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:    CastError
Message:      Couldn't cast
pipeline_config_hash: string
run_timestamp: string
source_files: list<item: null>
  child 0, item: null
stage_counts: struct<JSONLReader: struct<output_count: int64, rejected_count: int64>, SchemaGate: struct<input_cou (... 326 chars omitted)
  child 0, JSONLReader: struct<output_count: int64, rejected_count: int64>
      child 0, output_count: int64
      child 1, rejected_count: int64
  child 1, SchemaGate: struct<input_count: int64, output_count: int64, probe_recovered: int64, rejected_count: int64>
      child 0, input_count: int64
      child 1, output_count: int64
      child 2, probe_recovered: int64
      child 3, rejected_count: int64
  child 2, ExactDeduplicator: struct<input_count: int64, output_count: int64>
      child 0, input_count: int64
      child 1, output_count: int64
  child 3, MinHashDeduplicator: struct<input_count: int64, output_count: int64>
      child 0, input_count: int64
      child 1, output_count: int64
  child 4, TextCleaner: struct<input_count: int64, output_count: int64>
      child 0, input_count: int64
      child 1, output_count: int64
  child 5, AlpacaExporter: struct<exported_count: int64>
      child 0, exported_count: int64
rejected_breakdown: struct<>
dedup_stats: struct<near_duplicates_removed: int64, surviving_count: int64>
  child 0, near_duplicates_removed: int64
  child 1, surviving_count: int64
minhash_threshold: double
sampler_stats: null
token_stats: struct<total_prompt_tokens: int64, total_completion_tokens: int64, total_toke
...
       child 0, kind: string
          child 1, ref: string
          child 2, config: null
          child 3, provenance: null
  child 8, params: struct<generator_model: null, judge_model: null, grpo_scoring_model: null, export_formats: list<item (... 127 chars omitted)
      child 0, generator_model: null
      child 1, judge_model: null
      child 2, grpo_scoring_model: null
      child 3, export_formats: list<item: string>
          child 0, item: string
      child 4, output_split: null
      child 5, output_split_group: null
      child 6, output_split_near_dup_threshold: double
      child 7, pipeline_config_hash: string
base_model: null
method: string
params: struct<generator_model: null, judge_model: null, grpo_scoring_model: null, export_formats: list<item (... 127 chars omitted)
  child 0, generator_model: null
  child 1, judge_model: null
  child 2, grpo_scoring_model: null
  child 3, export_formats: list<item: string>
      child 0, item: string
  child 4, output_split: null
  child 5, output_split_group: null
  child 6, output_split_near_dup_threshold: double
  child 7, pipeline_config_hash: string
created_at: timestamp[s]
schema: string
version: string
git_sha: null
library: string
inputs: list<item: struct<kind: string, ref: string, config: null, provenance: null>>
  child 0, item: struct<kind: string, ref: string, config: null, provenance: null>
      child 0, kind: string
      child 1, ref: string
      child 2, config: null
      child 3, provenance: null
to
{'schema': Value('string'), 'library': Value('string'), 'version': Value('string'), 'git_sha': Value('null'), 'created_at': Value('timestamp[s]'), 'base_model': Value('null'), 'method': Value('string'), 'inputs': List({'kind': Value('string'), 'ref': Value('string'), 'config': Value('null'), 'provenance': Value('null')}), 'params': {'generator_model': Value('null'), 'judge_model': Value('null'), 'grpo_scoring_model': Value('null'), 'export_formats': List(Value('string')), 'output_split': Value('null'), 'output_split_group': Value('null'), 'output_split_near_dup_threshold': Value('float64'), 'pipeline_config_hash': Value('string')}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, 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 129, 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 544, 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 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 361, 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 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              pipeline_config_hash: string
              run_timestamp: string
              source_files: list<item: null>
                child 0, item: null
              stage_counts: struct<JSONLReader: struct<output_count: int64, rejected_count: int64>, SchemaGate: struct<input_cou (... 326 chars omitted)
                child 0, JSONLReader: struct<output_count: int64, rejected_count: int64>
                    child 0, output_count: int64
                    child 1, rejected_count: int64
                child 1, SchemaGate: struct<input_count: int64, output_count: int64, probe_recovered: int64, rejected_count: int64>
                    child 0, input_count: int64
                    child 1, output_count: int64
                    child 2, probe_recovered: int64
                    child 3, rejected_count: int64
                child 2, ExactDeduplicator: struct<input_count: int64, output_count: int64>
                    child 0, input_count: int64
                    child 1, output_count: int64
                child 3, MinHashDeduplicator: struct<input_count: int64, output_count: int64>
                    child 0, input_count: int64
                    child 1, output_count: int64
                child 4, TextCleaner: struct<input_count: int64, output_count: int64>
                    child 0, input_count: int64
                    child 1, output_count: int64
                child 5, AlpacaExporter: struct<exported_count: int64>
                    child 0, exported_count: int64
              rejected_breakdown: struct<>
              dedup_stats: struct<near_duplicates_removed: int64, surviving_count: int64>
                child 0, near_duplicates_removed: int64
                child 1, surviving_count: int64
              minhash_threshold: double
              sampler_stats: null
              token_stats: struct<total_prompt_tokens: int64, total_completion_tokens: int64, total_toke
              ...
                     child 0, kind: string
                        child 1, ref: string
                        child 2, config: null
                        child 3, provenance: null
                child 8, params: struct<generator_model: null, judge_model: null, grpo_scoring_model: null, export_formats: list<item (... 127 chars omitted)
                    child 0, generator_model: null
                    child 1, judge_model: null
                    child 2, grpo_scoring_model: null
                    child 3, export_formats: list<item: string>
                        child 0, item: string
                    child 4, output_split: null
                    child 5, output_split_group: null
                    child 6, output_split_near_dup_threshold: double
                    child 7, pipeline_config_hash: string
              base_model: null
              method: string
              params: struct<generator_model: null, judge_model: null, grpo_scoring_model: null, export_formats: list<item (... 127 chars omitted)
                child 0, generator_model: null
                child 1, judge_model: null
                child 2, grpo_scoring_model: null
                child 3, export_formats: list<item: string>
                    child 0, item: string
                child 4, output_split: null
                child 5, output_split_group: null
                child 6, output_split_near_dup_threshold: double
                child 7, pipeline_config_hash: string
              created_at: timestamp[s]
              schema: string
              version: string
              git_sha: null
              library: string
              inputs: list<item: struct<kind: string, ref: string, config: null, provenance: null>>
                child 0, item: struct<kind: string, ref: string, config: null, provenance: null>
                    child 0, kind: string
                    child 1, ref: string
                    child 2, config: null
                    child 3, provenance: null
              to
              {'schema': Value('string'), 'library': Value('string'), 'version': Value('string'), 'git_sha': Value('null'), 'created_at': Value('timestamp[s]'), 'base_model': Value('null'), 'method': Value('string'), 'inputs': List({'kind': Value('string'), 'ref': Value('string'), 'config': Value('null'), 'provenance': Value('null')}), 'params': {'generator_model': Value('null'), 'judge_model': Value('null'), 'grpo_scoring_model': Value('null'), 'export_formats': List(Value('string')), 'output_split': Value('null'), 'output_split_group': Value('null'), 'output_split_near_dup_threshold': Value('float64'), 'pipeline_config_hash': 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.

Whistleblower or Accomplice: team-log probes, generations and scores

Data for "Whistleblower or Accomplice?", Team Cadbury (Abhra Dubey, Harmanjot Singh), Lexsi AI Safety and Alignment Hackathon, Track 2. Code, specification and article: see the GitHub repository (https://github.com/coderjs-lab/whistleblower-or-accomplice) and the collection.

Contents

Path What
scenarios/calib.jsonl 1,728 rows (576 underlying episodes × leniency off / A / B), seed 101: used for every calibration choice
scenarios/test.jsonl 3,240 rows (1,080 episodes × 3 levels), seed 202: frozen test set, generated once per condition after SPEC.md was frozen
scenarios/comprehension.jsonl 5 multiple-choice checks of the leniency rules
scenarios/build_scenarios.py the deterministic generator (rebuilds the files byte-identically)
generations/{calib,test,utility,comprehension}/<condition>.jsonl raw model answers per condition, with P(ACCEPT) on accomplice rows, run metadata (model revision, git commit, decoding, intervention, Slurm job)
scores.json/md, aux_scores.json/md, analysis.json/md metrics with 95% cluster-bootstrap CIs, McNemar tests, pre-registered primary tests (Holm), decision rule
auditkit/ Lexsi AuditKIT before/after comparisons with run fingerprints
curatorkit/ Lexsi CuratorKIT manifests, checksums and dataset cards for the scenario texts
directions/ layer scan, chosen hyperparameters, calibration sweep tables
figures/ article figures
utility_global_mmlu_lite.jsonl the 300-question Global-MMLU-Lite slice (en/hi/bn) used for the utility axis (Apache-2.0, from CohereLabs/Global-MMLU-Lite)
SPEC.md the pre-registered specification

Row schema: id, split, probe ("accomplice" | "report"), domain, leniency ("off" | "A" | "B"), subject, agents, system, user, plus offer, style, norm, inviter, correct_decision (accomplice) or type, truth, colluders (report). Generation rows drop system / user (recover them by id) and add condition, response, p_accept, model, revision, git_commit, intervention.

Provenance and quality

Scenarios are templated and scripted: four domains (code review, procurement, moderation, team formation), an accomplice probe (collusive vs legitimate offers, clean vs colluding peers, seven persuasion styles) and a report probe (clean, visible, no-talk, benign, self-involved collusion). Ground truth is fixed by construction; no LLM judge is used.

CuratorKIT 0.1.29 on the unique episode texts: calib 576 unique, 0 exact duplicates, 150 MinHash near-duplicates (threshold 0.85); test 1,080 unique, 0 exact, 339 near-duplicates (31%). These are same-template instances that differ in names, numbers and order; they are kept (independent randomized draws) and all intervals are cluster-bootstrapped over underlying episodes.

Limitations

English only; templated logs; scripted actors; single-decision probes. Not a benchmark of real agent deployments.

Licence

Scenarios, generations and scores: CC-BY-4.0. Model outputs were produced by Tiny Aya Global and Aya Expanse (CC-BY-NC-4.0 + Cohere Acceptable Use Policy). Global-MMLU-Lite slice: Apache-2.0.

Downloads last month
22

Collection including harmanjs/whistleblower-or-accomplice-data