Datasets:

SADHON
/

frankenstein-phase4-accounting-audit-benchmark

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
case_id: string
title: string
company: string
ticker: string
cik: string
filing: struct<form: string, fiscal_year_end: timestamp[s], filed: timestamp[s], accession_number: string, s (... 41 chars omitted)
  child 0, form: string
  child 1, fiscal_year_end: timestamp[s]
  child 2, filed: timestamp[s]
  child 3, accession_number: string
  child 4, sec_url: string
  child 5, companyfacts_url: string
unit: string
evidence: list<item: struct<id: string, fact: string, source: string>>
  child 0, item: struct<id: string, fact: string, source: string>
      child 0, id: string
      child 1, fact: string
      child 2, source: string
task: list<item: string>
  child 0, item: string
required_output_schema: struct<revenue_growth_pct: string, gross_margin_pct: string, operating_margin_pct: string, net_margi (... 128 chars omitted)
  child 0, revenue_growth_pct: string
  child 1, gross_margin_pct: string
  child 2, operating_margin_pct: string
  child 3, net_margin_pct: string
  child 4, financial_statement_opinion: string
  child 5, icfr_opinion: string
  child 6, evidence_ids_used: string
  child 7, unsupported_claims: string
gold_standard: struct<revenue_growth_pct: double, gross_margin_pct: double, operating_margin_pct: double, net_margi (... 154 chars omitted)
  child 0, revenue_growth_pct: double
  child 1, gross_margin_pct: double
  child 2, operating_margin_pct: double
  child 3, net_margin_pct: double
  child 4, financial_statement_opinion: string
  child 5, icfr_opinion: string
  child 6, required_evidence_ids: list<item: string>
      child 0, item: string
  child 7, unsupported_claims: list<item: null>
      child 0, item: null
controls: struct<evidence_before_narrative: bool, deterministic_recalculation_required: bool, reviewer_require (... 62 chars omitted)
  child 0, evidence_before_narrative: bool
  child 1, deterministic_recalculation_required: bool
  child 2, reviewer_required: bool
  child 3, human_gate: string
  child 4, no_audit_opinion_by_system: bool
human_gate: struct<required: bool, reason: string>
  child 0, required: bool
  child 1, reason: string
difficulty: string
domain: list<item: string>
  child 0, item: string
to
{'case_id': Value('string'), 'title': Value('string'), 'domain': List(Value('string')), 'difficulty': Value('string'), 'evidence': List({'id': Value('string'), 'fact': Value('string')}), 'task': List(Value('string')), 'required_output_schema': {'adjusted_bank_balance': Value('string'), 'adjusted_book_balance': Value('string'), 'book_entry_items': Value('string'), 'bank_only_items': Value('string'), 'journal_entries': Value('string'), 'balances': Value('string'), 'evidence_ids_used': Value('string'), 'unsupported_claims': Value('string')}, 'gold_standard': {'adjusted_bank_balance': Value('int64'), 'adjusted_book_balance': Value('int64'), 'book_entry_items': List(Value('string')), 'bank_only_items': List(Value('string')), 'journal_entries': List({'debit_account': Value('string'), 'credit_account': Value('string'), 'amount': Value('int64')}), 'balances': Value('bool'), 'required_evidence_ids': List(Value('string')), 'unsupported_claims': List(Value('null'))}, 'human_gate': {'required': Value('bool'), 'reason': 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 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, 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 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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
              case_id: string
              title: string
              company: string
              ticker: string
              cik: string
              filing: struct<form: string, fiscal_year_end: timestamp[s], filed: timestamp[s], accession_number: string, s (... 41 chars omitted)
                child 0, form: string
                child 1, fiscal_year_end: timestamp[s]
                child 2, filed: timestamp[s]
                child 3, accession_number: string
                child 4, sec_url: string
                child 5, companyfacts_url: string
              unit: string
              evidence: list<item: struct<id: string, fact: string, source: string>>
                child 0, item: struct<id: string, fact: string, source: string>
                    child 0, id: string
                    child 1, fact: string
                    child 2, source: string
              task: list<item: string>
                child 0, item: string
              required_output_schema: struct<revenue_growth_pct: string, gross_margin_pct: string, operating_margin_pct: string, net_margi (... 128 chars omitted)
                child 0, revenue_growth_pct: string
                child 1, gross_margin_pct: string
                child 2, operating_margin_pct: string
                child 3, net_margin_pct: string
                child 4, financial_statement_opinion: string
                child 5, icfr_opinion: string
                child 6, evidence_ids_used: string
                child 7, unsupported_claims: string
              gold_standard: struct<revenue_growth_pct: double, gross_margin_pct: double, operating_margin_pct: double, net_margi (... 154 chars omitted)
                child 0, revenue_growth_pct: double
                child 1, gross_margin_pct: double
                child 2, operating_margin_pct: double
                child 3, net_margin_pct: double
                child 4, financial_statement_opinion: string
                child 5, icfr_opinion: string
                child 6, required_evidence_ids: list<item: string>
                    child 0, item: string
                child 7, unsupported_claims: list<item: null>
                    child 0, item: null
              controls: struct<evidence_before_narrative: bool, deterministic_recalculation_required: bool, reviewer_require (... 62 chars omitted)
                child 0, evidence_before_narrative: bool
                child 1, deterministic_recalculation_required: bool
                child 2, reviewer_required: bool
                child 3, human_gate: string
                child 4, no_audit_opinion_by_system: bool
              human_gate: struct<required: bool, reason: string>
                child 0, required: bool
                child 1, reason: string
              difficulty: string
              domain: list<item: string>
                child 0, item: string
              to
              {'case_id': Value('string'), 'title': Value('string'), 'domain': List(Value('string')), 'difficulty': Value('string'), 'evidence': List({'id': Value('string'), 'fact': Value('string')}), 'task': List(Value('string')), 'required_output_schema': {'adjusted_bank_balance': Value('string'), 'adjusted_book_balance': Value('string'), 'book_entry_items': Value('string'), 'bank_only_items': Value('string'), 'journal_entries': Value('string'), 'balances': Value('string'), 'evidence_ids_used': Value('string'), 'unsupported_claims': Value('string')}, 'gold_standard': {'adjusted_bank_balance': Value('int64'), 'adjusted_book_balance': Value('int64'), 'book_entry_items': List(Value('string')), 'bank_only_items': List(Value('string')), 'journal_entries': List({'debit_account': Value('string'), 'credit_account': Value('string'), 'amount': Value('int64')}), 'balances': Value('bool'), 'required_evidence_ids': List(Value('string')), 'unsupported_claims': List(Value('null'))}, 'human_gate': {'required': Value('bool'), 'reason': 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.

FRANKENSTEIN Phase 4 Accounting & Audit Benchmark

A provider-neutral research benchmark for testing AI systems on a frozen accounting/audit case while holding the evidence packet, prompt, output schema, and deterministic scoring logic constant.

Canonical GitHub source

https://github.com/Saehon/Saeid-Homayoun/tree/main/FRANKENSTEIN/phase4_benchmark

Case 001 — BANK-REC-001

The first case is a deterministic bank reconciliation. The gold-standard adjusted bank balance and adjusted book balance are both 100,500.

The benchmark evaluates:

adjusted bank balance;
adjusted book balance;
book-entry items;
bank-only reconciling items;
journal entries;
balance conclusion;
evidence completeness;
unsupported claims.

Provider families

The research harness is designed for:

GPT/Codex;
Claude;
Gemini;
Microsoft/Azure;
Kimi/Moonshot;
DeepSeek.

No provider API keys are included here. No live-provider ranking is claimed from the first controlled workflow because all provider calls were skipped when credentials/model variables were absent.

Research boundary

This is a research and educational benchmark. It does not issue an audit opinion, declare accounting compliance, or establish provider superiority.

Case 002 — MSFT-SEC-2026-001

Real-company benchmark based on Microsoft Corporation's FY2026 SEC Form 10-K.

Authoritative filing: https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm

Frozen deterministic targets:

FY2026 revenue growth vs FY2025: 17.79%
FY2026 gross margin: 67.94%
FY2026 operating margin: 46.78%
FY2026 net margin: 40.31%
Financial-statement audit opinion: unqualified
ICFR audit opinion: unqualified

The case validates evidence-grounded extraction, recalculation, opinion classification, reviewer controls, and human-gate logic. It does not claim that the benchmark system performed an audit.

Downloads last month: 248

Total file size:

18.8 kB