Forecast-Dojo: Replayable Environments for Benchmarking and Training LLM Forecasting Agents
Paper • 2609.28876 • Published
Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: CastError
Message: Couldn't cast
schemaVersion: int64
generatedAt: string
questions: int64
forecastRows: int64
notebookRows: int64
toolRows: int64
snapshot: string
sourceArchiveSha256: string
scoringPolicy: string
paths: struct<trajectories: string, details: string>
child 0, trajectories: string
child 1, details: string
source: struct<dataset: string, mirroredFrom: string, path: string>
child 0, dataset: string
child 1, mirroredFrom: string
child 2, path: string
tools: list<item: struct<runId: string, modelName: string, mode: string, retrieval: string, rolloutIndex: i (... 197 chars omitted)
child 0, item: struct<runId: string, modelName: string, mode: string, retrieval: string, rolloutIndex: int64, stepI (... 185 chars omitted)
child 0, runId: string
child 1, modelName: string
child 2, mode: string
child 3, retrieval: string
child 4, rolloutIndex: int64
child 5, stepIndex: int64
child 6, forecastDate: timestamp[s]
child 7, toolName: string
child 8, canonicalTool: string
child 9, known: bool
child 10, calls: int64
child 11, successes: int64
child 12, errors: int64
child 13, parseErrors: int64
child 14, latencySeconds: double
eventId: string
notebooks: list<item: struct<runId: string, modelName: string, mode: string, retrieval: string, rolloutIndex: i (... 141 chars omitted)
child 0, item: struct<runId: string, modelName: string, mode: string, retrieval: string, rolloutIndex: int64, stepI (... 129 chars omitted)
child 0, runId: string
child 1, modelName: string
child 2, mode: string
child 3, retrieval: string
child 4, rolloutIndex: int64
child 5, stepIndex: int64
child 6, forecastDate: timestamp[s]
child 7, notebook: string
child 8, characters: int64
child 9, blockCount: int64
child 10, formatOk: bool
child 11, jsonValid: bool
to
{'eventId': Value('string'), 'notebooks': List({'runId': Value('string'), 'modelName': Value('string'), 'mode': Value('string'), 'retrieval': Value('string'), 'rolloutIndex': Value('int64'), 'stepIndex': Value('int64'), 'forecastDate': Value('timestamp[s]'), 'notebook': Value('string'), 'characters': Value('int64'), 'blockCount': Value('int64'), 'formatOk': Value('bool'), 'jsonValid': Value('bool')}), 'tools': List({'runId': Value('string'), 'modelName': Value('string'), 'mode': Value('string'), 'retrieval': Value('string'), 'rolloutIndex': Value('int64'), 'stepIndex': Value('int64'), 'forecastDate': Value('timestamp[s]'), 'toolName': Value('string'), 'canonicalTool': Value('string'), 'known': Value('bool'), 'calls': Value('int64'), 'successes': Value('int64'), 'errors': Value('int64'), 'parseErrors': Value('int64'), 'latencySeconds': Value('float64')})}
because column names don't match
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1827, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
schemaVersion: int64
generatedAt: string
questions: int64
forecastRows: int64
notebookRows: int64
toolRows: int64
snapshot: string
sourceArchiveSha256: string
scoringPolicy: string
paths: struct<trajectories: string, details: string>
child 0, trajectories: string
child 1, details: string
source: struct<dataset: string, mirroredFrom: string, path: string>
child 0, dataset: string
child 1, mirroredFrom: string
child 2, path: string
tools: list<item: struct<runId: string, modelName: string, mode: string, retrieval: string, rolloutIndex: i (... 197 chars omitted)
child 0, item: struct<runId: string, modelName: string, mode: string, retrieval: string, rolloutIndex: int64, stepI (... 185 chars omitted)
child 0, runId: string
child 1, modelName: string
child 2, mode: string
child 3, retrieval: string
child 4, rolloutIndex: int64
child 5, stepIndex: int64
child 6, forecastDate: timestamp[s]
child 7, toolName: string
child 8, canonicalTool: string
child 9, known: bool
child 10, calls: int64
child 11, successes: int64
child 12, errors: int64
child 13, parseErrors: int64
child 14, latencySeconds: double
eventId: string
notebooks: list<item: struct<runId: string, modelName: string, mode: string, retrieval: string, rolloutIndex: i (... 141 chars omitted)
child 0, item: struct<runId: string, modelName: string, mode: string, retrieval: string, rolloutIndex: int64, stepI (... 129 chars omitted)
child 0, runId: string
child 1, modelName: string
child 2, mode: string
child 3, retrieval: string
child 4, rolloutIndex: int64
child 5, stepIndex: int64
child 6, forecastDate: timestamp[s]
child 7, notebook: string
child 8, characters: int64
child 9, blockCount: int64
child 10, formatOk: bool
child 11, jsonValid: bool
to
{'eventId': Value('string'), 'notebooks': List({'runId': Value('string'), 'modelName': Value('string'), 'mode': Value('string'), 'retrieval': Value('string'), 'rolloutIndex': Value('int64'), 'stepIndex': Value('int64'), 'forecastDate': Value('timestamp[s]'), 'notebook': Value('string'), 'characters': Value('int64'), 'blockCount': Value('int64'), 'formatOk': Value('bool'), 'jsonValid': Value('bool')}), 'tools': List({'runId': Value('string'), 'modelName': Value('string'), 'mode': Value('string'), 'retrieval': Value('string'), 'rolloutIndex': Value('int64'), 'stepIndex': Value('int64'), 'forecastDate': Value('timestamp[s]'), 'toolName': Value('string'), 'canonicalTool': Value('string'), 'known': Value('bool'), 'calls': Value('int64'), 'successes': Value('int64'), 'errors': Value('int64'), 'parseErrors': Value('int64'), 'latencySeconds': Value('float64')})}
because column names don't match
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1879, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
eventId string | notebooks list | tools list |
|---|---|---|
238391 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
238398 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
238406 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
238417 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
238642 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
240053 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
240066 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
240653 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
240668 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
240748 | [{"runId":"deepseek-v3.2-think-code.forecast_eval.sequential","modelName":"deepseek-v3.2-think-code"(...TRUNCATED) | [{"runId":"deepseek-v3.2-think-code.forecast_eval.independent","modelName":"deepseek-v3.2-think-code(...TRUNCATED) |
End of preview.
Forecast-Dojo Trajectories
Per-question agent trajectories and research records behind the interactive site forecast-dojo.github.io. The site loads these files on demand when a question is opened. The benchmark data itself (questions, splits, dated news corpus pointers) lives in the companion dataset liqinye/Forecast-Dojo.
Layout
paper-v1-cd98a865/
manifest.json snapshot metadata and scoring policy
README.md notes on the generated files
trajectories/{event_id}.json one file per evaluation question: every model's probability path across forecast dates
details/{event_id}.json full notebook entries and tool-call records for the same question
paper-v1-cd98a865 is the snapshot used for the paper's results: 230 evaluation questions, 12 models, every recorded
forecast step, scored with the paper's v1 policy (exact labels, clipped and normalized probabilities, uniform fallback
for unusable forecasts, missing records excluded). Future snapshots will be added as sibling folders.
Paper
Forecast-Dojo: Replayable Environments for Benchmarking and Training LLM Forecasting Agents (2026). arXiv: 2609.28876.
@misc{ye2026forecastdojo,
title={Forecast-Dojo: Replayable Environments for Benchmarking and Training LLM Forecasting Agents},
author={Liqin Ye and Haorui Wang and Fardin Ahmed and Rongzhi Zhang and Yuan He and Ziyuan Lin and
Yanbin Yin and Jing Peng and Michael Galarnyk and Sudheer Chava and Chao Zhang},
year={2026},
eprint={2609.28876},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2609.28876},
}
- Downloads last month
- 190