SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2607.01022 · 2026

A Unified Benchmarking Framework for Spatiotemporal Event Modeling

Sebastian Vollmer, Yahya Aalaila, Gerrit Grossmann

arXiv · PDF · Open in the Atlas

Code that ran

9 of 11 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, 2 violated it and 0 have no contract verdict; 7 failures of our instrument; 2 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/2607.01022.json · badge: /api/ran/2607.01022.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
YahyaAalaila/seahorse canonical (classifier label)
classifier basis: census
9 ran · 2 no recorded run
FunctionStatusWhere it lives
get_event_model_capabilities Ran
ran; our contract draft was wrong, not the code
YahyaAalaila/seahorse/seahorse/models/sampling.py
code served (permissive licence) · get_code("494ca1811ff31d0f")
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 dict. 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)

class EventModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.capabilities = {
            'supports_batching': True,
            'max_batch_size': 32,
            'features': ['event_tracking', 'logging']
        }

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.event_model = EventModel()
        self.linear = nn.Linear(4, 3)

model = Model()

ARGS = {
    'model': model
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
load_jsonl Ran
ran; our contract draft was wrong, not the code
YahyaAalaila/seahorse/seahorse/utils.py
code served (permissive licence) · get_code("fd2e5730d6853aae")
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 json
import tempfile
import os

# Create a temporary JSONL file with test data
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test.jsonl")

# Write test data to the temporary file
test_records = [
    # Record with events containing time, location, and mark
    {
        "events": [
            {"t": 0.1, "x": 1.0, "y": 2.0, "mark": "A"},
            {"t": 0.5, "x": 1.5, "y": 2.5, "mark": "B"}
        ]
    },
    # Record with compact aliases
    {
        "time": [0.2, 0.7],
        "x": [2.0, 2.5],
        "y": [3.0, 3.5]
    },
    # Record with canonical keys already present
    {
        "times": [0.3, 0.8],
        "locations": [[4.0, 5.0], [4.5, 5.5]]
    }
]

with open(temp_file, 'w') as f:
    for record in test_records:
        f.write(json.dumps(record) + '\n')

ARGS = {
    "path": temp_file
}
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 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
load_splits_dir Ran
ran; our contract draft was wrong, not the code
YahyaAalaila/seahorse/seahorse/utils.py
code served (permissive licence) · get_code("be831c6aeb7f4250")
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 dict. 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 tempfile
import json
from pathlib import Path

# Create a temporary directory structure for testing
tmpdir = tempfile.TemporaryDirectory()
splits_dir = tmpdir.name

# Create dataset_a with train.jsonl, val.jsonl, test.jsonl
dataset_a_dir = Path(splits_dir) / "dataset_a"
dataset_a_dir.mkdir(parents=True, exist_ok=True)

# Create sample JSONL records with canonicalized format
train_records = [
    {"times": [0.1, 0.5, 1.2], "locations": [[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], "marks": [1, 2, 1]},
    {"times": [0.2, 0.8], "locations": [[0.5, 1.5], [2.5, 3.5]]}
]

val_records = [
    {"times": [0.3, 0.9], "locations": [[1.5, 2.5], [4.0, 5.0]]}
]

test_records = [
    {"times": [0.4, 1.0, 1.5], "locations": [[2.0, 3.0], [5.0, 6.0], [7.0, 8.0]]}
]

# Write train.jsonl
with open(dataset_a_dir / "train.jsonl", "w") as f:
    for record in train_records:
        f.write(json.dumps(record) + "\n")

# Write val.jsonl
with open(dataset_a_dir / "val.jsonl", "w") as f:
    for record in val_records:
        f.write(json.dumps(record) + "\n")

# Write test.jsonl
with open(dataset_a_dir / "test.jsonl", "w") as f:
    for record in test_records:
        f.write(json.dumps(record) + "\n")

# Create dataset_b with train.jsonl, val.jsonl (no test.jsonl)
dataset_b_dir = Path(splits_dir) / "dataset_b"
dataset_b_dir.mkdir(parents=True, exist_ok=True)

train_records_b = [
    {"times": [0.1, 0.7], "locations": [[1.0, 1.5], [2.0, 2.5]]}
]

val_records_b = [
    {"times": [0.2], "locations": [[3.0, 3.5]]}
]

with open(dataset_b_dir / "train.jsonl", "w") as f:
    for record in train_records_b:
        f.write(json.dumps(record) + "\n")

with open(dataset_b_dir / "val.jsonl", "w") as f:
    for record in val_records_b:
        f.write(json.dumps(record) + "\n")

# Define ARGS for load_splits_dir
ARGS = {
    "splits_dir": splits_dir,
    "datasets": ["dataset_a", "dataset_b"]
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
parse_overrides Ran
ran; our contract draft was wrong, not the code
YahyaAalaila/seahorse/seahorse/utils.py
code served (permissive licence) · get_code("add840e516a362d7")
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 dict. 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 sys

def parse_overrides(override_list: list[str]) -> dict:
    """Convert ['training.n_epochs=50', 'model.hidden_dim=128'] to nested dict."""
    result = {}
    for item in override_list or []:
        if '=' not in item:
            print(f'WARNING: ignoring malformed override {item!r} (expected key=value)', file=sys.stderr)
            continue
        key, _, raw_val = item.partition('=')
        try:
            val = int(raw_val)
        except ValueError:
            try:
                val = float(raw_val)
            except ValueError:
                if raw_val.lower() in ('true', 'false'):
                    val = raw_val.lower() == 'true'
                else:
                    val = raw_val
        parts = key.split('.')
        node = result
        for part in parts[:-1]:
            node = node.setdefault(part, {})
        node[parts[-1]] = val
    return result


ARGS = {
    'override_list': [
        'training.n_epochs=50',
        'model.hidden_dim=128',
        'learning_rate=0.001',
        'use_dropout=true',
        'batch_size=32',
        'model.layers.0.activation=relu',
        'malformed_override',
    ]
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
register_event Ran
ran; our contract draft was wrong, not the code
YahyaAalaila/seahorse/seahorse/models/model_registry.py
code served (permissive licence) · get_code("1ae1b7f8f759ab09")
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 torch
import torch.nn as nn

torch.manual_seed(0)

# The function register_event is a decorator factory that takes a key string
# and returns a decorator. The decorator registers a class in _EVENT_REGISTRY.
# To call this function, we need to provide a key parameter (str).

# First, initialize the registry that the decorator uses
_EVENT_REGISTRY = {}

# Create a simple EventModel class to use as a test
class SimpleEventModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# The key parameter is simply a string identifier
ARGS = {
    "key": "simple_event_model"
}
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 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
register_spatial Ran
ran; our contract draft was wrong, not the code
YahyaAalaila/seahorse/seahorse/models/model_registry.py
code served (permissive licence) · get_code("2d53f8efb8c38157")
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 torch
import torch.nn as nn

# Initialize the registry that the decorator will use
_SPATIAL_REGISTRY = {}

# Define ARGS for the register_spatial function
ARGS = {
    "key": "test_spatial_decoder"
}
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 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
register_state Ran
ran; our contract draft was wrong, not the code
YahyaAalaila/seahorse/seahorse/models/model_registry.py
code served (permissive licence) · get_code("52d3432cec7f9ab2")
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 torch
import torch.nn as nn

# Initialize torch for reproducibility
torch.manual_seed(0)

# Create a simple StateModel class to use as a stand-in
class SimpleStateModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# The function register_state takes a single parameter: key (a string)
# The decorator pattern means we need to provide a key that will be used
# to register a StateModel class in a registry dictionary

ARGS = {
    "key": "simple_state_model"
}
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 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
supports_intensity_query Ran
ran, violated the contract we drafted
YahyaAalaila/seahorse/seahorse/models/sampling.py
code served (permissive licence) · get_code("bee148d07ac6c2ac")
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 torch.nn as nn

torch.manual_seed(0)

class Capabilities:
    def __init__(self, has_intensity=True):
        self.has_intensity = has_intensity

class EventModel:
    def __init__(self, capabilities=None):
        self.capabilities = capabilities

class TestModel(nn.Module):
    def __init__(self, event_model=None):
        super().__init__()
        self.event_model = event_model

# Test case 1: Model with event_model that has capabilities with has_intensity=True
capabilities_with_intensity = Capabilities(has_intensity=True)
event_model_with_caps = EventModel(capabilities=capabilities_with_intensity)
model_with_intensity = TestModel(event_model=event_model_with_caps)

ARGS = {
    "model": model_with_intensity
}
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 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
supports_native_sampling Ran
ran, violated the contract we drafted
YahyaAalaila/seahorse/seahorse/models/sampling.py
code served (permissive licence) · get_code("5a098f726c669185")
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 torch.nn as nn

class Capabilities:
    def __init__(self, has_native_sampler=False):
        self.has_native_sampler = has_native_sampler

class EventModel(nn.Module):
    def __init__(self, capabilities=None):
        super().__init__()
        self.capabilities = capabilities

class ModelWithEventModel(nn.Module):
    def __init__(self, event_model=None):
        super().__init__()
        self.event_model = event_model

class ModelWithoutEventModel(nn.Module):
    def __init__(self):
        super().__init__()

torch.manual_seed(0)

# Case 1: Model with event_model that has capabilities with has_native_sampler=True
caps_true = Capabilities(has_native_sampler=True)
event_model_true = EventModel(capabilities=caps_true)
model_with_native_sampling = ModelWithEventModel(event_model=event_model_true)

# Case 2: Model with event_model that has capabilities with has_native_sampler=False
caps_false = Capabilities(has_native_sampler=False)
event_model_false = EventModel(capabilities=caps_false)
model_without_native_sampling = ModelWithEventModel(event_model=event_model_false)

# Case 3: Model with event_model but no capabilities
event_model_no_caps = EventModel(capabilities=None)
model_with_event_model_no_caps = ModelWithEventModel(event_model=event_model_no_caps)

# Case 4: Model without event_model
model_without_event_model = ModelWithoutEventModel()

ARGS = {
    "model": model_with_native_sampling
}
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 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
friendly_name_for_preset Not yet run
unverified (harvested, no recorded run)
YahyaAalaila/seahorse/seahorse/api/model_class_map.py
code served (permissive licence) · get_code("1e5cd7b265d06e2a")
How it ran

no invocation recorded.

resolve_preset Not yet run
unverified (harvested, no recorded run)
YahyaAalaila/seahorse/seahorse/api/model_class_map.py
code served (permissive licence) · get_code("f276ba834a4d6935")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 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 11 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

Spatiotemporal point processes (STPPs) model event data in continuous time and space, with applications in mobility, epidemiology, and public safety. Recent neural STPPs span expressive intensity models, conditional density models, continuoustime latent dynamics, normalizing-flow spatial decoders, and score-based generative mechanisms. Yet comparison remains fragile because implementations differ in preprocessing, coordinate normalization, splits, likelihood conventions, and evaluation protocols. We present SEAHORSE, a unified framework for reproducible STPP experimentation. SEAHORSE formalizes neural STPPs through a common encode-evolve-decode interface and trains, tunes, and evaluates every model family under a single executable benchmark protocol with raw-coordinate likelihood reporting. This enables fair comparisons but, more importantly, controlled diagnostic studies. We pair SEAHORSE with HawkesNest, a synthetic stress-test suite, and show that increasing event-pattern complexity exposes each family's inductive bias, degrading some models sharply and leaving others stable. Code. SEAHORSE is available at github.com/YahyaAalaila/seahorse; install with pip install seahorse-stpp. The v0.1.0 release is archived on Zenodo [1].

For agents

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

get_harvested_code_for_paper("2607.01022")
get_code_for_paper("2607.01022")
have("2607.01022")

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

curl https://syntology.ai/api/ran/2607.01022.json

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

[![Syntology run record](https://syntology.ai/api/ran/2607.01022.svg)](https://syntology.ai/paper/2607.01022)

Connect an agent — have() is free.