Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
config: string
run: string
prefix_steps: int64
commit_rate_cap50: double
commit_rate_cap100: double
toolcap_hit_cap50: double
toolcap_hit_cap100: double
tool_calls_cap50: double
tool_calls_cap100: double
SR_prefix_cap50: double
SR_prefix_cap100: double
per_run: list<item: struct<harness: string, backbone: string, run: int64, n_cases: int64, n_scored: int64, in (... 101 chars omitted)
child 0, item: struct<harness: string, backbone: string, run: int64, n_cases: int64, n_scored: int64, incomplete: i (... 89 chars omitted)
child 0, harness: string
child 1, backbone: string
child 2, run: int64
child 3, n_cases: int64
child 4, n_scored: int64
child 5, incomplete: int64
child 6, final: bool
child 7, ACC: double
child 8, SER: double
child 9, EER: double
child 10, CER: double
child 11, limit_hits: int64
across_runs: list<item: struct<harness: string, backbone: string, runs_final: int64, runs_seen: int64, ACC_mean: (... 154 chars omitted)
child 0, item: struct<harness: string, backbone: string, runs_final: int64, runs_seen: int64, ACC_mean: double, ACC (... 142 chars omitted)
child 0, harness: string
child 1, backbone: string
child 2, runs_final: int64
child 3, runs_seen: int64
child 4, ACC_mean: double
child 5, ACC_std: double
child 6, SER_mean: double
child 7, SER_std: double
child 8, EER_mean: double
child 9, EER_std: double
child 10, CER_mean: double
child 11, CER_std: double
child 12, case_agreement: double
to
{'per_run': List({'harness': Value('string'), 'backbone': Value('string'), 'run': Value('int64'), 'n_cases': Value('int64'), 'n_scored': Value('int64'), 'incomplete': Value('int64'), 'final': Value('bool'), 'ACC': Value('float64'), 'SER': Value('float64'), 'EER': Value('float64'), 'CER': Value('float64'), 'limit_hits': Value('int64')}), 'across_runs': List({'harness': Value('string'), 'backbone': Value('string'), 'runs_final': Value('int64'), 'runs_seen': Value('int64'), 'ACC_mean': Value('float64'), 'ACC_std': Value('float64'), 'SER_mean': Value('float64'), 'SER_std': Value('float64'), 'EER_mean': Value('float64'), 'EER_std': Value('float64'), 'CER_mean': Value('float64'), 'CER_std': Value('float64'), 'case_agreement': Value('float64')})}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
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
config: string
run: string
prefix_steps: int64
commit_rate_cap50: double
commit_rate_cap100: double
toolcap_hit_cap50: double
toolcap_hit_cap100: double
tool_calls_cap50: double
tool_calls_cap100: double
SR_prefix_cap50: double
SR_prefix_cap100: double
per_run: list<item: struct<harness: string, backbone: string, run: int64, n_cases: int64, n_scored: int64, in (... 101 chars omitted)
child 0, item: struct<harness: string, backbone: string, run: int64, n_cases: int64, n_scored: int64, incomplete: i (... 89 chars omitted)
child 0, harness: string
child 1, backbone: string
child 2, run: int64
child 3, n_cases: int64
child 4, n_scored: int64
child 5, incomplete: int64
child 6, final: bool
child 7, ACC: double
child 8, SER: double
child 9, EER: double
child 10, CER: double
child 11, limit_hits: int64
across_runs: list<item: struct<harness: string, backbone: string, runs_final: int64, runs_seen: int64, ACC_mean: (... 154 chars omitted)
child 0, item: struct<harness: string, backbone: string, runs_final: int64, runs_seen: int64, ACC_mean: double, ACC (... 142 chars omitted)
child 0, harness: string
child 1, backbone: string
child 2, runs_final: int64
child 3, runs_seen: int64
child 4, ACC_mean: double
child 5, ACC_std: double
child 6, SER_mean: double
child 7, SER_std: double
child 8, EER_mean: double
child 9, EER_std: double
child 10, CER_mean: double
child 11, CER_std: double
child 12, case_agreement: double
to
{'per_run': List({'harness': Value('string'), 'backbone': Value('string'), 'run': Value('int64'), 'n_cases': Value('int64'), 'n_scored': Value('int64'), 'incomplete': Value('int64'), 'final': Value('bool'), 'ACC': Value('float64'), 'SER': Value('float64'), 'EER': Value('float64'), 'CER': Value('float64'), 'limit_hits': Value('int64')}), 'across_runs': List({'harness': Value('string'), 'backbone': Value('string'), 'runs_final': Value('int64'), 'runs_seen': Value('int64'), 'ACC_mean': Value('float64'), 'ACC_std': Value('float64'), 'SER_mean': Value('float64'), 'SER_std': Value('float64'), 'EER_mean': Value('float64'), 'EER_std': Value('float64'), 'CER_mean': Value('float64'), 'CER_std': Value('float64'), 'case_agreement': Value('float64')})}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Herculean benchmark runs
Raw run store of the Herculean agentic financial benchmark experiments, mirrored while the experiments run (updated every ~15 minutes; numbers are partial until every configuration has three complete runs).
- Workflows:
auditing(XBRL filing audit, 67 cases),trading(4 stocks x 61 trading days, Jan-Mar 2026),hedging(pair hedging over 8 stocks, 61 days),insights(weekly market-insights reports, 4 stocks x 13 weeks). - Harnesses: ReAct, Claude Code, Codex, Hermes, OpenClaw. Every harness talks to the same MCP exam server.
- Backbones: Claude Sonnet 5, GPT-5.4, DeepSeek-V3.2, DeepSeek-V4-Pro (API), Qwen3.5-9B and Qwen3.5-4B (self-hosted vLLM).
- Three independent runs per configuration.
Layout
<experiment>/<workflow>/<harness>/<backbone>/run<r>/<case>/step_<k>/
| file | content |
|---|---|
task.json |
the assignment the agent received from get_task |
commit.json |
the agent's write-once submission |
status.json |
outcome, limits hit, attempts, token usage, estimated cost |
manifest.json (per run) |
harness/backbone/limits/route provenance |
ledger.jsonl |
budget ledger (estimated list-price cost) |
Statuses: committed (submitted), no_commit (finished or hit a turn/tool/token limit without submitting),
timeout, harness_error / skipped_* (infrastructure; re-run, never scored).
post/ holds the post-release window runs (Aug-Sep 2026; environment and instances in
TheFinAI/Herculean-post).
ablation_toolcap50/ holds hedging runs that hit the earlier 50-tool-call cap (re-run in main with 100).
bug_codex_qwen_reasoning/ holds Codex+Qwen runs affected by a bridge bug (Qwen's reasoning was dropped); re-run in main.
Full traces are added when the experiments finish.
- Downloads last month
- 295