SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2409.16223 · NeurIPS · 2024

Fine-Tuning is Fine, if Calibrated

Wei-Lun Chao, Hong-You Chen, Cheng-Hao Tu, Yu Su, Zheda Mai, Vardaan Pahuja, Tanya Berger-Wolf, Song Gao, Charles Stewart, Arpita Chowdhury, Ping Zhang

arXiv · PDF · Open in the Atlas

Code that ran

9 of 15 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; 1 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 6 failures of our instrument; 6 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/2409.16223.json · badge: /api/ran/2409.16223.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
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated canonical (classifier label)
classifier basis: census
9 ran · 6 no recorded run
FunctionStatusWhere it lives
all_in Ran
ran, violated the contract we drafted
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/util/common.py
code served (permissive licence) · get_code("7d3790fd60c5077a")
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 bool, 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 all_in checks if all elements of `a` are contained in `b`
# Both `a` and `b` should be iterable collections

a = [1, 2, 3]
b = [1, 2, 3, 4, 5]

ARGS = {
    'a': a,
    'b': b
}
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 bool, 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
dtn Ran
ran; our contract draft was wrong, not the code
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/model/_digits.py
code served (permissive licence) · get_code("654b3d00bbf1c923")
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 DTN. 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 = {
    "pretrained": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DTN — a DTN output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DTN 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
flatten_list Ran
ran, honoured the contract we drafted
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/util/common.py
code served (permissive licence) · get_code("1d66cc48b63fca52")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family int; 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 numpy as np

torch.manual_seed(0)

# flatten_list takes a parameter 'l' which is a list of lists
# The function iterates through each sublist and flattens them into a single list
# So we need to provide a list of iterables

ARGS = {
    "l": [
        [1, 2, 3],
        [4, 5],
        [6, 7, 8, 9]
    ]
}
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 list, ndim 1, dtype family int — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_class_mean Ran
ran; our contract draft was wrong, not the code
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/util/evaluate.py
code served (permissive licence) · get_code("1f88ab8c26c03b73")
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 (3, 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)

class DomainInfo:
    def __init__(self, all_classes, num_classes):
        self.all_classes = all_classes
        self.num_classes = num_classes

# Small batch with 2 samples, 4 feature dimensions
batch_size = 2
feature_dim = 4
num_classes = 3

features = torch.randn(batch_size, feature_dim, dtype=torch.float32)
labels = torch.tensor([0, 1], dtype=torch.long)
all_classes = [0, 1, 2]

domain_info = DomainInfo(all_classes=all_classes, num_classes=num_classes)

ARGS = {
    "domain_info": domain_info,
    "features": features,
    "labels": labels,
}
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 (3, 4) — shape (3, 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
get_parameters Ran
ran; our contract draft was wrong, not the code
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/util/model.py
code served (permissive licence) · get_code("d829c6631f3d92c6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import logging

logging.basicConfig(level=logging.DEBUG)

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.backbone = nn.Linear(4, 3)
        self.bottleneck = nn.Linear(3, 2)
        self.head = nn.Linear(2, 1)
    
    def forward(self, x):
        x = self.backbone(x)
        x = self.bottleneck(x)
        x = self.head(x)
        return x

model = SimpleModel()

ARGS = {
    'model': model,
    'base_lr': 1.0,
    'fixedcls': False,
    'fixed_backbone': False,
    'finetune': 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 list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
is_json Ran
ran, violated the contract we drafted
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/util/common.py
code served (permissive licence) · get_code("70367b6dda994032")
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 bool, 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 json

# The function is_json(s) takes a string parameter and tries to parse it as JSON
# It returns True if the string is valid JSON, False otherwise

# Create test cases for the parameter s
valid_json_string = '{"key": "value"}'
invalid_json_string = 'not valid json'
empty_json_object = '{}'
json_array = '[1, 2, 3]'
json_number = '42'
json_null = 'null'

ARGS = {
    's': valid_json_string
}
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 bool, 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
lenet Ran
ran; our contract draft was wrong, not the code
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/model/_digits.py
code served (permissive licence) · get_code("87c67b9512a15d16")
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 LeNet. 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)

# The lenet function accepts **kwargs which are passed to LeNet
# LeNet.__init__ doesn't take any parameters beyond self
# The forward method expects an input tensor of shape (batch, 1, 28, 28)
# based on the docstring mentioning "input image size must be 28 x 28"
# and the architecture expecting 1 input channel

# Create a small test input tensor (batch_size=2, channels=1, height=28, width=28)
test_input = torch.randn(2, 1, 28, 28)

ARGS = {
    'pretrained': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LeNet — a LeNet output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LeNet 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
mask_score Ran
ran; our contract draft was wrong, not the code
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/util/evaluate.py
code served (permissive licence) · get_code("095e3291b31a3a1e")
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 copy

torch.manual_seed(0)

# score: a 2D tensor that will be indexed and masked
score = torch.randn(2, 4)

# row_mask: boolean tensor with shape matching first dimension of score
row_mask = torch.tensor([True, False])

# column_mask: boolean tensor with shape matching second dimension of score
column_mask = torch.tensor([True, True, False, True])

ARGS = {
    "score": score,
    "row_mask": row_mask,
    "column_mask": column_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
resnet34_ibn_a Ran
ran; our contract draft was wrong, not the code
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/model/_ibn_models.py
code served (permissive licence) · get_code("b2ae4997a88217a1")
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 IBNNet. 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)

class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, ibn=None, stride=1, downsample=None):
        super(BasicBlock, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(planes)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = x
        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)
        out = self.conv2(out)
        out = self.bn2(out)
        if self.downsample is not None:
            residual = self.downsample(x)
        out += residual
        out = self.relu(out)
        return out


class InstanceBatchNorm2d(nn.Module):
    def __init__(self, planes, ratio=0.5):
        super(InstanceBatchNorm2d, self).__init__()
        self.half = int(planes * ratio)
        self.IN = nn.InstanceNorm2d(self.half, affine=True)
        self.BN = nn.BatchNorm2d(planes - self.half)

    def forward(self, x):
        split = torch.split(x, self.half, 1)
        out1 = self.IN(split[0].contiguous())
        out2 = self.BN(split[1].contiguous())
        out = torch.cat((out1, out2), 1)
        return out


class IBNNet(nn.Module):
    def __init__(self, block, layers, ibn_cfg=('a', 'a', 'a', None)):
        self.inplanes = 64
        super(IBNNet, self).__init__()
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
        if ibn_cfg[0] == 'b':
            self.bn1 = nn.InstanceNorm2d(64, affine=True)
        else:
            self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0], ibn=ibn_cfg[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2, ibn=ibn_cfg[1])
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2, ibn=ibn_cfg[2])
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2, ibn=ibn_cfg[3])
        self._out_features = 512 * block.expansion
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                m.weight.data.normal_(0, math.sqrt(2.0 / n))
            elif isinstance(m, nn.BatchNorm2d) or isinstance(m, nn.InstanceNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()

    def _make_layer(self, block, planes, blocks, stride=1, ibn=None):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(planes * block.expansion),
            )
        layers = []
        layers.append(block(self.inplanes, planes, ibn=ibn, stride=stride, downsample=downsample))
        self.inplanes = planes * block.expansion
        for _ in range(1, blocks):
            layers.append(block(self.inplanes, planes, ibn=ibn))
        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        return x


def resnet34_ibn_a(pretrained=False, **kwargs):
    model = IBNNet(BasicBlock, [3, 4, 6, 3], ibn_cfg=('a', 'a', 'a', None), **kwargs)
    return model


ARGS = {
    'pretrained': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type IBNNet — a IBNNet output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a IBNNet 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_HARDCODED_CLASSES Not yet run
unverified (harvested, no recorded run)
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/HT_HARDCODED.py
code served (permissive licence) · get_code("8b7318d02fdd57f5")
How it ran

no invocation recorded.

create_model Not yet run
unverified (harvested, no recorded run)
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/util/model.py
code served (permissive licence) · get_code("15b6ce87cea6a1a2")
How it ran

no invocation recorded.

get_model Not yet run
unverified (harvested, no recorded run)
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/util/model.py
code served (permissive licence) · get_code("619286e8e6e8c6ad")
How it ran

no invocation recorded.

mask_predict_score Not yet run
unverified (harvested, no recorded run)
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/util/evaluate.py
code served (permissive licence) · get_code("a95000b4197680ef")
How it ran

no invocation recorded.

resnet18_ibn_a Not yet run
unverified (harvested, no recorded run)
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/model/_ibn_models.py
code served (permissive licence) · get_code("c7b47196a0364d1b")
How it ran

no invocation recorded.

resnet50_ibn_a Not yet run
unverified (harvested, no recorded run)
OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated/py/model/_ibn_models.py
code served (permissive licence) · get_code("3a805a0c566b4de7")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 0 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
dtn 1 other paper
1 dated earlier
2203.11819 (2022-03)
A Broad Study of Pre-training for Domain Generalization and Adaptation
lenet 1 other paper
1 dated earlier
2203.11819 (2022-03)
A Broad Study of Pre-training for Domain Generalization and Adaptation

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

Fine-tuning is arguably the most straightforward way to tailor a pre-trained model (e.g., a foundation model) to downstream applications, but it also comes with the risk of losing valuable knowledge the model had learned in pre-training. For example, fine-tuning a pre-trained classifier capable of recognizing a large number of classes to master a subset of classes at hand is shown to drastically degrade the model's accuracy in the other classes it had previously learned. As such, it is hard to further use the fine-tuned model when it encounters classes beyond the fine-tuning data. In this paper, we systematically dissect the issue, aiming to answer the fundamental question, "What has been damaged in the fine-tuned model?" To our surprise, we find that the fine-tuned model neither forgets the relationship among the other classes nor degrades the features to recognize these classes. Instead, the fine-tuned model often produces more discriminative features for these other classes, even if they were missing during fine-tuning! What really hurts the accuracy is the discrepant logit scales between the fine-tuning classes and the other classes, implying that a simple post-processing calibration would bring back the pre-trained model's capability and at the same time unveil the feature improvement over all classes. We conduct an extensive empirical study to demonstrate the robustness of our findings and provide preliminary explanations underlying them, suggesting new directions for future theoretical analysis. Our code is available at https://github.com/OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated.

For agents

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

get_harvested_code_for_paper("2409.16223")
get_code_for_paper("2409.16223")
have("2409.16223")

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

curl https://syntology.ai/api/ran/2409.16223.json

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

[![Syntology run record](https://syntology.ai/api/ran/2409.16223.svg)](https://syntology.ai/paper/2409.16223)

Connect an agent — have() is free.