Dataset Preview
Duplicate
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:    TypeError
Message:      'str' object is not a mapping
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 283, in _generate_tables
                  examples = [json_encode_field(example, json_field_path) for example in examples]
                              ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 34, in json_encode_field
                  return {**example, field: json_encode_field(example.get(field), json_field_path)}
                                            ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 32, in json_encode_field
                  return [json_encode_field(x, json_field_path) for x in example]
                          ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 34, in json_encode_field
                  return {**example, field: json_encode_field(example.get(field), json_field_path)}
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
              TypeError: 'str' object is not a mapping
              
              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 1880, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

id
int64
prompt
string
language
string
target_word_count
int64
word_count
int64
method
string
complete
bool
outline
list
story
list
final_state
dict
1,114
"Write a new tale featuring a clever pawnbroker in a floating sky bazaar who strikes deals with dari(...TRUNCATED)
en
20,000
22,092
NstAgentAblation
true
[{"id":0,"name":"The Ledger of Lost Things","description":"Introduces the floating sky bazaar of Aer(...TRUNCATED)
[{"id":0,"name":"The Ledger of Lost Things","content":"Aerovane did not float so much as it hung, a (...TRUNCATED)
{"character_states":[{"id":1,"name":"Elian Voss","description":"Trusted pawnbroker and appraiser, pr(...TRUNCATED)
1,122
"Create an original story about someone who wakes up after the most incredible day of their life onl(...TRUNCATED)
en
20,000
23,191
NstAgentAblation
true
[{"id":0,"name":"The Last Morning of the Festival","description":"Nora, a 34-year-old landscape arch(...TRUNCATED)
[{"id":0,"name":"The Last Morning of the Festival","content":"The sign for Port Ember appeared out o(...TRUNCATED)
{"character_states":[{"id":1,"name":"Nora","description":"34-year-old former landscape architect, no(...TRUNCATED)
1,133
"Generate a narrative about two rival wielders of primal energies whose conflict threatens to unrave(...TRUNCATED)
en
20,000
23,527
NstAgentAblation
true
[{"id":0,"name":"The Weave and the Wound","description":"Aethra is introduced as a world woven from (...TRUNCATED)
[{"id":0,"name":"The Weave and the Wound","content":"Before there were gods, or names for gods, or c(...TRUNCATED)
{"character_states":[{"id":1,"name":"Aelith","description":"Aelith, the Radiant Mother of Lumen, one(...TRUNCATED)
1,140
"Imagine a story where two hitmen on the run crash together inside a remote mountain lodge. One is a(...TRUNCATED)
en
20,000
22,075
NstAgentAblation
true
[{"id":0,"name":"The Storm and the Lodge","description":"An avalanche of snow isolates a mountain pa(...TRUNCATED)
[{"id":0,"name":"The Storm and the Lodge","content":"The mountain did not take sides. It buried the (...TRUNCATED)
{"character_states":[{"id":1,"name":"Julian Mercer","description":"Perfectionist ex-detective turned(...TRUNCATED)
1,163
"Imagine a story where a fifteen-year-old gaming champ lands an epic killstreak only to discover his(...TRUNCATED)
en
20,000
23,081
NstAgentAblation
true
[{"id":0,"name":"King of the Grid","description":"In 2042, competitive gaming rules the world, and f(...TRUNCATED)
[{"id":0,"name":"King of the Grid","content":"The ceiling of Kade Tanaka's bedroom was a screen, and(...TRUNCATED)
{"character_states":[{"id":1,"name":"Kade 'Vex' Tanaka","description":"Fifteen-year-old former Globa(...TRUNCATED)
1,168
"Compose a story that explores the life of an interstellar agent whose mission is to nudge emerging (...TRUNCATED)
en
20,000
21,642
NstAgentAblation
true
[{"id":0,"name":"The Necessary Mathematics","description":"Introduces Vesper, the Concord's most con(...TRUNCATED)
[{"id":0,"name":"The Necessary Mathematics","content":"Vesper had been dreaming in arithmetic again.(...TRUNCATED)
{"character_states":[{"id":1,"name":"Vesper","description":"At last returned to Auva, lying on the s(...TRUNCATED)
1,169
"Write a new tale featuring a young recruit’s journey through his first and final days in the tren(...TRUNCATED)
en
20,000
23,997
NstAgentAblation
true
[{"id":0,"name":"The Schoolteacher's War","description":"Introduces Γ‰mile Charpentier, a 22-year-ol(...TRUNCATED)
[{"id":0,"name":"The Schoolteacher's War","content":"The news reached Saint-CΓ©lerin on a Sunday in (...TRUNCATED)
{"character_states":[{"id":1,"name":"Γ‰mile Charpentier","description":"26 years old, the schoolteac(...TRUNCATED)
1,183
"Imagine a story about a lone insectoid warrior who suddenly rejects the telepathic hive mind contro(...TRUNCATED)
en
20,000
21,846
NstAgentAblation
true
[{"id":0,"name":"The Geometry of Obedience","description":"On the toxic, storm-scoured world Cinderf(...TRUNCATED)
[{"id":0,"name":"The Geometry of Obedience","content":"The storm was a constant pressure at Krill's (...TRUNCATED)
{"character_states":[{"id":1,"name":"Krill","description":"Gone. Her body dissolved entirely inside (...TRUNCATED)
1,198
"Imagine a story where every student’s GPA glows as a number above their head on a busy college ca(...TRUNCATED)
en
20,000
22,035
NstAgentAblation
true
[{"id":0,"name":"The Number Above","description":"Meridian University glows with the numbers hoverin(...TRUNCATED)
[{"id":0,"name":"The Number Above","content":"The first thing Maya Chen noticed about Meridian Unive(...TRUNCATED)
{"character_states":[{"id":1,"name":"Maya Chen","description":"Sophomore transfer student, movement (...TRUNCATED)
1,201
"Imagine a story where the organs inside a recovering patient come to life as characters vying for c(...TRUNCATED)
en
20,000
22,572
NstAgentAblation
true
[{"id":0,"name":"The Operation's Echo","description":"In the aftermath of emergency cardiac surgery,(...TRUNCATED)
[{"id":0,"name":"The Operation's Echo","content":"The last sound the Patient knew was a voice saying(...TRUNCATED)
{"character_states":[{"id":1,"name":"Brain","description":"The command center inside the Patient's s(...TRUNCATED)
End of preview.

NstAgent

🌐 Github | πŸ“– Paper | πŸ€— Data

Introduction

NstAgent (Narrative State Tracking Agent) is a training-free agentic framework for scaling long-form story generation toward full-length novels.

This dataset contains all 2,030 stories evaluated in the paper, generated by NstAgent, its ablations and the baselines (RollSum, Direct, DOME, StoryWriter) with DeepSeek-V4-Flash and GPT-5.6 Luna, from 10K to 100K words. The prompts are from ConStory-Bench. The judge outputs and evaluation code are in the Github repository.

Data

You can download the data into the root of the Github repository:

huggingface-cli download zhennan1/NstAgent --repo-type dataset --local-dir . --include "results/stories/*"

The data is organized as follows:

./
└── results/
    └── stories/
        β”œβ”€β”€ deepseek/
        β”‚   β”œβ”€β”€ nstagent_{10k,20k,50k,100k}.jsonl
        β”‚   β”œβ”€β”€ rollsum_{10k,20k,50k,100k}.jsonl
        β”‚   β”œβ”€β”€ direct_10k.jsonl
        β”‚   β”œβ”€β”€ storywriter_10k.jsonl
        β”‚   β”œβ”€β”€ dome_10k_20prompts.jsonl
        β”‚   └── ablation_{no_state,no_lookback}_20k.jsonl
        └── luna/
            └── ...

Each line is one story with the following fields:

  • id: Prompt id in ConStory-Bench.
  • prompt: Story prompt.
  • outline: Frozen chapter outline shared by NstAgent and RollSum, null for Direct, DOME and StoryWriter.
  • story: List of chapters, or a single string for Direct, DOME and StoryWriter.
  • final_state: Final narrative state of NstAgent, including characters, past events and future requirements.
  • word_count, target_word_count, complete, language, method: Metadata.

For NstAgent, each chapter also contains tool_trace, the tool calls made while writing it, from which the state after any chapter can be rebuilt. At 100K words, each method is run on 50 of the 100 prompts. DOME is run on an earlier set of 20 prompts.

Citation

If you find this project helpful, please cite it as follows:

@article{wan2026nstagent,
  title={Scaling Long-Form Story Generation via Narrative State Tracking},
  author={Zhennan Wan and Jianfei Chen},
  year={2026}
}
Downloads last month
29