SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.16828 · 2024

Ragnarök: A Reusable RAG Framework and Baselines for TREC 2024 Retrieval-Augmented Generation Track

arXiv · PDF · Open in the Atlas

Code that ran

19 of 23 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; 1 honoured the contract we drafted, 1 violated it and 10 have no contract verdict; 7 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/2406.16828.json · badge: /api/ran/2406.16828.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
castorini/ragnarok canonical (classifier label)
classifier basis: census
9 ran · 3 no recorded run
castorini/nuggetizer community (classifier label)
classifier basis not recorded on the link
10 ran · 1 no recorded run
FunctionStatusWhere it lives
build_assign_nuggetizer_kwargs Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/execution.py
code served (permissive licence) · get_code("b072bca5935fc26d")
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.

from dataclasses import dataclass
from typing import Any, Literal

ExecutionMode = Literal['sync', 'async']

@dataclass(frozen=True)
class AssignExecutionConfig:
    model: str
    execution_mode: ExecutionMode = 'sync'
    log_level: int = 0
    use_azure_openai: bool = False
    use_openrouter: bool = False
    reasoning_effort: str | None = None
    include_trace: bool = False
    include_reasoning: bool = False
    redact_prompts: bool = False
    quiet: bool = False
    output: str = 'json'

def build_assign_nuggetizer_kwargs(config: AssignExecutionConfig) -> dict[str, Any]:
    return {'assigner_model': config.model, 'log_level': config.log_level, 'use_azure_openai': config.use_azure_openai, 'use_openrouter': config.use_openrouter, 'store_trace': config.include_trace, 'store_reasoning': config.include_reasoning, 'reasoning_effort': config.reasoning_effort}

config = AssignExecutionConfig(
    model='gpt-4',
    execution_mode='sync',
    log_level=1,
    use_azure_openai=False,
    use_openrouter=False,
    reasoning_effort=None,
    include_trace=False,
    include_reasoning=False,
    redact_prompts=False,
    quiet=False,
    output='json'
)

ARGS = {
    'config': config
}
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
build_create_nuggetizer_kwargs Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/execution.py
code served (permissive licence) · get_code("c775437a8c8fe07d")
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.

from dataclasses import dataclass
from typing import Literal

ExecutionMode = Literal['sync', 'async']

@dataclass(frozen=True)
class CreateExecutionConfig:
    model: str
    creator_model: str | None = None
    scorer_model: str | None = None
    window_size: int | None = None
    max_nuggets: int | None = None
    min_judgment: int = 2
    execution_mode: ExecutionMode = 'sync'
    log_level: int = 0
    use_azure_openai: bool = False
    use_openrouter: bool = False
    reasoning_effort: str | None = None
    include_trace: bool = False
    include_reasoning: bool = False
    redact_prompts: bool = False
    quiet: bool = False
    output: str = 'json'

# Test case 1: minimal config with only model specified
config_minimal = CreateExecutionConfig(
    model='gpt-4'
)

# Test case 2: config with creator and scorer models
config_with_models = CreateExecutionConfig(
    model='gpt-4',
    creator_model='gpt-3.5-turbo',
    scorer_model='gpt-4',
    window_size=512,
    max_nuggets=10,
    log_level=1,
    use_azure_openai=True,
    use_openrouter=False,
    reasoning_effort='high',
    include_trace=True,
    include_reasoning=True
)

# Test case 3: config without creator/scorer but with window and max_nuggets
config_with_params = CreateExecutionConfig(
    model='claude-3-opus',
    window_size=256,
    max_nuggets=5,
    log_level=2
)

ARGS = {
    'config': config_minimal
}
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
collect_nonempty_reasoning_traces Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/adapters.py
code served (permissive licence) · get_code("53b36b0cccd27107")
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.

from collections.abc import Sequence

ARGS = {
    "traces": [
        "reasoning step 1",
        None,
        "  reasoning step 2  ",
        "",
        "reasoning step 3",
        None,
        "   ",
    ]
}
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
compute_elo Ran
ran, honoured the contract we drafted
castorini/ragnarok/src/ragnarok/api/elo.py
code served (permissive licence) · get_code("d76974dd615c64e1")
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, 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

# The function compute_elo takes three parameters:
# - sa: int (score for player a, appears to be 0 or 1 for loss/win)
# - ra: int (rating for player a, defaults to INIT_RATING=1000)
# - rb: int (rating for player b, defaults to INIT_RATING=1000)
# 
# The function body only does arithmetic operations on these integers,
# so we just need valid int values.

ARGS = {
    "sa": 1,           # Player A's score (0 for loss, 1 for win)
    "ra": 1000,        # Player A's current rating
    "rb": 1200,        # Player B's current rating
}
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, ndim 1, dtype family float — a tuple 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
detect_command Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/main_support.py
code served (permissive licence) · get_code("145f2c77b2912062")
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 str. 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
from collections.abc import Sequence

torch.manual_seed(0)

# The function detect_command takes argv: Sequence[str]
# It iterates through tokens in argv and returns the first one found in KNOWN_COMMANDS
# If none found, returns 'unknown'

# Test case 1: argv containing a known command
argv_with_command = ['--help', 'create', '--verbose']

# Test case 2: argv with no known commands
argv_without_command = ['--help', '--verbose', '--output', 'file.txt']

# Test case 3: argv with command at the start
argv_command_first = ['serve', '--port', '8080']

# Test case 4: empty argv
argv_empty = []

# Test case 5: argv with only known command
argv_only_command = ['metrics']

# Use one of these as the main test case
ARGS = {
    'argv': argv_with_command
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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_run_id Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/io.py
code served (permissive licence) · get_code("8fa7a665d31a83dc")
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 str. 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
from pathlib import Path

# The function extracts a run identifier from a file path using Path.stem
# which returns the filename without extension

ARGS = {
    "file_path": "/some/path/to/experiment_run_001.txt"
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
make_data_artifact Ran
ran; our contract draft was wrong, not the code
castorini/ragnarok/src/ragnarok/cli/adapters.py
code served (permissive licence) · get_code("dd91db53416878f5")
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)

# The function make_data_artifact simply takes a name and data,
# and returns a dict with those values plus a 'kind' field.
# No special attributes or methods are called on the data object,
# so we can pass any type of data.

# Create diverse data examples
tensor_data = torch.randn(2, 4)
numpy_data = np.array([[1, 2, 3], [4, 5, 6]])
list_data = [1, 2, 3, 4, 5]
dict_data = {'key1': 'value1', 'key2': 'value2'}

ARGS = {
    'name': 'my_dataset',
    'data': tensor_data
}
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
make_file_artifact Ran
ran; our contract draft was wrong, not the code
castorini/ragnarok/src/ragnarok/cli/adapters.py
code served (permissive licence) · get_code("7544006c2fb2eb8d")
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
from pathlib import Path

ARGS = {
    'name': 'test_artifact',
    'path': '/tmp/test_file.txt'
}
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
prepare_output_path Ran
ran; our contract draft was wrong, not the code
castorini/ragnarok/src/ragnarok/cli/dispatch.py
code served (permissive licence) · get_code("4a89617557cd8d60")
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 str. 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 argparse
from pathlib import Path
import tempfile
import os

# Define a minimal CLIError class since it's referenced in the function
class CLIError(Exception):
    def __init__(self, message, exit_code=1, status=None, error_code=None, command=None, details=None):
        self.message = message
        self.exit_code = exit_code
        self.status = status
        self.error_code = error_code
        self.command = command
        self.details = details
        super().__init__(message)

# Define EXIT_CODES dict since it's referenced
EXIT_CODES = {
    'invalid_arguments': 1,
    'validation_error': 2,
}

# Create a temporary directory for output
temp_dir = tempfile.mkdtemp()
output_file_path = os.path.join(temp_dir, "output.txt")

# Create args namespace with output_file and no resume/overwrite/fail_if_exists flags
args = argparse.Namespace(
    output_file=output_file_path,
    resume=False,
    overwrite=False,
    fail_if_exists=False
)

ARGS = {
    'args': args,
    'command': 'test_command'
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
read_direct_payload Ran
ran; our contract draft was wrong, not the code
castorini/ragnarok/src/ragnarok/cli/dispatch.py
code served (permissive licence) · get_code("f3bfd9301b14fe26")
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 argparse
import json
import sys
from io import StringIO
from typing import Any, cast

# Define the CLIError class that the function references
class CLIError(Exception):
    def __init__(self, message, exit_code=1, status='error', error_code='unknown', command='test', details=None):
        self.message = message
        self.exit_code = exit_code
        self.status = status
        self.error_code = error_code
        self.command = command
        self.details = details or {}
        super().__init__(self.message)

# Define EXIT_CODES dict that the function references
EXIT_CODES = {
    'invalid_arguments': 2,
    'success': 0,
}

# Create a valid args namespace with stdin=True and proper JSON input
namespace = argparse.Namespace(
    stdin=True,
    input_json=None,
    command='test_command'
)

# Set up stdin with valid JSON payload
test_payload = {'key': 'value', 'nested': {'data': 123}}
original_stdin = sys.stdin
sys.stdin = StringIO(json.dumps(test_payload))

ARGS = {
    'args': namespace
}
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
read_jsonl Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/io.py
code served (permissive licence) · get_code("719779e74d1ac660")
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 for testing
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test.jsonl")

# Write some test data to the JSONL file
test_data = [
    {"id": 1, "name": "Alice", "value": 42},
    {"id": 2, "name": "Bob", "value": 3.14},
    {"id": 3, "name": "Charlie", "value": None}
]

with open(temp_file, 'w', encoding='utf-8') as f:
    for item in test_data:
        f.write(json.dumps(item) + '\n')

# Define the arguments for read_jsonl
ARGS = {
    "file_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
read_results_from_file Ran
ran; our contract draft was wrong, not the code
castorini/ragnarok/src/ragnarok/data.py
code served (permissive licence) · get_code("e653a76dcbab5bab")
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 JSON file with valid data structure
temp_dir = tempfile.mkdtemp()
json_file_path = os.path.join(temp_dir, "results.json")

# Create sample data matching the expected structure
sample_data = [
    {
        "topic": "What is machine learning?",
        "topic_id": 1,
        "references": ["doc1", "doc2"],
        "answer": [
            {
                "text": "Machine learning is a subset of AI.",
                "citations": [0]
            },
            {
                "text": "It enables systems to learn from data.",
                "citations": [1]
            }
        ]
    },
    {
        "topic": "How does deep learning work?",
        "topic_id": 2,
        "references": ["doc3", "doc4"],
        "answer": [
            {
                "text": "Deep learning uses neural networks.",
                "citations": [0, 1]
            }
        ]
    }
]

# Write the sample data to the temp JSON file
with open(json_file_path, 'w', encoding='utf-8') as f:
    json.dump(sample_data, f)

# Define ARGS dictionary with the file_path parameter
ARGS = {
    "file_path": json_file_path
}
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
resolve_write_policy Ran
ran; our contract draft was wrong, not the code
castorini/ragnarok/src/ragnarok/cli/dispatch.py
code served (permissive licence) · get_code("33cd494e3b5331cf")
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 str. 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 argparse

ARGS = {
    'args': argparse.Namespace(
        resume=False,
        overwrite=False,
        fail_if_exists=False
    )
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
result_to_dict Ran
ran; our contract draft was wrong, not the code
castorini/ragnarok/src/ragnarok/data.py
code served (permissive licence) · get_code("4a11e6b65475e860")
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
from dataclasses import dataclass, field
from typing import Any

@dataclass
class Query:
    text: str
    qid: str | int

@dataclass
class Candidate:
    docid: str | int
    score: float
    doc: dict[str, Any]

@dataclass
class CitedSentence:
    text: str
    citations: list[int] = field(default_factory=list)

@dataclass
class RAGExecInfo:
    prompt: Any
    response: Any
    input_token_count: int
    output_token_count: int
    reasoning: str | None = None
    candidates: list[Candidate] = field(default_factory=list)

@dataclass
class Result:
    query: Query
    references: list[str | int] = field(default_factory=list)
    answer: list[CitedSentence] = field(default_factory=list)
    rag_exec_summary: RAGExecInfo | None = None

# Construct minimal valid arguments
query = Query(text="What is machine learning?", qid=42)

answer_sentences = [
    CitedSentence(text="Machine learning is a field of AI.", citations=[0, 1]),
    CitedSentence(text="It involves training models on data.", citations=[1, 2]),
]

rag_exec_info = RAGExecInfo(
    prompt="Answer the question: What is machine learning?",
    response="Machine learning is a field of AI. It involves training models on data.",
    input_token_count=15,
    output_token_count=25,
    reasoning="Retrieved relevant documents about ML concepts",
    candidates=[
        Candidate(docid="doc1", score=0.95, doc={"title": "ML Basics", "text": "Introduction to ML"}),
        Candidate(docid="doc2", score=0.87, doc={"title": "ML Advanced", "text": "Advanced ML techniques"}),
    ]
)

result = Result(
    query=query,
    references=["doc1", "doc2", "doc3"],
    answer=answer_sentences,
    rag_exec_summary=rag_exec_info
)

ARGS = {
    "result": result,
    "run_id": "run_001",
    "include_trace": True,
    "redact_prompts": False,
}
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
setup_logging Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/logging_utils.py
code served (permissive licence) · get_code("b74163bc3735c7ee")
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 Logger. 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 logging

ARGS = {
    'log_level': 1,
    'quiet': False,
}
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 Logger — a Logger output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Logger 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
unwrap_direct_create_payload Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/normalize.py
code served (permissive licence) · get_code("074a68a9b4bcba9f")
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
from typing import Any

# Create a valid payload that matches the schema_version and structure expected
# The function looks for artifacts with 'query' and 'candidates' keys, or a 'judgments' artifact

payload = {
    'schema_version': 'castorini.cli.v1',
    'artifacts': [
        {
            'name': 'search_results',
            'data': {
                'query': 'what is machine learning',
                'candidates': [
                    {'id': '1', 'text': 'Machine learning is a subset of AI'},
                    {'id': '2', 'text': 'ML involves training models on data'}
                ]
            }
        }
    ]
}

ARGS = {
    'payload': payload
}
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
unwrap_generation_record Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/normalize.py
code served (permissive licence) · get_code("74873c74cd75260c")
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.

# Construct valid arguments for unwrap_generation_record

ARGS = {
    'payload': {
        'schema_version': 'castorini.cli.v1',
        'artifacts': [
            {
                'name': 'generation-results',
                'data': {
                    'query': 'test query',
                    'generated_text': 'test output',
                    'score': 0.95
                }
            }
        ]
    }
}
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
unwrap_nugget_record Ran
ran on a synthesized input
castorini/nuggetizer/src/nuggetizer/cli/normalize.py
code served (permissive licence) · get_code("fd5baa7869c054bc")
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

# Build a valid payload that satisfies unwrap_nugget_record
# The function expects a payload dict with:
# - schema_version: 'castorini.cli.v1'
# - artifacts: list of dicts where one has name='create-result'
# The artifact's data/value should be a dict or a list with exactly one dict

payload = {
    'schema_version': 'castorini.cli.v1',
    'artifacts': [
        {
            'name': 'create-result',
            'data': {
                'id': 'test-id',
                'text': 'test nugget content',
                'metadata': {'key': 'value'}
            }
        }
    ]
}

ARGS = {
    'payload': payload
}
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
wants_json Ran
ran, violated the contract we drafted
castorini/ragnarok/src/ragnarok/cli/errors.py
code served (permissive licence) · get_code("c53a0ca89658e42c")
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.

from collections.abc import Sequence

def wants_json(argv: Sequence[str]) -> bool:
    for index, token in enumerate(argv):
        if token == '--output' and index + 1 < len(argv):
            return argv[index + 1] == 'json'
    return False

# Test case 1: argv with --output json
argv_with_json = ['program', '--output', 'json', '--verbose']

ARGS = {
    'argv': argv_with_json
}
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 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
detect_command Not yet run
unverified (harvested, no recorded run)
castorini/ragnarok/src/ragnarok/cli/errors.py
code served (permissive licence) · get_code("258e9e13fcbf5b84")
How it ran

no invocation recorded.

get_processed_values Not yet run
unverified (harvested, no recorded run)
castorini/nuggetizer/src/nuggetizer/cli/io.py
code served (permissive licence) · get_code("7f2886141fa690bf")
How it ran

no invocation recorded.

validate_generate_batch_file Not yet run
unverified (harvested, no recorded run)
castorini/ragnarok/src/ragnarok/cli/introspection.py
code served (permissive licence) · get_code("2016f3275f584d17")
How it ran

no invocation recorded.

validate_generate_payload Not yet run
unverified (harvested, no recorded run)
castorini/ragnarok/src/ragnarok/cli/introspection.py
code served (permissive licence) · get_code("03080e064ef748c8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 23 samples lifted from this paper's linked repositories, 23 carry a function identity and 0 do not. Of the 22 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 22 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

Did you try out the new Bing Search? Or maybe you fiddled around with Google AI~Overviews? These might sound familiar because the modern-day search stack has recently evolved to include retrieval-augmented generation (RAG) systems. They allow searching and incorporating real-time data into large language models (LLMs) to provide a well-informed, attributed, concise summary in contrast to the traditional search paradigm that relies on displaying a ranked list of documents. Therefore, given these recent advancements, it is crucial to have an arena to build, test, visualize, and systematically evaluate RAG-based search systems. With this in mind, we propose the TREC 2024 RAG Track to foster innovation in evaluating RAG systems. In our work, we lay out the steps we've made towards making this track a reality -- we describe the details of our reusable framework, Ragnarök, explain the curation of the new MS MARCO V2.1 collection choice, release the development topics for the track, and standardize the I/O definitions which assist the end user. Next, using Ragnarök, we identify and provide key industrial baselines such as OpenAI's GPT-4o or Cohere's Command R+. Further, we introduce a web-based user interface for an interactive arena allowing benchmarking pairwise RAG systems by crowdsourcing. We open-source our Ragnarök framework and baselines to achieve a unified standard for future RAG systems.

For agents

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

get_harvested_code_for_paper("2406.16828")
get_code_for_paper("2406.16828")
have("2406.16828")

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

curl https://syntology.ai/api/ran/2406.16828.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.16828.svg)](https://syntology.ai/paper/2406.16828)

Connect an agent — have() is free.