SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1412.7449 · 2014

Grammar as a Foreign Language

arXiv · PDF · Open in the Atlas

Code that ran

9 of 17 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 9 have no contract verdict; 0 failures of our instrument; 8 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/1412.7449.json · badge: /api/ran/1412.7449.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
sambit9238/deep_text_corrector community (classifier label)
classifier basis not recorded on the link
4 ran · 2 no recorded run
atpaino/deep-text-corrector community (classifier label)
classifier basis not recorded on the link
3 ran · 3 no recorded run
kmadathil/sanskrit_parser community (classifier label)
classifier basis not recorded on the link
2 ran · 3 no recorded run
FunctionStatusWhere it lives
apply_input_bias_and_extract_argmax_fn_factory Ran
ran on a synthesized input
atpaino/deep-text-corrector/text_corrector_models.py
code served (permissive licence) · get_code("d1058b37b338fbbf")
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 function. 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 tensorflow as tf
import numpy as np

# Set seed for reproducibility
tf.random.set_seed(0)
np.random.seed(0)

# Create small tensors for testing
batch_size = 2
vocab_size = 5
embedding_dim = 4
hidden_dim = 3

# Create input_bias tensor - should have shape compatible with probability distribution
# This will be multiplied element-wise with softmax output, so shape should be (batch_size, vocab_size)
input_bias = tf.constant(np.random.rand(batch_size, vocab_size).astype(np.float32))

ARGS = {
    "input_bias": input_bias
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
basic_rnn_seq2seq Ran
ran on a synthesized input
sambit9238/deep_text_corrector/seq2seq.py
code served (permissive licence) · get_code("7044f78aac4e77b0")
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 tensorflow as tf
from tensorflow.python.framework import dtypes
from tensorflow.python.ops import rnn_cell
import numpy as np

# Set random seeds for reproducibility
tf.compat.v1.set_random_seed(0)
np.random.seed(0)

# Define a minimal RNN cell that satisfies the interface
class MinimalRNNCell(rnn_cell.RNNCell):
    def __init__(self, input_size, output_size, state_size):
        self._input_size = input_size
        self._output_size = output_size
        self._state_size = state_size
    
    @property
    def state_size(self):
        return self._state_size
    
    @property
    def output_size(self):
        return self._output_size
    
    def __call__(self, inputs, state, scope=None):
        # Simple dummy RNN: output = tanh(input + state)
        with tf.compat.v1.variable_scope(scope or "minimal_rnn_cell"):
            # Create a simple weight matrix for transformation
            w = tf.compat.v1.get_variable(
                "w",
                [self._input_size + self._state_size, self._output_size],
                dtype=dtypes.float32
            )
            # Concatenate input and state
            combined = tf.concat([inputs, state], 1)
            output = tf.nn.tanh(tf.matmul(combined, w))
            
            # Update state (simple: use output as next state)
            new_state = output[:, :self._state_size]
            
            return output, new_state

# Parameters
batch_size = 2
input_size = 4
output_size = 5
state_size = 3
seq_length = 3

# Create encoder inputs: list of 2D Tensors [batch_size x input_size]
encoder_inputs = [
    tf.constant(np.random.randn(batch_size, input_size), dtype=dtypes.float32)
    for _ in range(seq_length)
]

# Create decoder inputs: list of 2D Tensors [batch_size x input_size]
decoder_inputs = [
    tf.constant(np.random.randn(batch_size, input_size), dtype=dtypes.float32)
    for _ in range(seq_length)
]

# Create RNN cell
cell = MinimalRNNCell(input_size=input_size, output_size=output_size, state_size=state_size)

# Create arguments dictionary
ARGS = {
    'encoder_inputs': encoder_inputs,
    'decoder_inputs': decoder_inputs,
    'cell': cell,
    'dtype': dtypes.float32,
    'scope': None
}
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
cleanText Ran
ran on a synthesized input
sambit9238/deep_text_corrector/preprocessors/preprocess_movie_dialogs.py
code served (permissive licence) · get_code("cb0ae36059a1edc1")
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 re

def replace_contraction(text):
    contraction_patterns = [("won\\'t", 'will not'), ("can\\'t", 'can not'), ("i\\'m", 'i am'), ("ain\\'t", 'is not'), ("(\\w+)\\'ll", '\\g<1> will'), ("(\\w+)n\\'t", '\\g<1> not'), ("(\\w+)\\'ve", '\\g<1> have'), ("(\\w+)\\'s", '\\g<1> is'), ("(\\w+)\\'re", '\\g<1> are'), ("(\\w+)\\'d", '\\g<1> would'), ('&', 'and'), ('dammit', 'damn it'), ('dont', 'do not'), ('wont', 'will not')]
    patterns = [(re.compile(regex), repl) for regex, repl in contraction_patterns]
    for pattern, repl in patterns:
        text, count = re.subn(pattern, repl, text)
    return text

def replace_links(text, filler='links'):
    text = re.sub('((http|https)\\:\\/\\/)?[a-zA-Z0-9\\.\\/\\?\\:@\\-_=#]+\\.([a-zA-Z]){2,6}([a-zA-Z0-9\\.\\&\\/\\?\\:@\\-_=#])*', filler, text).strip()
    return text

def cleanText(text):
    text = text.strip().replace('\n', ' ').replace('\r', ' ')
    text = replace_contraction(text)
    text = replace_links(text, 'link')
    text = text.lower()
    return text

ARGS = {
    'text': "I won't go to https://example.com today, it ain't right & I'm tired.\nThis is dammit."
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
getSandhiArgs Ran
ran on a synthesized input
kmadathil/sanskrit_parser/sanskrit_parser/cmd_line.py
code served (permissive licence) · get_code("05d4c96440a087cb")
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 Namespace. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
from argparse import ArgumentParser

torch.manual_seed(0)

# The function getSandhiArgs takes argv as an optional parameter
# It parses command-line arguments for a Sandhi Analyzer
# We need to construct valid argv that would be passed to the parser

ARGS = {
    'argv': [
        'astyuttarasyAMdishidevatAtmA',
        '--input-encoding', 'utf-8',
        '--max-paths', '10',
        '--lexical-lookup', 'combined',
        '--pre-segmented',
        '--strict-io',
        '--dot-file', 'output.dot'
    ]
}
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 Namespace — a Namespace output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Namespace 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
get_corrective_tokens Ran
ran on a synthesized input
atpaino/deep-text-corrector/correct_text.py
code served (permissive licence) · get_code("36f700e311ac6061")
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 set. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

class MockDataReader:
    def read_samples_by_string(self, train_path):
        # Returns an iterator of (source_tokens, target_tokens) tuples
        samples = [
            (["hello", "world"], ["hello", "world", "foo"]),
            (["cat", "dog"], ["cat", "dog", "bar", "baz"]),
        ]
        return samples

data_reader = MockDataReader()
train_path = "/dummy/path/train.txt"

ARGS = {
    "data_reader": data_reader,
    "train_path": train_path,
}
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 set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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
guna Ran
ran on a synthesized input
kmadathil/sanskrit_parser/sanskrit_parser/generator/paribhasha.py
code served (permissive licence) · get_code("4d28261bf66161c9")
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.

# First, let's understand what the function does:
# - It takes a string parameter `s`
# - It checks if `s` is in 'fF', 'xX', or calls adesha() for other cases
# - The function adesha is not defined in the provided code, so we need to provide it

# Define the missing adesha function that guna depends on
def adesha(s, param2, param3):
    """Stand-in function for adesha that guna calls"""
    return 'result'

# Now define valid arguments for guna
ARGS = {
    's': 'f'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
replace_contraction Ran
ran on a synthesized input
sambit9238/deep_text_corrector/preprocessors/preprocess_movie_dialogs.py
code served (permissive licence) · get_code("32fddafac018f902")
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 re

def replace_contraction(text):
    contraction_patterns = [("won\\'t", 'will not'), ("can\\'t", 'can not'), ("i\\'m", 'i am'), ("ain\\'t", 'is not'), ("(\\w+)\\'ll", '\\g<1> will'), ("(\\w+)n\\'t", '\\g<1> not'), ("(\\w+)\\'ve", '\\g<1> have'), ("(\\w+)\\'s", '\\g<1> is'), ("(\\w+)\\'re", '\\g<1> are'), ("(\\w+)\\'d", '\\g<1> would'), ('&', 'and'), ('dammit', 'damn it'), ('dont', 'do not'), ('wont', 'will not')]
    patterns = [(re.compile(regex), repl) for regex, repl in contraction_patterns]
    for pattern, repl in patterns:
        text, count = re.subn(pattern, repl, text)
    return text

ARGS = {
    "text": "I won't go there, can't you see? It ain't right & I'm done. He'd better not dammit."
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
replace_links Ran
ran on a synthesized input
sambit9238/deep_text_corrector/preprocessors/preprocess_movie_dialogs.py
code served (permissive licence) · get_code("153755cbc8e726de")
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 re

def replace_links(text, filler='links'):
    text = re.sub('((http|https)\\:\\/\\/)?[a-zA-Z0-9\\.\\/\\?\\:@\\-_=#]+\\.([a-zA-Z]){2,6}([a-zA-Z0-9\\.\\&\\/\\?\\:@\\-_=#])*', filler, text).strip()
    return text

ARGS = {
    'text': 'Check out https://example.com and google.com for more info',
    'filler': 'links'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
rnn_decoder Ran
ran on a synthesized input
atpaino/deep-text-corrector/seq2seq.py
code served (permissive licence) · get_code("cf938d1d0e9a8c10")
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 torch
import tensorflow as tf
import numpy as np

torch.manual_seed(0)

class SimpleRNNCell:
    """Stand-in RNN cell that mimics the interface expected by rnn_decoder."""
    def __init__(self, input_size, output_size, state_size):
        self.input_size = input_size
        self.output_size = output_size
        self.state_size = state_size
        
    def __call__(self, inp, state):
        """
        Args:
            inp: 2D Tensor [batch_size x input_size]
            state: 2D Tensor [batch_size x state_size]
        
        Returns:
            output: 2D Tensor [batch_size x output_size]
            new_state: 2D Tensor [batch_size x state_size]
        """
        batch_size = inp.shape[0]
        # Simple deterministic computation for stand-in
        output = tf.ones([batch_size, self.output_size], dtype=tf.float32)
        new_state = tf.ones([batch_size, self.state_size], dtype=tf.float32)
        return output, new_state

batch_size = 2
input_size = 3
output_size = 4
state_size = 5
sequence_length = 3

decoder_inputs = [
    tf.constant(np.random.randn(batch_size, input_size), dtype=tf.float32)
    for _ in range(sequence_length)
]

initial_state = tf.constant(np.random.randn(batch_size, state_size), dtype=tf.float32)

cell = SimpleRNNCell(input_size=input_size, output_size=output_size, state_size=state_size)

def loop_function(prev, i):
    """
    Args:
        prev: 2D Tensor [batch_size x output_size]
        i: integer step number
    
    Returns:
        next: 2D Tensor [batch_size x input_size]
    """
    batch_size = prev.shape[0]
    return tf.constant(np.random.randn(batch_size, input_size), dtype=tf.float32)

ARGS = {
    "decoder_inputs": decoder_inputs,
    "initial_state": initial_state,
    "cell": cell,
    "loop_function": loop_function,
    "scope": None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow, 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: none by the harness; any seed is in the setup block
basic_rnn_seq2seq Not yet run
unverified (harvested, no recorded run)
atpaino/deep-text-corrector/seq2seq.py
code served (permissive licence) · get_code("8429d7c06389b807")
How it ran

no invocation recorded.

dirgha Not yet run
unverified (harvested, no recorded run)
kmadathil/sanskrit_parser/sanskrit_parser/generator/paribhasha.py
code served (permissive licence) · get_code("a865a8db7110f228")
How it ran

no invocation recorded.

getVakyaArgs Not yet run
unverified (harvested, no recorded run)
kmadathil/sanskrit_parser/sanskrit_parser/cmd_line.py
code served (permissive licence) · get_code("b95718cadb9b8eaa")
How it ran

no invocation recorded.

hrasva Not yet run
unverified (harvested, no recorded run)
kmadathil/sanskrit_parser/sanskrit_parser/generator/paribhasha.py
code served (permissive licence) · get_code("1dffd7db3e987bbe")
How it ran

no invocation recorded.

project_and_apply_input_bias Not yet run
unverified (harvested, no recorded run)
atpaino/deep-text-corrector/text_corrector_models.py
code served (permissive licence) · get_code("5aaefa2f51c41369")
How it ran

no invocation recorded.

rnn_decoder Not yet run
unverified (harvested, no recorded run)
sambit9238/deep_text_corrector/seq2seq.py
code served (permissive licence) · get_code("f44847e8b1c4bd91")
How it ran

no invocation recorded.

tied_rnn_seq2seq Not yet run
unverified (harvested, no recorded run)
atpaino/deep-text-corrector/seq2seq.py
code served (permissive licence) · get_code("8c64b893aa64969f")
How it ran

no invocation recorded.

tied_rnn_seq2seq Not yet run
unverified (harvested, no recorded run)
sambit9238/deep_text_corrector/seq2seq.py
code served (permissive licence) · get_code("772f0a140d5d7ed8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 16 distinct functions with an identity, 3 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 3 later) and 13 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
rnn_decoder 2 other papers
2 later
1609.04938 (2016-09)
Image-to-Markup Generation with Coarse-to-Fine Attention
2204.05805 (2022-04)
Learning to Express in Knowledge-Grounded Conversation
basic_rnn_seq2seq 1 other paper
1 later
2204.05805 (2022-04)
Learning to Express in Knowledge-Grounded Conversation
tied_rnn_seq2seq 1 other paper
1 later
2204.05805 (2022-04)
Learning to Express in Knowledge-Grounded Conversation

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

Syntactic constituency parsing is a fundamental problem in natural language processing and has been the subject of intensive research and engineering for decades. As a result, the most accurate parsers are domain specific, complex, and inefficient. In this paper we show that the domain agnostic attention-enhanced sequence-to-sequence model achieves state-of-the-art results on the most widely used syntactic constituency parsing dataset, when trained on a large synthetic corpus that was annotated using existing parsers. It also matches the performance of standard parsers when trained only on a small human-annotated dataset, which shows that this model is highly data-efficient, in contrast to sequence-to-sequence models without the attention mechanism. Our parser is also fast, processing over a hundred sentences per second with an unoptimized CPU implementation.

For agents

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

get_harvested_code_for_paper("1412.7449")
get_code_for_paper("1412.7449")
have("1412.7449")

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

curl https://syntology.ai/api/ran/1412.7449.json

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

[![Syntology run record](https://syntology.ai/api/ran/1412.7449.svg)](https://syntology.ai/paper/1412.7449)

Connect an agent — have() is free.