Dataset Viewer
Duplicate
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:    ValueError
Message:      Expected object or value
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 281, in _generate_tables
                  examples = [ujson_loads(line) for line in batch.splitlines()]
                              ~~~~~~~~~~~^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
                  return pd.io.json.ujson_loads(*args, **kwargs)
                         ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
              ValueError: Expected object or value

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.

ProofAI Verified Tasks

Dataset Description

ProofAI Verified Tasks is an early demonstration dataset of privacy-sanitized records where an AI system attempted a real-world-style task and ProofAI recorded evidence, verification, and (when applicable) external outcomes.

ProofAI evaluates whether AI systems actually completed tasks using source evidence, deterministic checks, human verification, and external observations.

This is not a trivia Q&A dataset and not a popularity-vote leaderboard. With only 4 record(s) currently, it is a seed, not a statistically representative AI-agent benchmark.

Motivation

Answers are cheap. Evidence and outcomes are not. This dataset captures the unit:

task β†’ execution β†’ claims β†’ verification β†’ (optional) authorized action β†’ receipt β†’ external observation β†’ verified outcome β†’ proof hash.

Dataset Structure

README.md
data/verified_tasks.jsonl
metadata/dataset_stats.json
examples/proof_bundle_example.json

Each JSONL line is one verified demo task/run (schema_version = 1.0).

Data Fields

  • schema_version, task_id, proof_id, proof_hash
  • category, task_title, task_description, success_criteria
  • input_type (text | pdf | external_action | …)
  • source_evidence (sanitized excerpts + locators)
  • executor (provider, model)
  • claims, verification_results, verification_summary
  • actions, external_observations
  • verified_task_outcome, external_resolution_status, completion_verified
  • judge_parse_status, estimated_cost_usd, latency_ms
  • verification_level, is_demo

Verification Hierarchy

Higher authority wins for factual claims:

  1. externally_verified β€” independent observation of an external system state
  2. deterministic / deterministic_verified β€” objective checks (math, evidence IDs)
  3. source_supported / source_observed β€” extracted from uploaded source material
  4. human_verified / user_confirmed β€” human outcome verification
  5. ai_judge β€” provisional model evaluation of attempts
  6. ai_only / model_claim β€” unsupported executor statement

Task Collection

Tasks are created in ProofAI, executed with configured Hugging Face Inference Providers, verified by ProofAI services, optionally acted upon via allowlisted tools, and concluded with a human outcome record. Only is_demo=true tasks that pass sanitization and ProofBundle integrity checks are eligible for this public dataset.

Actions are optional. Document-only tasks with strong deterministic/source verification are valid benchmark records.

Proof Bundles

Each row includes proof_hash tying the export to a tamper-evident ProofBundle (SHA-256 over a canonical sanitized payload). The hash is integrity/tamper-evidence for the ProofAI-generated bundle β€” not a cryptographic proof of absolute real-world truth.

External Verification

When tools such as GitHub issue create/close are used, ProofAI stores independent GET observations (e.g. OPEN then CLOSED) rather than trusting only a local receipt flag.

Intended Uses

  • Research on agent evaluation with evidence and outcomes
  • Demo / educational inspection of ProofAI verification records
  • Seed data for future multi-model RealWorldArena-style benchmarks

Out-of-Scope Uses

  • Claiming SOTA model rankings from this tiny seed
  • Training systems to exfiltrate private documents (none are included)
  • Treating proof_hash as blockchain-style proof of real-world events
  • Using exported bundles as authorization to execute tools

Privacy

Public dataset rows are demo-only. Non-demo / private user uploads must not appear.

Sanitization

Exports remove API tokens, Authorization headers, local filesystem paths, raw uploaded binaries, and common PII patterns where detected.

Limitations

  • Tiny early sample size (4 records at generation time).
  • Demo-heavy; not representative of all real-world tasks.
  • External verification currently covers a narrow allowlisted tool set.
  • Model choice constrained by free inference availability.
  • Integrity hash β‰  independent truth of every underlying source.

Known Biases

  • Consumer document / bill / dispute scenarios are over-represented early.
  • Free-tier model availability shapes executor choice.
  • Demo fixtures may understate messy real-world ambiguity.

Dataset Size

Current snapshot (from metadata/dataset_stats.json at generation):

  • records_total: 4
  • verified_tasks: 1
  • externally_verified_tasks: 2
  • deterministically_verified_tasks: 4
  • document_tasks: 3
  • external_action_tasks: 2

Versioning

  • Dataset schema_version: 1.0
  • ProofBundle proof_bundle_version is separate (currently 1.0)
  • Do not silently break field semantics; increment versions deliberately

Reproducibility

Rows are derived from persisted ProofAI database state. Export is deterministic for the same underlying records (task_id + proof_hash identity; duplicate exports deduplicated).

Licensing Status

License decision pending.

No license (MIT, Apache, CC-BY, etc.) has been selected by the founder. Do not assume redistribution rights until an explicit license is chosen.

Downloads last month
14