SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2010.02830 · EMNLP · 2020

PROVER: Proof Generation for Interpretable Reasoning over Rules

Mohit Bansal, Sayan Ghosh, Swarnadeep Saha, Shashank Srivastava

arXiv · PDF · Open in the Atlas

Code that ran

5 of 12 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 3 have no contract verdict; 2 failures of our instrument; 7 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/2010.02830.json · badge: /api/ran/2010.02830.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
swarnaHub/PRover canonical (classifier label)
classifier basis: census
5 ran · 7 no recorded run
FunctionStatusWhere it lives
compute_metrics Ran
ran on a synthesized input
swarnaHub/PRover/utils.py
code served (permissive licence) · get_code("edca3480666257aa")
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)

# Analyze the function:
# - compute_metrics takes: task_name (str), preds (array-like), labels (array-like)
# - It checks len(preds) == len(labels)
# - For task_name 'rr' or 'rr_qa', it calls simple_accuracy(preds, labels)
# - simple_accuracy does: (preds == labels).mean()
# - This requires preds and labels to support == comparison and .mean()

# Create small numpy arrays that support these operations
preds = np.array([0, 1, 2, 1])
labels = np.array([0, 1, 2, 1])

ARGS = {
    'task_name': 'rr',
    'preds': preds,
    'labels': labels
}
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
get_index_component_maps Ran
ran on a synthesized input
swarnaHub/PRover/evaluation/print_graphs.py
code served (permissive licence) · get_code("89fc1d14be606f95")
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 os
import json
import tempfile

# Create a temporary directory for test files
temp_dir = tempfile.mkdtemp()

# Create test.jsonl file
test_data = [
    {
        "meta": {
            "sentenceScramble": [1, 2, 3]
        },
        "questions": [
            {"id": 1},
            {"id": 2}
        ]
    },
    {
        "meta": {
            "sentenceScramble": [2, 1, 3]
        },
        "questions": [
            {"id": 3}
        ]
    }
]

test_file_path = os.path.join(temp_dir, 'test.jsonl')
with open(test_file_path, 'w', encoding='utf-8') as f:
    for record in test_data:
        f.write(json.dumps(record) + '\n')

# Create meta-test.jsonl file
meta_test_data = [
    {
        "NFact": 2,
        "NRule": 1
    },
    {
        "NFact": 2,
        "NRule": 1
    }
]

meta_test_file_path = os.path.join(temp_dir, 'meta-test.jsonl')
with open(meta_test_file_path, 'w', encoding='utf-8') as f:
    for record in meta_test_data:
        f.write(json.dumps(record) + '\n')

ARGS = {
    'data_dir': temp_dir
}
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
get_natlang_mappings Ran
ran on a synthesized input
swarnaHub/PRover/evaluation/eval_natlang.py
code served (permissive licence) · get_code("5fe3351fd927fc60")
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 OrderedDict. 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 os
from collections import OrderedDict

# Create a temporary TSV file with sample natlang metadata
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.tsv', encoding='utf-8')

# Write header and sample data rows
# Format: id-suffix \t name \t col2 \t col3 \t description \t col5 \t rf_ids
temp_file.write("id\tname\tcol2\tcol3\tdescription\tcol5\trf_ids\n")
temp_file.write("Att1-suffix1\ttest_name1\tvalue2\tvalue3\ttest_desc1\tvalue5\trf001 rf002\n")
temp_file.write("Att1-suffix2\ttest_name2\tvalue2\tvalue3\ttest_desc2\tvalue5\trf003\n")
temp_file.write("AttNoneg2-suffix3\ttest_name3\tvalue2\tvalue3\ttest_desc3\tvalue5\trf004 rf005 rf006\n")
temp_file.write("\n")
temp_file.write("Att3-suffix4\ttest_name4\tvalue2\tvalue3\ttest_desc4\tvalue5\trf007\n")

temp_file.close()

ARGS = {
    'natlang_metadata': temp_file.name
}
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 OrderedDict — a OrderedDict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a OrderedDict 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_proof_graph Ran
ran; our contract draft was wrong, not the code
swarnaHub/PRover/proof_utils.py
code served (permissive licence) · get_code("c13d0749a5472646")
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.

# Setup code to construct valid arguments for get_proof_graph function

# The function takes a single parameter: proof_str (a string representing a proof structure)
# Looking at the function body, it:
# - Replaces parentheses and splits the string
# - Processes tokens including '(', ')', '[', ']', '->', and node names
# - Builds a graph of nodes and edges

# Create a valid proof string that exercises the function logic
proof_str = "(node1 -> node2 (node3 -> node4))"

ARGS = {
    "proof_str": proof_str
}
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 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
get_proof_graph_with_fail Ran
ran; our contract draft was wrong, not the code
swarnaHub/PRover/proof_utils.py
code served (permissive licence) · get_code("5621c0d34ff2bff6")
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

def get_proof_graph_with_fail(proof_str):
    proof_str = proof_str[:-2].split('=')[1].strip()[1:-1]
    nodes = proof_str.split(' <- ')
    all_nodes = []
    all_edges = []
    for i in range(len(nodes) - 1):
        all_nodes.append(nodes[i])
        if nodes[i + 1] != 'FAIL':
            all_edges.append((nodes[i + 1], nodes[i]))
    return (all_nodes, all_edges)


# Construct a valid proof_str argument
# The function expects a string that ends with "= [...]"
# Inside the brackets are nodes separated by " <- "
# Example: "proof= [node1 <- node2 <- node3 <- FAIL]"
proof_str = "proof= [step1 <- step2 <- step3 <- FAIL]"

ARGS = {
    "proof_str": proof_str
}
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 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
convert_examples_to_features_RR Not yet run
unverified (harvested, no recorded run)
swarnaHub/PRover/utils.py
code served (permissive licence) · get_code("ecb790feacadaaad")
How it ran

no invocation recorded.

convert_examples_to_features_RR Not yet run
unverified (harvested, no recorded run)
swarnaHub/PRover/utils_natlang.py
code served (permissive licence) · get_code("3f327d827ee113e3")
How it ran

no invocation recorded.

filter_context Not yet run
unverified (harvested, no recorded run)
swarnaHub/PRover/evaluation/eval_natlang.py
code served (permissive licence) · get_code("abf6d8b134f4edd6")
How it ran

no invocation recorded.

get_fact_rule_identifiers Not yet run
unverified (harvested, no recorded run)
swarnaHub/PRover/ilp_infer/inference.py
code served (permissive licence) · get_code("555f983801abb905")
How it ran

no invocation recorded.

simple_accuracy Not yet run
unverified (harvested, no recorded run)
swarnaHub/PRover/utils.py
code served (permissive licence) · get_code("3c241ecfe3749a6d")
How it ran

no invocation recorded.

solve_LP Not yet run
unverified (harvested, no recorded run)
swarnaHub/PRover/ilp_infer/inference.py
code served (permissive licence) · get_code("fcbaf738dc6fc42a")
How it ran

no invocation recorded.

solve_LP_unconstrained Not yet run
unverified (harvested, no recorded run)
swarnaHub/PRover/ilp_infer/inference.py
code served (permissive licence) · get_code("f7b1e7bc7db74fa7")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 4 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 1 the same month, 4 later) and 8 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
simple_accuracy 59 other papers
19 dated earlier, 2 the same month, 37 later, 1 same year, order not determinable
1506.05865 (2015-06)
LCSTS: A Large Scale Chinese Short Text Summarization Dataset
1509.01626 (2015-09)
Character-level Convolutional Networks for Text Classification *
1810.04805 (2018-10)
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
and 56 more not named here
get_proof_graph 3 other papers
3 later
2106.01354 (2021-06)
MULTIPROVER: Generating Multiple Proofs for Improved Interpretability in Rule Reasoning
2203.10261 (2022-03)
FaiRR: Faithful and Robust Deductive Reasoning over Natural Language
2310.14491 (2023-10)
Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Models
get_proof_graph_with_fail 2 other papers
2 later
2106.01354 (2021-06)
MULTIPROVER: Generating Multiple Proofs for Improved Interpretability in Rule Reasoning
2310.14491 (2023-10)
Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Models
convert_examples_to_features_RR 1 other paper
1 later
2106.01354 (2021-06)
MULTIPROVER: Generating Multiple Proofs for Improved Interpretability in Rule Reasoning

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

shows that transformers can act as "soft theorem provers" by answering questions over explicitly provided knowledge in natural language. In our work, we take a step closer to emulating formal theorem provers, by proposing PROVER, an interpretable transformer-based model that jointly answers binary questions over rule-bases and generates the corresponding proofs. Our model learns to predict nodes and edges corresponding to proof graphs in an efficient constrained training paradigm. During inference, a valid proof, satisfying a set of global constraints is generated. We conduct experiments on synthetic, hand-authored, and human-paraphrased rule-bases to show promising results for QA and proof generation, with strong generalization performance. First, PROVER generates proofs with an accuracy of 87%, while retaining or improving performance on the QA task, compared to RuleTakers (up to 6% improvement on zero-shot evaluation). Second, when trained on questions requiring lower depths of reasoning, it generalizes significantly better to higher depths (up to 15% improvement). Third, PROVER obtains near perfect QA accuracy of 98% using only 40% of the training data. However, generating proofs for questions requiring higher depths of reasoning becomes challenging, and the accuracy drops to 65% for "depth 5", indicating significant scope for future work. 1 Facts : F 1 : The bald eagle eats the lion. F2: The bald eagle sees the tiger. F3: The lion chases the bald eagle. F 4 : The lion eats the mouse. F5: The mouse eats the tiger. F6: The tiger eats the bald eagle. F 7 : The tiger is red. Rules : R1: If the lion is green and the lion is not kind then the lion sees the bald eagle. R2: If someone sees the lion then they eat the mouse. R 3 : If someone is kind and not green then they see the bald eagle. R4: If someone is rough then they see the lion. R5: If someone sees the lion and they do not eat the tiger then the tiger is rough. R 6 : If someone eats the bald eagle and the bald eagle is not kind then the bald eagle is rough. R7: If someone does not eat the lion then the lion is big. R8: If someone is kind then they do not eat the mouse. The bald eagle eats the mouse. [ Answer : T ] Q5: The tiger does not eat the mouse.

For agents

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

get_harvested_code_for_paper("2010.02830")
get_code_for_paper("2010.02830")
have("2010.02830")

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

curl https://syntology.ai/api/ran/2010.02830.json

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

[![Syntology run record](https://syntology.ai/api/ran/2010.02830.svg)](https://syntology.ai/paper/2010.02830)

Connect an agent — have() is free.