SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2211.08671 · NeurIPS · 2022

LEMMA: Bootstrapping High-Level Mathematical Reasoning with Learned Symbolic Abstractions

Armando Solar-Lezama, Noah Goodman, Gabriel Poesia, Zhening Li, Omar Costilla-Reyes

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 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 4 have no contract verdict; 0 failures of our instrument; 0 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/2211.08671.json · badge: /api/ran/2211.08671.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
uranium11010/lemma canonical (classifier label)
classifier basis not recorded on the link
4 ran · 0 no recorded run
FunctionStatusWhere it lives
load_axioms Ran
ran on a synthesized input
uranium11010/lemma/abs_util.py
code served (permissive licence) · get_code("85b9cc3b36e518db")
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 json
import tempfile
import os

# Create a temporary JSON file with valid axiom data
temp_dir = tempfile.gettempdir()
temp_file = os.path.join(temp_dir, "test_axioms.json")

axiom_data = {
    "num": 3,
    "axioms": [
        "axiom_1: forall x, P(x)",
        "axiom_2: exists y, Q(y)",
        "axiom_3: R(a, b)"
    ]
}

with open(temp_file, 'w') as f:
    json.dump(axiom_data, f)

ARGS = {
    "f_name": 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 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
load_solutions Ran
ran on a synthesized input
uranium11010/lemma/abs_util.py
code served (permissive licence) · get_code("e10feb9746bf36c1")
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 json
import tempfile
import os

# Create a temporary JSON file with sample solutions
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "solutions.json")

solutions_data = {
    "solution_1": {"answer": 42, "steps": [1, 2, 3]},
    "solution_2": {"answer": "test", "steps": [4, 5, 6]}
}

with open(temp_file, 'w') as f:
    json.dump(solutions_data, f)

ARGS = {
    "f_name": 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 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
log Ran
ran on a synthesized input
uranium11010/lemma/compress.py
code served (permissive licence) · get_code("a3670b6ecc8de682")
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 float, 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 math

def log(x):
    """ For entropy calculations """
    return -1000.0 if x == 0 else math.log(x)

ARGS = {
    "x": 0.5
}
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 float, 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
make_rule_trie Ran
ran on a synthesized input
uranium11010/lemma/abs_util.py
code served (permissive licence) · get_code("c04834f139718d89")
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 Trie. 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 warnings

class Trie:

    def __init__(self, key=None, value=None, comment=None):
        self.key = key
        self.value = value
        self.comment = comment
        self.is_term = False
        self.children = {}

    def find(self, keys):
        node = self
        for key in keys:
            node = node.children.get(key)
            if node is None:
                return None
        return node if node.is_term else None

    def add(self, keys, value=None, comment=None):
        node = self
        path_exists = True
        for i, key in enumerate(keys):
            if node.children.get(key) is None:
                path_exists = False
                break
            node = node.children[key]
        if path_exists:
            if node.is_term == True and node.value != value:
                warnings.warn('Changing value of existing key')
            node.is_term = True
            node.value = value
            node.comment = comment
            return node
        final_child = old_child = Trie(keys[-1], value, comment)
        old_child.is_term = True
        for j in range(len(keys) - 2, i - 1, -1):
            new_child = Trie(keys[j])
            new_child.children[old_child.key] = old_child
            old_child = new_child
        node.children[old_child.key] = old_child
        return final_child

    def __str__(self):
        self_str = str(self.key) if self.value is None else str(self.key) + ': ' + str(self.value)
        if not self.children:
            return f'    {self_str}    '
        children = [list(reversed(str(child).split('\n'))) for child in self.children.values()]
        widths = [len(child[0]) for child in children]
        width = sum(widths)
        left_spaces = (width - len(self_str)) // 2 - 2
        right_spaces = (width - len(self_str) + 1) // 2 - 2
        out_str = '  ' + '_' * left_spaces + self_str + '_' * right_spaces + '  '
        while True:
            row = ''.join((child.pop() if child else ' ' * w for child, w in zip(children, widths)))
            out_str += f'\n{row}'
            if all((not child for child in children)):
                break
        return out_str

def make_rule_trie(rules):
    """
    Convert rules into trie
    """
    trie = Trie()
    for rule in rules:
        trie.add(list(rule), rule)
    return trie


rules = [
    "abc",
    "abd",
    "xyz"
]

ARGS = {
    "rules": rules
}
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 Trie — a Trie output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Trie 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

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 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 4 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

Humans tame the complexity of mathematical reasoning by developing hierarchies of abstractions. With proper abstractions, solutions to hard problems can be expressed concisely, thus making them more likely to be found. In this paper, we propose Learning Mathematical Abstractions (LEMMA): an algorithm that implements this idea for reinforcement learning agents in mathematical domains. LEMMA augments Expert Iteration with an abstraction step, where solutions found so far are revisited and rewritten in terms of new higher-level actions, which then become available to solve new problems. We evaluate LEMMA on two mathematical reasoning tasks-equation solving and fraction simplification-in a step-by-step fashion. In these two domains, LEMMA improves the ability of an existing agent, both solving more problems and generalizing more effectively to harder problems than those seen during training.

For agents

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

get_harvested_code_for_paper("2211.08671")
get_code_for_paper("2211.08671")
have("2211.08671")

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

curl https://syntology.ai/api/ran/2211.08671.json

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

[![Syntology run record](https://syntology.ai/api/ran/2211.08671.svg)](https://syntology.ai/paper/2211.08671)

Connect an agent — have() is free.