Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot extract the features (columns) for the split 'test' of the config 'component_evaluation' of the dataset.
Error code:   FeaturesError
Exception:    ArrowInvalid
Message:      JSON parse error: Invalid value. in row 0
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 324, in _generate_tables
                  df = pandas_read_json(f)
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 38, in pandas_read_json
                  return pd.read_json(path_or_buf, **kwargs)
                         ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 815, in read_json
                  return json_reader.read()
                         ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 1014, in read
                  obj = self._get_object_parser(self.data)
                File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 1040, in _get_object_parser
                  obj = FrameParser(json, **kwargs).parse()
                File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 1176, in parse
                  self._parse()
                  ~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 1392, in _parse
                  ujson_loads(json, precise_float=self.precise_float), dtype=None
                  ~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
              ValueError: Unexpected character found when decoding 'true'
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 244, in compute_first_rows_from_streaming_response
                  iterable_dataset = iterable_dataset._resolve_features()
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 4408, in _resolve_features
                  features = _infer_features_from_batch(self.with_format(None)._head())
                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2679, in _head
                  return next(iter(self.iter(batch_size=n)))
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2861, in iter
                  for key, pa_table in ex_iterable.iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2395, in _iter_arrow
                  yield from 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 327, in _generate_tables
                  raise e
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 290, in _generate_tables
                  pa_table = paj.read_json(
                      io.BytesIO(batch), read_options=paj.ReadOptions(block_size=block_size)
                  )
                File "pyarrow/_json.pyx", line 342, in pyarrow._json.read_json
                File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
                  raise convert_status(status)
              pyarrow.lib.ArrowInvalid: JSON parse error: Invalid value. in row 0

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.

RAISER Responsibility and Repair Benchmark

This dataset supports research on event-level responsibility and checker-validated repair in LLM multi-agent traces. It separates four evidence layers that should not be treated as interchangeable:

  1. six-role model-consensus and silver labels;
  2. independent human MV/CV labels with blind expert adjudication;
  3. blinded broad-rubric Event-Source cross-annotation; and
  4. text-free outcomes from frozen repair evaluations.

Historical artifact identifiers may retain the TRACE-R prefix for stable lineage. No third-party trace text is redistributed.

Contents

Configuration Split Unit Rows
model_benchmark train event 43,077
model_benchmark test event 5,516
human_gold test candidate-role pair 1,600
event_source test event 1,267
repair_validation validation trace 346
component_evaluation test trace-configuration 4,800
native_evaluation test framework-selector trace 600

The model benchmark covers 3,121 traces from AEGIS, CORRECT-Error, MAST-Data, TraceElephant, TRAIL, and Who&When. Its held-out model-consensus portion contains 484 traces; the model-silver portion contains 2,637 traces.

Configurations

model_benchmark

Each event receives six binary labels: source, propagator, missed_verifier, contract_violation, symptom, and repair_leverage. The test rows use three-model cross-adjudicated labels. Training rows use the model-silver protocol. Confidence and label-tier metadata are retained.

human_gold

This set contains 850 MissedVerifier candidates with 114 positives and 750 ContractViolation candidates with 200 positives. Two humans labeled every candidate independently; 585 disagreements or uncertain cases were resolved by a blind expert. Free-text evidence and rationales are excluded.

event_source

Two blinded model annotators independently applied a broad Event-Source rubric to 1,267 events from 116 repairable failures. The file preserves both labels, confidence, boundary flags, strict intersection, positive union, and resolution status. This is a model cross-annotation sensitivity set, not human Gold.

Repair evaluation configurations

  • untouched_346.csv contains paired RAISER/query-only outcomes on a frozen policy-validation cohort.
  • component_1200.csv contains four prespecified candidate/ranker/verifier configurations per trace with Oracle@3, Rank@1, and Repair@1 outcomes.
  • native_checkpoint_150.csv contains matched LangGraph and AutoGen outcomes for RAISER and majority under checkpointed selected-event intervention.

These files contain hashes, identifiers, categorical configuration names, and binary outcomes only. They do not contain questions, answers, conversation text, prompts, completions, or reference labels.

Loading

from datasets import load_dataset

roles = load_dataset("RogerYao/RAISER", "model_benchmark")
human_gold = load_dataset("RogerYao/RAISER", "human_gold")
event_source = load_dataset("RogerYao/RAISER", "event_source")
repair_validation = load_dataset("RogerYao/RAISER", "repair_validation")
components = load_dataset("RogerYao/RAISER", "component_evaluation")
native = load_dataset("RogerYao/RAISER", "native_evaluation")

Joining source traces

  1. Obtain source datasets listed in metadata/source_datasets.csv.
  2. Normalize their event identifiers with the RAISER code repository.
  3. Join on dataset_source, trace_id, and event_id where applicable.
  4. Respect every source repository's current terms and access controls.

TRAIL content must remain in an approved gated or private repository and is intended for evaluation. This release grants no rights to upstream content.

Leakage controls

Events from one trace are never split across role-model partitions. Human Gold, Event-Source labels, checker answers, and repair outcomes are outside the role model's training boundary. Frozen repair cohorts retain their original status; the 346-trace cohort is validation evidence and the 1,200-trace component cohort is a later confirmatory test.

Limitations

  • Model-consensus labels are not interchangeable with human Gold.
  • Event-Source is not human-expert Source Gold and is rubric-sensitive.
  • Raw traces are required for text-based modeling but are not redistributed.
  • Repair files expose binary outcomes, not source questions or answers.
  • Upstream licenses and access terms may change; verify them before use.

Licensing and attribution

See LICENSE_DATA.md, DATA_STATEMENT.md, PROVENANCE.md, and metadata/source_datasets.csv. The other license tag is deliberate because the release combines original annotations with identifiers referring to datasets governed by different terms.

Validation

python scripts/validate_dataset.py .

The validator checks schemas, row counts, unique keys, binary fields, paired coverage, and the absence of raw-content fields.

Downloads last month
49