            p_attn = dropout(p_attn)
        return (torch.matmul(p_attn, value), p_attn)

# Create valid arguments for MultiHeadedAttention.__init__
nh = 2  # number of heads
d_model = 4  # model dimension (must be divisible by nh)
dropout = 0.1  # dropout rate

ARGS = {
    'nh': nh,
    'd_model': d_model,
    'dropout': dropout
}
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 MultiHeadedAttention — 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 MultiHeadedAttention 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
PositionEncoding Ran
ran on a synthesized input · constructed an object; no result was computed
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("b9683b87afb37df5")
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, PositionEncoding: 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
import math

torch.manual_seed(0)

# PositionEncoding only takes n_filters and max_len as parameters
# n_filters: int, dimension of the embeddings
# max_len: int, maximum sequence length

ARGS = {
    'n_filters': 8,
    'max_len': 16
}
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 PositionEncoding — 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 PositionEncoding 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
StructuredAttention_bi Ran
ran on a synthesized input · constructed an object; no result was computed
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("d606630a26a361c2")
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, StructuredAttention_bi: 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
import torch.nn.functional as F

torch.manual_seed(0)

class StructuredAttention_bi(nn.Module):

    def __init__(self, dropout=0.1, scale=100):
        super(StructuredAttention_bi, self).__init__()
        self.dropout = dropout
        self.scale = scale

    def forward(self, C, Q, c_mask, q_mask):
        bsz, _, num_img, num_region, hsz = Q.shape
        S, S_mask = self.similarity(C, Q, c_mask, q_mask)
        S_c = F.softmax(S * self.scale, dim=-1)
        S_q = F.softmax(S * self.scale, dim=-2)
        S_c = S_c * S_mask
        S_q = S_q * S_mask
        A_c = torch.matmul(S_c, Q)
        A_q = torch.matmul(S_q.transpose(-2, -1), C)
        return (A_c, A_q, S_mask, S_mask.transpose(-2, -1))

    def similarity(self, C, Q, c_mask, q_mask):
        C = F.dropout(F.normalize(C, p=2, dim=-1), p=self.dropout, training=self.training)
        Q = F.dropout(F.normalize(Q, p=2, dim=-1), p=self.dropout, training=self.training)
        S_mask = torch.matmul(c_mask.unsqueeze(-1), q_mask.unsqueeze(-2))
        S = torch.matmul(C, Q.transpose(-2, -1))
        masked_S = S - 10000000000.0 * (1 - S_mask)
        return (masked_S, S_mask)


# Create model instance
model = StructuredAttention_bi(dropout=0.1, scale=100)
model.eval()

# Create input tensors
# Q shape: (bsz, _, num_img, num_region, hsz) = (2, 3, 4, 5, 8)
bsz, dim1, num_img, num_region, hsz = 2, 3, 4, 5, 8

# C shape should be compatible with Q for matmul in similarity
# C will be (bsz, c_seq_len, hsz) = (2, 7, 8)
C = torch.randn(2, 7, 8)
Q = torch.randn(2, 3, 4, 5, 8)

# c_mask shape: (bsz, c_seq_len) = (2, 7)
# q_mask shape: (bsz, num_img, num_region) = (2, 4, 5)
c_mask = torch.ones(2, 7)
q_mask = torch.ones(2, 4, 5)

ARGS = {
    'dropout': 0.1,
    'scale': 100
}
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 StructuredAttention_bi — 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 StructuredAttention_bi 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
StructuredAttention_frame Ran
ran on a synthesized input · constructed an object; no result was computed
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("ca5ccdc1e66312e7")
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, StructuredAttention_frame: 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)

# Analyze the function:
# - forward() takes C, Q, c_mask, q_mask
# - Q.shape is unpacked as (bsz, _, num_img, hsz)
# - C is used in similarity() with matmul against Q.transpose(-2, -1)
# - c_mask and q_mask are used in similarity()
# - C and Q are normalized and dropout applied
# - S_mask comes from c_mask.unsqueeze(-1)
# - matmul(S_, Q) is performed where S_ is (bsz, c_len, num_img) and Q is (bsz, q_len, num_img, hsz)

# Shape requirements:
# Q: (bsz, q_len, num_img, hsz)
# C: must be compatible for matmul with Q.transpose(-2, -1)
#    Q.transpose(-2, -1) = (bsz, q_len, hsz, num_img)
#    For matmul to work: C should be (bsz, c_len, hsz)
# c_mask: (bsz, c_len) or broadcastable for subtraction
# q_mask: used in similarity but not directly in computation

bsz = 2
c_len = 5
q_len = 3
num_img = 4
hsz = 6

# Create tensors with distinct dimensions
C = torch.randn(bsz, c_len, hsz)  # (2, 5, 6)
Q = torch.randn(bsz, q_len, num_img, hsz)  # (2, 3, 4, 6)

# Masks: typically binary masks indicating valid positions
c_mask = torch.ones(bsz, c_len, 1)  # (2, 5, 1) - will be used in unsqueeze
q_mask = torch.ones(bsz, q_len, 1)  # (2, 3, 1)

ARGS = {
    "dropout": 0.1,
    "scale": 100,
}

# Instantiate and test
model = nn.Module.__new__(nn.Module)
# Manually set up the instance to verify arguments work
for key, value in ARGS.items():
    setattr(model, key, value)

# Forward pass test to verify tensor shapes work
model_instance = nn.Sequential()  # dummy, we'll call forward directly

# Just verify the tensors are compatible
assert C.shape == (bsz, c_len, hsz)
assert Q.shape == (bsz, q_len, num_img, hsz)
assert c_mask.shape == (bsz, c_len, 1)

ARGS = {
    "dropout": 0.1,
    "scale": 100,
}
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 StructuredAttention_frame — 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 StructuredAttention_frame 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
clones Ran
ran; our contract draft was wrong, not the code
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("792073de4c443fa8")
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 ModuleList. 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 copy

torch.manual_seed(0)

# The clones function takes a module and clones it n times
# It uses: copy.deepcopy() on the module, and nn.ModuleList()
# So we need a simple nn.Module and an integer n

class SimpleModule(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

module = SimpleModule()
n = 5

ARGS = {
    "module": module,
    "n": n
}
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 ModuleList — a ModuleList output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ModuleList 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
mask_logits Ran
ran, honoured the contract we drafted
this paper's copy was not recorded; identical code first harvested from songpipi/amdnet
pointer only; the stored copy get_code serves, from songpipi/amdnet, is pointer only (licence: NONE) · get_code("1c17153fb5488283")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. 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.

mask_logits_sum Ran
ran, honoured the contract we drafted
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("ef908c1d80306f2d")
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 torch.nn as nn

torch.manual_seed(0)

# Create small tensors for the function parameters
target = torch.randn(2, 4)  # batch size 2, dim 4
mask = torch.randint(0, 2, (2, 4)).float()  # binary mask same shape as target

ARGS = {
    "target": target,
    "mask": mask,
}
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
EncoderBlock Not yet run
unverified (harvested, no recorded run)
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("141e2be2575e53f1")
How it ran

no invocation recorded.

StackedEncoder Not yet run
unverified (harvested, no recorded run)
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("7bbe54b9dd2d671a")
How it ran

no invocation recorded.

TVQANet Not yet run
unverified (harvested, no recorded run)
hyounghk/VideoQADenseCapFrameGate-ACL2020/qanet/tvqanet.py
code served (permissive licence) · get_code("5a91cb89645d7259")
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 12 distinct functions with an identity, 4 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 1 the same month, 4 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
mask_logits
brick of this paper's assembly: helper
13 other papers
2 dated earlier, 11 later · 1 only as an assembly brick
1904.11574 (2019-04)
TVQA+: Spatio-Temporal Grounding for Video Question Answering
2001.09099 (2020-01)
TVR: A Large-Scale Dataset for Video-Subtitle Moment Retrieval
2105.06247 (2021-05)
Video Corpus Moment Retrieval with Contrastive Learning
and 10 more not named here
PositionEncoding
brick of this paper's assembly: helper
3 other papers
1 the same month, 2 later · 1 only as an assembly brick
2005.05402 (2020-05)
MART: Memory-Augmented Recurrent Transformer for Coherent Video Paragraph Captioning
2210.08465 (2022-10)
Character-Centric Story Visualization via Visual Planning and Token Alignment
2603.05969 (2026-03)
Imagine How To Change: Explicit Procedure Modeling for Change Captioning
clones
brick of this paper's assembly: helper
2 other papers
1 dated earlier, 1 later
1904.11574 (2019-04)
TVQA+: Spatio-Temporal Grounding for Video Question Answering
2203.07216 (2022-03)
A Novel Perspective to Look At Attention: Bi-level Attention-based Explainable Topic Modeling for News Classification
DepthwiseSeparableConv
brick of this paper's assembly: helper
1 other paper
1 later
2108.00061 (2021-08)
MTVR: Multilingual Moment Retrieval in Videos

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

Videos convey rich information. Dynamic spatio-temporal relationships between people/objects, and diverse multimodal events are present in a video clip. Hence, it is important to develop automated models that can accurately extract such information from videos. Answering questions on videos is one of the tasks which can evaluate such AI abilities. In this paper, we propose a video question answering model which effectively integrates multi-modal input sources and finds the temporally relevant information to answer questions. Specifically, we first employ dense image captions to help identify objects and their detailed salient regions and actions, and hence give the model useful extra information (in explicit textual format to allow easier matching) for answering questions. Moreover, our model is also comprised of duallevel attention (word/object and frame level), multi-head self/cross-integration for different sources (video and dense captions), and gates which pass more relevant information to the classifier. Finally, we also cast the frame selection problem as a multi-label classification task and introduce two loss functions, In-and-Out Frame Score Margin (IOFSM) and Balanced Binary Cross-Entropy (BBCE), to better supervise the model with human importance annotations. We evaluate our model on the challenging TVQA dataset, where each of our model components provides significant gains, and our overall model outperforms the stateof-the-art by a large margin (74.09% versus 70.52%). We also present several word, object, and frame level visualization studies. 1 Frame-Level Att.

For agents

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

get_harvested_code_for_paper("2005.06409")
get_code_for_paper("2005.06409")
have("2005.06409")

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

curl https://syntology.ai/api/ran/2005.06409.json

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

[![Syntology run record](https://syntology.ai/api/ran/2005.06409.svg)](https://syntology.ai/paper/2005.06409)

Connect an agent — have() is free.