SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1608.02644 · 2016

Holophrasm: a neural Automated Theorem Prover for higher-order logic

arXiv · PDF · Open in the Atlas

Code that ran

10 of 14 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 10 have no contract verdict; 0 failures of our instrument; 4 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/1608.02644.json · badge: /api/ran/1608.02644.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
dwhalen/holophrasm canonical (classifier label)
classifier basis: census
10 ran · 4 no recorded run
FunctionStatusWhere it lives
dictionary_merge Ran
ran on a synthesized input
dwhalen/holophrasm/data_utils5.py
code served (permissive licence) · get_code("7d0b96fa6f5331d0")
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 numpy as np

torch.manual_seed(0)

# Analyzing the function:
# - dictionary_merge takes two parameters: current and new
# - Both are dictionaries (based on usage: iteration with "for x in new", checking "x in current", comparing values with !=)
# - The function merges 'new' into 'current', returning None if there's a conflict, otherwise returning the merged dict

# Create two dictionaries with compatible values
current = {
    "model_dim": 512,
    "num_layers": 6,
    "learning_rate": 0.001,
    "batch_size": 32
}

new = {
    "num_layers": 6,  # Same value as current - no conflict
    "dropout": 0.1,    # New key not in current
    "activation": "relu"  # New key not in current
}

ARGS = {
    "current": current,
    "new": new
}
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 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
list_to_data Ran
ran on a synthesized input
dwhalen/holophrasm/payout_model_5_train.py
code served (permissive licence) · get_code("e4f3ecb14bd415e0")
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. 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

# Create a small stand-in class to represent items in list_of_pds
class PredictionData:
    def __init__(self, correct, wrong, hyps):
        self.correct = correct
        self.wrong = wrong
        self.hyps = hyps

# Create sample data
torch.manual_seed(0)

# Create some dummy tree objects (can be any object, function just iterates over them)
tree1 = {"type": "tree", "value": 1}
tree2 = {"type": "tree", "value": 2}
tree3 = {"type": "tree", "value": 3}
tree4 = {"type": "tree", "value": 4}

# Create hypothesis data (can be any object)
hyps1 = torch.randn(2, 4)
hyps2 = torch.randn(3, 4)

# Create PredictionData instances
pd1 = PredictionData(
    correct=[tree1, tree2],
    wrong=[tree3],
    hyps=hyps1
)

pd2 = PredictionData(
    correct=[tree4],
    wrong=[],
    hyps=hyps2
)

# Construct the ARGS dictionary
ARGS = {
    "list_of_pds": [pd1, pd2]
}
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 — a list 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
merge_graph_structures Ran
ran on a synthesized input
dwhalen/holophrasm/models.py
code served (permissive licence) · get_code("53bedbded781bd89")
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 numpy as np

class GraphStructure:
    def __init__(self, string, parents, left_sibling, right_sibling, depth, parent_arity, leaf_position, arity):
        self.string = string
        self.parents = parents
        self.left_sibling = left_sibling
        self.right_sibling = right_sibling
        self.depth = depth
        self.parent_arity = parent_arity
        self.leaf_position = leaf_position
        self.arity = arity

# Create first graph structure
gs1 = GraphStructure(
    string=['node_a', 'node_b', 'node_c'],
    parents=[-1, 0, 0],
    left_sibling=[-1, -1, 1],
    right_sibling=[-1, 2, -1],
    depth=[0, 1, 1],
    parent_arity=[0, 2, 2],
    leaf_position=[0, 1, 2],
    arity=[2, 0, 0]
)

# Create second graph structure
gs2 = GraphStructure(
    string=['node_d', 'node_e'],
    parents=[-1, 0],
    left_sibling=[-1, -1],
    right_sibling=[-1, -1],
    depth=[0, 1],
    parent_arity=[0, 1],
    leaf_position=[0, 1],
    arity=[1, 0]
)

# Create parameter tensors
torch.manual_seed(0)
param1 = torch.randn(4, 3)
param2 = torch.randn(2, 5)

ARGS = {
    'gs_list': [gs1, gs2],
    'params_list': [param1, param2]
}
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 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
slow_delete_duplicates Ran
ran on a synthesized input
dwhalen/holophrasm/build_payout_data_set.py
code served (permissive licence) · get_code("7f6b640ff2600f3b")
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

torch.manual_seed(0)

# slow_delete_duplicates takes a list and removes duplicates while preserving order
# It checks membership with "in" operator, so we need hashable items
# Let's use a simple list of integers

inlist = [1, 2, 3, 2, 4, 1, 5, 3]

ARGS = {
    "inlist": inlist
}
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 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
test_data_set Ran
ran on a synthesized input
dwhalen/holophrasm/gen_model_train.py
code served (permissive licence) · get_code("3ec00b3b7aa08090")
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 numpy as np

torch.manual_seed(0)

class Proposition:
    def __init__(self, arity):
        self._arity = arity
    
    def unconstrained_arity(self):
        return self._arity

class ProofStep:
    def __init__(self, prop_arity):
        self.prop = Proposition(prop_arity)

class LanguageModel:
    def __init__(self):
        self.test_proof_steps = [
            ProofStep(0),
            ProofStep(1),
            ProofStep(0),
            ProofStep(2),
            ProofStep(3),
        ]

lm = LanguageModel()

ARGS = {
    'lm': lm
}
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 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
test_data_set Ran
ran on a synthesized input
dwhalen/holophrasm/pred_model.py
code served (permissive licence) · get_code("9ee1ed4c5a709c17")
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. 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 MockLanguageModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
        self.test_proof_steps = [
            (torch.randn(2, 4), torch.randn(2, 3)),
            (torch.randn(2, 4), torch.randn(2, 3))
        ]
    
    def forward(self, x):
        return self.linear(x)

lm = MockLanguageModel()

ARGS = {
    "lm": lm
}
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 — a list 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
training_data_set Ran
ran on a synthesized input
dwhalen/holophrasm/gen_model_train.py
code served (permissive licence) · get_code("3781549407b86440")
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 ndarray, ndim 1, dtype family float, shape (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 numpy as np

class MockProposition:
    def __init__(self, arity):
        self._arity = arity
    
    def unconstrained_arity(self):
        return self._arity

class MockProofStep:
    def __init__(self, arity):
        self.prop = MockProposition(arity)

class MockLanguageModel:
    def __init__(self):
        self.training_proof_steps = [
            MockProofStep(0),
            MockProofStep(1),
            MockProofStep(0),
            MockProofStep(2),
            MockProofStep(3),
            MockProofStep(1),
        ]

lm = MockLanguageModel()

ARGS = {
    "lm": lm
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (4,) — shape (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
training_data_set Ran
ran on a synthesized input
dwhalen/holophrasm/pred_model.py
code served (permissive licence) · get_code("9f0d681dfe32d4c8")
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 ndarray, ndim 1, dtype family int, shape (8,) 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 numpy as np

class MockLanguageModel:
    def __init__(self):
        # training_proof_steps should be a numpy array or list that can be permuted
        self.training_proof_steps = np.array([1, 2, 3, 4, 5, 6, 7, 8])

lm = MockLanguageModel()

ARGS = {
    "lm": lm
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (8,) — shape (8,) 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
validation_data_set Ran
ran on a synthesized input
dwhalen/holophrasm/gen_model_train.py
code served (permissive licence) · get_code("7f3e5f63ef491e6d")
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

torch.manual_seed(0)

class Prop:
    def __init__(self, arity):
        self._arity = arity
    
    def unconstrained_arity(self):
        return self._arity

class ProofStep:
    def __init__(self, arity):
        self.prop = Prop(arity)

class LanguageModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.validation_proof_steps = [
            ProofStep(0),
            ProofStep(1),
            ProofStep(0),
            ProofStep(2),
            ProofStep(3),
        ]

lm = LanguageModel()

ARGS = {
    "lm": lm
}
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
validation_data_set Ran
ran on a synthesized input
dwhalen/holophrasm/pred_model.py
code served (permissive licence) · get_code("d68d4b10e52cc4fd")
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. 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 LanguageModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
        # validation_proof_steps is accessed as an attribute
        self.validation_proof_steps = [
            (torch.randn(2, 4), torch.randn(2, 3)),
            (torch.randn(2, 4), torch.randn(2, 3))
        ]
    
    def forward(self, x):
        return self.linear(x)

lm = LanguageModel()

ARGS = {
    'lm': lm
}
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 — a list 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
get_wrong Not yet run
unverified (harvested, no recorded run)
dwhalen/holophrasm/build_payout_data_set.py
code served (permissive licence) · get_code("83f3b97d976fb076")
How it ran

no invocation recorded.

prop_applies_to_statement Not yet run
unverified (harvested, no recorded run)
dwhalen/holophrasm/data_utils5.py
code served (permissive licence) · get_code("ad8fc7eac91efe6f")
How it ran

no invocation recorded.

test_data_set Not yet run
unverified (harvested, no recorded run)
dwhalen/holophrasm/payout_model_5_train.py
code served (permissive licence) · get_code("6518d7d784e7aaa9")
How it ran

no invocation recorded.

validation_data_set Not yet run
unverified (harvested, no recorded run)
dwhalen/holophrasm/payout_model_5_train.py
code served (permissive licence) · get_code("bdf57915ccc9a068")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 2 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 2 later) and 12 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
get_wrong 1 other paper
1 later
2002.07019 (2020-02)
Learning to Prove Theorems by Learning to Generate Theorems
slow_delete_duplicates 1 other paper
1 later
2002.07019 (2020-02)
Learning to Prove Theorems by Learning to Generate Theorems

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

I propose a system for Automated Theorem Proving in higher order logic using deep learning and eschewing hand-constructed features. Holophrasm exploits the formalism of the Metamath language and explores partial proof trees using a neural-network-augmented bandit algorithm and a sequence-to-sequence model for action enumeration. The system proves 14% of its test theorems from Metamath's set.mm module.

For agents

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

get_harvested_code_for_paper("1608.02644")
get_code_for_paper("1608.02644")
have("1608.02644")

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

curl https://syntology.ai/api/ran/1608.02644.json

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

[![Syntology run record](https://syntology.ai/api/ran/1608.02644.svg)](https://syntology.ai/paper/1608.02644)

Connect an agent — have() is free.