Datasets:

vibrantlabsai
/

usaspending-recon

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
reference: string
task_ids: list<item: string>
  child 0, item: string
fingerprints: struct<usr-l1-06: string, usr-l1-07: string, usr-l1-09: string, usr-l2-08: string, usr-l2-12: string (... 514 chars omitted)
  child 0, usr-l1-06: string
  child 1, usr-l1-07: string
  child 2, usr-l1-09: string
  child 3, usr-l2-08: string
  child 4, usr-l2-12: string
  child 5, usr-l2-13: string
  child 6, usr-l3-04: string
  child 7, usr-l3-10: string
  child 8, usr-l4-01: string
  child 9, usr-l4-02: string
  child 10, usr-l4-03: string
  child 11, usr-l4-04: string
  child 12, usr-l4-10: string
  child 13, usr-l4-12: string
  child 14, usr-l4-13: string
  child 15, usr-l5-01: string
  child 16, usr-l5-02: string
  child 17, usr-l5-03: string
  child 18, usr-l5-06: string
  child 19, usr-l5-07: string
  child 20, usr-l5-10: string
  child 21, usr-l5-12: string
  child 22, usr-l5-13: string
  child 23, usr-l6-01: string
  child 24, usr-l6-02: string
  child 25, usr-l6-03: string
  child 26, usr-l6-04: string
  child 27, usr-l6-06: string
  child 28, usr-l6-07: string
  child 29, usr-l6-08: string
  child 30, usr-l6-10: string
  child 31, usr-l6-11: string
tasks: list<item: struct<task_id: string, level: int64, base: string, reference: string, verdict: string>>
  child 0, item: struct<task_id: string, level: int64, base: string, reference: string, verdict: string>
      child 0, task_id: string
      child 1, level: int64
      child 2, base: string
      child 3, reference: string
      child 4, verdict: string
to
{'reference': Value('string'), 'tasks': List({'task_id': Value('string'), 'level': Value('int64'), 'base': Value('string'), 'reference': Value('string'), 'verdict': Value('string')})}
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
              reference: string
              task_ids: list<item: string>
                child 0, item: string
              fingerprints: struct<usr-l1-06: string, usr-l1-07: string, usr-l1-09: string, usr-l2-08: string, usr-l2-12: string (... 514 chars omitted)
                child 0, usr-l1-06: string
                child 1, usr-l1-07: string
                child 2, usr-l1-09: string
                child 3, usr-l2-08: string
                child 4, usr-l2-12: string
                child 5, usr-l2-13: string
                child 6, usr-l3-04: string
                child 7, usr-l3-10: string
                child 8, usr-l4-01: string
                child 9, usr-l4-02: string
                child 10, usr-l4-03: string
                child 11, usr-l4-04: string
                child 12, usr-l4-10: string
                child 13, usr-l4-12: string
                child 14, usr-l4-13: string
                child 15, usr-l5-01: string
                child 16, usr-l5-02: string
                child 17, usr-l5-03: string
                child 18, usr-l5-06: string
                child 19, usr-l5-07: string
                child 20, usr-l5-10: string
                child 21, usr-l5-12: string
                child 22, usr-l5-13: string
                child 23, usr-l6-01: string
                child 24, usr-l6-02: string
                child 25, usr-l6-03: string
                child 26, usr-l6-04: string
                child 27, usr-l6-06: string
                child 28, usr-l6-07: string
                child 29, usr-l6-08: string
                child 30, usr-l6-10: string
                child 31, usr-l6-11: string
              tasks: list<item: struct<task_id: string, level: int64, base: string, reference: string, verdict: string>>
                child 0, item: struct<task_id: string, level: int64, base: string, reference: string, verdict: string>
                    child 0, task_id: string
                    child 1, level: int64
                    child 2, base: string
                    child 3, reference: string
                    child 4, verdict: string
              to
              {'reference': Value('string'), 'tasks': List({'task_id': Value('string'), 'level': Value('int64'), 'base': Value('string'), 'reference': Value('string'), 'verdict': Value('string')})}
              because column names don't match

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.

USAspending contract reconciliation

An OpenEnv environment for the OpenEnv Arena. Each episode is one real U.S. EPA vendor. The model queries two tables with read-only SQL and reports every contract action whose finance-ledger postings do not sum to its obligated amount.

What it teaches

Multi-step reconciliation of two independent views of the same facts, the routine core of accounts-payable and contract-finance work: joining on keys that are formatted differently in each system (P00005 vs M5, 0 vs BASE), tolerating legitimate differences (split postings, posting lags, vendor-name variants, de-obligations recorded as positive DEOB entries), filtering out other vendors' records, and computing exact signed differences.

Raw source

contract_actions rows are real FY2024 EPA contract actions from the USAspending.gov award data archive (FY2024_068_Contracts_Full_20260906.zip, sha256 ce708e70…dc1), a U.S. federal work in the public domain (17 U.S.C. §105). Only awards whose actions reconcile exactly to the reported award total are used, so the raw data contributes no discrepancies of its own. The ledger is derived from the same actions; discrepancies (missing, duplicated, mis-keyed and sign-flipped postings) are planted, and the expected answer is recomputed from the final tables.

Reward

Binary. 1.0 when the submitted list of {award, mod, difference} equals the expected set exactly (formatting of keys and amounts is forgiven; differences within half a cent). 0.0 otherwise, including when the 30-call budget runs out.

Files

tasks.jsonl: one task per line (task_id, level, brief, source award keys)
tasks/*.sqlite: each task's database; answers.json: expected answers
build_tasks.py: the generator, deterministic from the pinned raw file and seed

Downloads last month: 124

Total file size:

1.49 MB