SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.21043 · 2026

T1: One-to-One Channel-Head Binding for Multivariate Time-Series Imputation

Hyunwoo Ryu, Hyung-Sin Kim, Keondo Park, Dongik Park, Suahn Bae, P Conv, D Conv

arXiv · PDF · Open in the Atlas

Code that ran

15 of 20 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; 3 honoured the contract we drafted, 0 violated it and 10 have no contract verdict; 2 failures of our instrument; 5 no recorded run), measured 2026-11-22; 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/2602.21043.json · badge: /api/ran/2602.21043.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
Oppenheimerdinger/T1 community
Papers with Code archive
15 ran · 5 no recorded run
FunctionStatusWhere it lives
ClassHead Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("5cbf305b5e20f778")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float, shape (2, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

class Swish(nn.Module):
    def forward(self, x):
        return x * torch.sigmoid(x)

class ClassHead(nn.Module):
    def __init__(self, cfg, T_out: int):
        super().__init__()
        input_features_to_linear = cfg.enc_in * cfg.n_heads
        proj_dim = cfg.projection_head_dim
        self.projection = nn.Linear(input_features_to_linear, proj_dim)
        self.activation = Swish()
        self.norm = nn.LayerNorm(proj_dim)
        self.dropout = nn.Dropout(cfg.drop_head)
        self.linear = nn.Linear(proj_dim, cfg.num_class)
        self.final_pool_type = getattr(cfg, 'final_pool_type', 'gap').lower()
        if self.final_pool_type == 'gap':
            self.final_pool = lambda x: x.mean(dim=1)
        elif self.final_pool_type == 'maxpool':
            self.final_pool = lambda x: torch.max(x, dim=1).values
        else:
            raise ValueError(f"Unsupported final pooling type: {self.final_pool_type}. Choose 'gap' or 'maxpool'.")

    def forward(self, x):
        B, M, H, T = x.shape
        x = x.reshape(B, M * H, T)
        x = self.projection(x.transpose(1, 2))
        x = self.activation(x)
        x = self.final_pool(x)
        x = self.norm(x)
        x = self.dropout(x)
        x = self.linear(x)
        return x

class Config:
    def __init__(self):
        self.enc_in = 4
        self.n_heads = 3
        self.projection_head_dim = 8
        self.drop_head = 0.1
        self.num_class = 5
        self.final_pool_type = 'gap'

cfg = Config()

INIT_ARGS = {
    'cfg': cfg,
    'T_out': 16
}

ARGS = {
    'x': torch.randn(2, 3, 4, 16)
}

if __name__ == '__main__':
    model = ClassHead(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 5) — shape (2, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
DepthwiseMix Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("a7ba430d1c90ac52")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, 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

# INIT_ARGS: parameters for DepthwiseMix.__init__
INIT_ARGS = {
    'Cin': 6,
    'Cout': 6,
    'kL': 5,
    'kS': 3,
    'bias': True
}

# ARGS: parameters for DepthwiseMix.forward
# Input shape: (batch, Cin, sequence_length)
ARGS = {
    'x': torch.randn(2, 6, 8)
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
DownSample Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("f94921018ed8273d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

def ceil_div(a: int, b: int) -> int:
    """Ceiling division: returns ⌈a/b⌉"""
    return (a + b - 1) // b

class DownSample(nn.Module):

    def __init__(self, k: int, s: int, C: int):
        super().__init__()
        self.k, self.s = (k, s)
        self.dw = nn.Conv1d(C, C, k, s, groups=C)

    def forward(self, x, *args):
        B, M, C, T = x.shape
        out_len = ceil_div(T, self.s)
        total_len_needed = (out_len - 1) * self.s + self.k
        pad = max(0, total_len_needed - T)
        if pad:
            x = torch.cat([x[..., -1:].repeat(1, 1, 1, pad), x], -1)
        x = self.dw(x.reshape(B * M, C, -1))
        return (x.reshape(B, M, C, x.size(-1)),) + args


INIT_ARGS = {
    'k': 3,
    's': 2,
    'C': 4,
}

ARGS = {
    'x': torch.randn(2, 3, 4, 8),
}

if __name__ == '__main__':
    model = DownSample(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shape:", output[0].shape)
    print("Output type:", type(output))
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
DropPath Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("96592475595fbed1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 4, dtype family float, shape (2, 4, 3, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

class DropPath(nn.Module):

    def __init__(self, drop_prob: float):
        super().__init__()
        self.drop_prob = drop_prob

    def forward(self, x):
        if not self.training or self.drop_prob == 0.0:
            return x
        keep_prob = 1 - self.drop_prob
        random_tensor = keep_prob + torch.rand(x.shape[0], 1, 1, 1, device=x.device, dtype=x.dtype)
        random_tensor.floor_()
        return x.div(keep_prob) * random_tensor


INIT_ARGS = {
    'drop_prob': 0.2
}

ARGS = {
    'x': torch.randn(2, 4, 3, 5)
}

if __name__ == '__main__':
    model = DropPath(**INIT_ARGS)
    model.train()
    output = model(**ARGS)
    print(f"Input shape: {ARGS['x'].shape}")
    print(f"Output shape: {output.shape}")
    print(f"Output:\n{output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 4, 3, 5) — shape (2, 4, 3, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
FeedForward Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("07ced2912044bc6c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 4, 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

class Config:
    def __init__(self):
        self.ffn_ratio = 2.0
        self.drop_ffn = 0.1

cfg = Config()
H = 8

INIT_ARGS = {
    'cfg': cfg,
    'H': H
}

B = 2
M = 3
T = 5

x = torch.randn(B, M, H, T)

ARGS = {
    'x': x
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ForecastHead Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("9f7cbb589c1b7f85")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, 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

class Config:
    def __init__(self):
        self.n_heads = 8
        self.projection_head_dim = 16
        self.pred_len = 12
        self.drop_head = 0.1

cfg = Config()
T_out = 5

INIT_ARGS = {
    'cfg': cfg,
    'T_out': T_out,
}

ARGS = {
    'x': torch.randn(2, 3, 8, 5),
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
FrontPadding Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("e1d41609713d2902")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 3, 18) 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

def ceil_div(a: int, b: int) -> int:
    """Ceiling division: returns ⌈a/b⌉"""
    return (a + b - 1) // b

class FrontPadding(nn.Module):

    def __init__(self, patch_size: int, stride: int):
        super().__init__()
        self.k, self.s = (patch_size, stride)

    def forward(self, x):
        T = x.size(-1)
        out_len = ceil_div(T, self.s)
        total_len_needed = (out_len - 1) * self.s + self.k
        pad = max(0, total_len_needed - T)
        if pad == 0:
            return x
        return torch.cat([x[..., -1:].repeat(*(1,) * (x.dim() - 1), pad), x], dim=-1)


INIT_ARGS = {
    'patch_size': 4,
    'stride': 2,
}

ARGS = {
    'x': torch.randn(2, 3, 16),
}

if __name__ == '__main__':
    model = FrontPadding(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Input shape: {ARGS['x'].shape}")
    print(f"Output shape: {output.shape}")
    print(f"Output:\n{output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 3, 18) — shape (2, 3, 18) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
PixelShuffle1D Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("19f9640048c6b72c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 2, 6) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

class PixelShuffle1D(nn.Module):

    def __init__(self, r: int):
        super().__init__()
        self.r = r

    def forward(self, x):
        B, C, L = x.shape
        assert C % self.r == 0
        out = x.reshape(B, C // self.r, self.r, L).permute(0, 1, 3, 2)
        return out.reshape(B, C // self.r, L * self.r)


INIT_ARGS = {
    'r': 2
}

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

if __name__ == '__main__':
    module = PixelShuffle1D(**INIT_ARGS)
    output = module(**ARGS)
    print(f"Input shape: {ARGS['x'].shape}")
    print(f"Output shape: {output.shape}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 2, 6) — shape (2, 2, 6) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ReconHead Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("d73ba875476c7a57")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 3, 12) 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

# Define a minimal config object
class Config:
    def __init__(self):
        self.seq_len = 8
        self.pred_len = 4
        self.task_name = 'forecast'
        self.n_heads = 4
        self.bottle_neck_n_heads = 0
        self.head_params_shared = True
        self.drop_head = 0.1

INIT_ARGS = {
    'cfg': Config(),
    'T_out': 3
}

# For forward pass: B=2, M=3, H=4, T=3
# x shape should be (B, M, H, T) = (2, 3, 4, 3)
ARGS = {
    'x': torch.randn(2, 3, 4, 3)
}

# Construct and call
if __name__ == '__main__':
    # Import the classes (they would be defined above in actual use)
    def ceil_div(a: int, b: int) -> int:
        return (a + b - 1) // b

    class PixelShuffle1D(nn.Module):
        def __init__(self, r: int):
            super().__init__()
            self.r = r

        def forward(self, x):
            B, C, L = x.shape
            assert C % self.r == 0
            out = x.reshape(B, C // self.r, self.r, L).permute(0, 1, 3, 2)
            return out.reshape(B, C // self.r, L * self.r)

    class ReconHead(nn.Module):
        def __init__(self, cfg, T_out: int):
            super().__init__()
            self.pred_len = cfg.seq_len + cfg.pred_len if cfg.task_name.endswith('forecast') else cfg.pred_len
            bottle_neck_n_heads = getattr(cfg, 'bottle_neck_n_heads', 0)
            if bottle_neck_n_heads > 0 and cfg.task_name == 'anomaly_detection':
                self.use_bottleneck = True
                self.bottleneck_dim = bottle_neck_n_heads
                self.proj_down = nn.Conv1d(cfg.n_heads, self.bottleneck_dim, 1)
                self.proj_up = nn.Conv1d(self.bottleneck_dim, cfg.n_heads, 1)
            else:
                self.use_bottleneck = False
            self.up = ceil_div(self.pred_len, T_out)
            self.adjusted_channels = (cfg.n_heads + self.up - 1) // self.up * self.up
            self.channel_adjust = nn.Conv1d(cfg.n_heads, self.adjusted_channels, 1)
            self.ps = PixelShuffle1D(self.up)
            self.ps_output_len = T_out * self.up
            outC = self.adjusted_channels // self.up
            self.proj = nn.Linear(outC, 1) if cfg.head_params_shared else nn.Conv1d(self.pred_len, self.pred_len, kernel_size=outC, groups=self.pred_len)
            self.dp = nn.Dropout(cfg.drop_head)

        def forward(self, x):
            B, M, H, T = x.shape
            if self.use_bottleneck:
                x = x.reshape(B * M, H, T)
                x = self.proj_down(x)
                x = self.proj_up(x)
                x = x.reshape(B, M, H, T)
            x = x.reshape(B * M, H, T)
            x = self.channel_adjust(x)
            y = self.ps(x.reshape(B, M * self.adjusted_channels, T)).reshape(B * M, self.adjusted_channels // self.up, self.ps_output_len)
            if self.ps_output_len > self.pred_len:
                crop_start = (self.ps_output_len - self.pred_len) // 2
                y = y[..., crop_start:crop_start + self.pred_len]
            y = y.transpose(1, 2)
            return self.dp(self.proj(y)).reshape(B, M, self.pred_len)

    model = ReconHead(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 3, 12)")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 3, 12) — shape (2, 3, 12) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
SACAReparameterization Ran
ran on a synthesized input
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("41c00d62d64041ca")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 4, 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

INIT_ARGS = {
    'enc_in': 5,
    'n_heads': 3,
    'mode': 1,
    'hidden_dim': 8,
    'drop_rate': 0.1,
    'param_type': 'sig'
}

ARGS = {
    'stats': torch.randn(2, 1, 5),
    'param_type': 'sig'
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
_get_saca_output_dim Ran
ran, honoured the contract we drafted
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("56e1096dafc14aa2")
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 int, ndim 0, dtype family int, 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 torch.nn as nn

torch.manual_seed(0)

# Construct arguments for _get_saca_output_dim
# The function takes: mode (int), enc_in (int), n_heads (int)
# and returns an int based on the mode

ARGS = {
    "mode": 2,
    "enc_in": 8,
    "n_heads": 4,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, 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
_reshape_saca_param Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("d4aeff43548b65fc")
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 4, dtype family float, shape (2, 4, 3, 1) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Test case 1: mode 0 with param
param_mode0 = torch.randn(2, 4)
mode_0 = 0
enc_in_0 = 4
n_heads_0 = 2

# Test case 2: mode 1 - reshape to [B, H, enc_in, 1]
param_mode1 = torch.randn(2, 12)  # B=2, enc_in=3, n_heads=4, so 3*4=12
mode_1 = 1
enc_in_1 = 3
n_heads_1 = 4

# Test case 3: mode 2 - reshape to [B, 1, enc_in, 1]
param_mode2 = torch.randn(2, 3)  # B=2, enc_in=3
mode_2 = 2
enc_in_2 = 3
n_heads_2 = 4

# Test case 4: mode 3 - reshape to [B, H, 1, 1]
param_mode3 = torch.randn(2, 4)  # B=2, n_heads=4
mode_3 = 3
enc_in_3 = 3
n_heads_3 = 4

# Test case 5: mode 4 - reshape to [B, 1, 1, 1]
param_mode4 = torch.randn(2, 1)  # B=2
mode_4 = 4
enc_in_4 = 3
n_heads_4 = 4

ARGS = {
    "param": param_mode1,
    "mode": mode_1,
    "enc_in": enc_in_1,
    "n_heads": n_heads_1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 4, 3, 1) — shape (2, 4, 3, 1) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ceil_div Ran
ran, honoured the contract we drafted
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("3f6428b8556da60f")
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 int, ndim 0, dtype family int, 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)

# ceil_div takes two integers and performs ceiling division
# We need to provide valid int arguments for both parameters

ARGS = {
    "a": 10,
    "b": 3,
}
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 int, ndim 0, dtype family int, 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
compute_out_len Ran
ran, honoured the contract we drafted
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("e6816109da80f6af")
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 int, ndim 0, dtype family int, 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)

# The function compute_out_len takes three integer parameters: L, k, s
# L: input length
# k: kernel size
# s: stride
# The function computes ceiling division of L by s

ARGS = {
    "L": 10,
    "k": 3,
    "s": 2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, 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
pick Ran
ran; our contract draft was wrong, not the code
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("5b15c79d1a0c3712")
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 int, ndim 0, dtype family int, 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 pick takes:
# - val: can be a list, tuple, or a scalar value
# - idx: an integer index
# It returns an int by either indexing into val (if it's a list/tuple) or converting val directly to int

# Test case 1: val is a list
val_list = [10, 20, 30, 40]
idx_value = 2

ARGS = {
    "val": val_list,
    "idx": idx_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, 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
Attention Not yet run
unverified (harvested, no recorded run)
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("a32c89a2966be2db")
How it ran

no invocation recorded.

Model Not yet run
unverified (harvested, no recorded run)
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("a2db8dfdfe40f459")
How it ran

no invocation recorded.

SACAModule Not yet run
unverified (harvested, no recorded run)
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("9a9342f7a853fe6c")
How it ran

no invocation recorded.

T1Block Not yet run
unverified (harvested, no recorded run)
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("213029c14c5f7221")
How it ran

no invocation recorded.

T1Stage Not yet run
unverified (harvested, no recorded run)
Oppenheimerdinger/T1/models/T1.py
code served (permissive licence) · get_code("71e3f87d20e2922e")
How it ran

no invocation recorded.

Also present in other papers' code

Of 20 samples lifted from this paper's linked repositories, 20 carry a function identity and 0 do not. Of the 20 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 20 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.

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

Imputing missing values in multivariate time series remains challenging, especially under diverse missing patterns and heavy missingness. Existing methods suffer from suboptimal performance as corrupted temporal features hinder effective cross-variable information transfer, amplifying reconstruction errors. Robust imputation requires both extracting temporal patterns from sparse observations within each variable and selectively transferring information across variables-yet current approaches excel at one while compromising the other. We introduce T1 (Time series imputation with 1-to-1 channel-head binding), a CNN-Transformer hybrid architecture that achieves robust imputation through Channel-Head Binding-a mechanism creating one-to-one correspondence between CNN channels and attention heads. This design enables selective information transfer: attention pathways adapt based on observable patterns, down-weighting corrupted channels while maintaining reliable cross-variable connections. Experiments on 11 benchmark datasets demonstrate that T1 achieves state-of-the-art performance, reducing MSE by 46% on average compared to the second-best baseline, with particularly strong gains under extreme sparsity (70% missing ratio). The model generalizes to unseen missing patterns without retraining and uses a consistent hyperparameter configuration across all datasets. The code is available at https://github.com/Oppenheimerdinger/T1.

For agents

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

get_harvested_code_for_paper("2602.21043")
get_code_for_paper("2602.21043")
have("2602.21043")

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

curl https://syntology.ai/api/ran/2602.21043.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.21043.svg)](https://syntology.ai/paper/2602.21043)

Connect an agent — have() is free.