Hugging Face's logo

Datasets:
csoai
/
gspc-academy-agi

Tasks:
Text Classification
Languages:
English
Size:
n<1K
Tags:
gspc
council-of-ai
measurement
safety
License:
Dataset card Files
xet
 Community
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
item_id: string
scenario: string
gold_label: string
arms: struct<deepseek-r1:8b: struct<prediction: string, correct: bool>, gemma3:12b: struct<prediction: str (... 1139 chars omitted)
  child 0, deepseek-r1:8b: struct<prediction: string, correct: bool>
      child 0, prediction: string
      child 1, correct: bool
  child 1, gemma3:12b: struct<prediction: string, correct: bool>
      child 0, prediction: string
      child 1, correct: bool
  child 2, llama3.2:3b: struct<prediction: string, correct: bool>
      child 0, prediction: string
      child 1, correct: bool
  child 3, mistral:7b: struct<prediction: string, correct: bool>
      child 0, prediction: string
      child 1, correct: bool
  child 4, qwen2.5:0.5b-instruct: struct<prediction: string, correct: bool>
      child 0, prediction: string
      child 1, correct: bool
  child 5, qwen2.5:3b: struct<prediction: string, correct: bool>
      child 0, prediction: string
      child 1, correct: bool
  child 6, sov6-abstraction-v3-light: struct<prediction: string, correct: bool>
      child 0, prediction: string
      child 1, correct: bool
  child 7, sov6-aesthetics-v3-light: struct<prediction: string, correct: bool>
      child 0, prediction: string
      child 1, correct: bool
  child 8, sov6-agency-v3-light: struct<prediction: string, correct: bool>
      child 0, prediction: string
      child 1, correct: bool
  child 9, sov6-creation-v3-light: struct<prediction: string, correct: bool>
      child 0, prediction: strin
...
 accuracy: double
      child 4, wilson95: list<item: double>
          child 0, item: double
  child 6, next_best: struct<model: string, k: int64, n: int64, accuracy: double, wilson95: list<item: double>>
      child 0, model: string
      child 1, k: int64
      child 2, n: int64
      child 3, accuracy: double
      child 4, wilson95: list<item: double>
          child 0, item: double
  child 7, paired_items: int64
  child 8, discordant: struct<leader_only_correct: int64, next_best_only_correct: int64>
      child 0, leader_only_correct: int64
      child 1, next_best_only_correct: int64
  child 9, mcnemar_p: double
  child 10, control: struct<what: string, nperm: int64, seed: int64, fixed_pair_rate: double, reselect_rate: double>
      child 0, what: string
      child 1, nperm: int64
      child 2, seed: int64
      child 3, fixed_pair_rate: double
      child 4, reselect_rate: double
  child 11, recompute: string
macro_f1: double
status: string
separation_sentence: string
dataset_url: string
interval: list<item: double>
  child 0, item: double
mde: struct<value: null, state: string, reason: string, unit: string, alpha: double, power: double, test: (... 71 chars omitted)
  child 0, value: null
  child 1, state: string
  child 2, reason: string
  child 3, unit: string
  child 4, alpha: double
  child 5, power: double
  child 6, test: string
  child 7, paired_items: int64
  child 8, discordance_rate: double
  child 9, source: string
distinct_items_source: string
hue: int64
to
{'axis': Value('string'), 'family': Value('string'), 'kind': Value('string'), 'bench': Value('string'), 'task': Value('string'), 'n': Value('int64'), 'separation': Value('string'), 'fleet_mean': Value('float64'), 'mean_harm': Value('float64'), 'cvar05_harm': Value('null'), 'status': Value('string'), 'dataset': Value('string'), 'colour': Value('string'), 'hue': Value('int64'), 'method_status': Value('string'), 'leader': Value('string'), 'accuracy': Value('float64'), 'interval': List(Value('float64')), 'macro_f1': Value('float64'), 'unparsed_rate': Value('float64'), 'separation_p': Value('float64'), 'separation_method': Value('string'), 'separation_sentence': Value('string'), 'separation_evidence': {'source': Value('string'), 'file': Value('string'), 'file_sha256': Value('string'), 'peritem_sha256': Value('string'), 'fleet': Value('string'), 'leader': {'model': Value('string'), 'k': Value('int64'), 'n': Value('int64'), 'accuracy': Value('float64'), 'wilson95': List(Value('float64'))}, 'next_best': {'model': Value('string'), 'k': Value('int64'), 'n': Value('int64'), 'accuracy': Value('float64'), 'wilson95': List(Value('float64'))}, 'paired_items': Value('int64'), 'discordant': {'leader_only_correct': Value('int64'), 'next_best_only_correct': Value('int64')}, 'mcnemar_p': Value('float64'), 'control': {'what': Value('string'), 'nperm': Value('int64'), 'seed': Value('int64'), 'fixed_pair_rate': Value('float64'), 'reselect_rate': Value('float64')}, 'recompute': Value('string')}, 'peritem_file_sha256': Value('string'), 'leader_source': Value('string'), 'leader_card_state': Value('string'), 'leader_card_note': Value('string'), 'note': Value('string'), 'distinct_items': Value('int64'), 'distinct_items_source': Value('string'), 'mde': {'value': Value('null'), 'state': Value('string'), 'reason': Value('string'), 'unit': Value('string'), 'alpha': Value('float64'), 'power': Value('float64'), 'test': Value('string'), 'paired_items': Value('int64'), 'discordance_rate': Value('float64'), 'source': Value('string')}, 'measurement_time': {'state': Value('string'), 'precision': Value('string'), 'observed_on': Value('timestamp[s]'), 'source': Value('string'), 'source_field': Value('string'), 'note': Value('string')}, 'dataset_url': 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
              item_id: string
              scenario: string
              gold_label: string
              arms: struct<deepseek-r1:8b: struct<prediction: string, correct: bool>, gemma3:12b: struct<prediction: str (... 1139 chars omitted)
                child 0, deepseek-r1:8b: struct<prediction: string, correct: bool>
                    child 0, prediction: string
                    child 1, correct: bool
                child 1, gemma3:12b: struct<prediction: string, correct: bool>
                    child 0, prediction: string
                    child 1, correct: bool
                child 2, llama3.2:3b: struct<prediction: string, correct: bool>
                    child 0, prediction: string
                    child 1, correct: bool
                child 3, mistral:7b: struct<prediction: string, correct: bool>
                    child 0, prediction: string
                    child 1, correct: bool
                child 4, qwen2.5:0.5b-instruct: struct<prediction: string, correct: bool>
                    child 0, prediction: string
                    child 1, correct: bool
                child 5, qwen2.5:3b: struct<prediction: string, correct: bool>
                    child 0, prediction: string
                    child 1, correct: bool
                child 6, sov6-abstraction-v3-light: struct<prediction: string, correct: bool>
                    child 0, prediction: string
                    child 1, correct: bool
                child 7, sov6-aesthetics-v3-light: struct<prediction: string, correct: bool>
                    child 0, prediction: string
                    child 1, correct: bool
                child 8, sov6-agency-v3-light: struct<prediction: string, correct: bool>
                    child 0, prediction: string
                    child 1, correct: bool
                child 9, sov6-creation-v3-light: struct<prediction: string, correct: bool>
                    child 0, prediction: strin
              ...
               accuracy: double
                    child 4, wilson95: list<item: double>
                        child 0, item: double
                child 6, next_best: struct<model: string, k: int64, n: int64, accuracy: double, wilson95: list<item: double>>
                    child 0, model: string
                    child 1, k: int64
                    child 2, n: int64
                    child 3, accuracy: double
                    child 4, wilson95: list<item: double>
                        child 0, item: double
                child 7, paired_items: int64
                child 8, discordant: struct<leader_only_correct: int64, next_best_only_correct: int64>
                    child 0, leader_only_correct: int64
                    child 1, next_best_only_correct: int64
                child 9, mcnemar_p: double
                child 10, control: struct<what: string, nperm: int64, seed: int64, fixed_pair_rate: double, reselect_rate: double>
                    child 0, what: string
                    child 1, nperm: int64
                    child 2, seed: int64
                    child 3, fixed_pair_rate: double
                    child 4, reselect_rate: double
                child 11, recompute: string
              macro_f1: double
              status: string
              separation_sentence: string
              dataset_url: string
              interval: list<item: double>
                child 0, item: double
              mde: struct<value: null, state: string, reason: string, unit: string, alpha: double, power: double, test: (... 71 chars omitted)
                child 0, value: null
                child 1, state: string
                child 2, reason: string
                child 3, unit: string
                child 4, alpha: double
                child 5, power: double
                child 6, test: string
                child 7, paired_items: int64
                child 8, discordance_rate: double
                child 9, source: string
              distinct_items_source: string
              hue: int64
              to
              {'axis': Value('string'), 'family': Value('string'), 'kind': Value('string'), 'bench': Value('string'), 'task': Value('string'), 'n': Value('int64'), 'separation': Value('string'), 'fleet_mean': Value('float64'), 'mean_harm': Value('float64'), 'cvar05_harm': Value('null'), 'status': Value('string'), 'dataset': Value('string'), 'colour': Value('string'), 'hue': Value('int64'), 'method_status': Value('string'), 'leader': Value('string'), 'accuracy': Value('float64'), 'interval': List(Value('float64')), 'macro_f1': Value('float64'), 'unparsed_rate': Value('float64'), 'separation_p': Value('float64'), 'separation_method': Value('string'), 'separation_sentence': Value('string'), 'separation_evidence': {'source': Value('string'), 'file': Value('string'), 'file_sha256': Value('string'), 'peritem_sha256': Value('string'), 'fleet': Value('string'), 'leader': {'model': Value('string'), 'k': Value('int64'), 'n': Value('int64'), 'accuracy': Value('float64'), 'wilson95': List(Value('float64'))}, 'next_best': {'model': Value('string'), 'k': Value('int64'), 'n': Value('int64'), 'accuracy': Value('float64'), 'wilson95': List(Value('float64'))}, 'paired_items': Value('int64'), 'discordant': {'leader_only_correct': Value('int64'), 'next_best_only_correct': Value('int64')}, 'mcnemar_p': Value('float64'), 'control': {'what': Value('string'), 'nperm': Value('int64'), 'seed': Value('int64'), 'fixed_pair_rate': Value('float64'), 'reselect_rate': Value('float64')}, 'recompute': Value('string')}, 'peritem_file_sha256': Value('string'), 'leader_source': Value('string'), 'leader_card_state': Value('string'), 'leader_card_note': Value('string'), 'note': Value('string'), 'distinct_items': Value('int64'), 'distinct_items_source': Value('string'), 'mde': {'value': Value('null'), 'state': Value('string'), 'reason': Value('string'), 'unit': Value('string'), 'alpha': Value('float64'), 'power': Value('float64'), 'test': Value('string'), 'paired_items': Value('int64'), 'discordance_rate': Value('float64'), 'source': Value('string')}, 'measurement_time': {'state': Value('string'), 'precision': Value('string'), 'observed_on': Value('timestamp[s]'), 'source': Value('string'), 'source_field': Value('string'), 'note': Value('string')}, 'dataset_url': 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.

Council of AI (CSOAI)

GSPC Academy — Safety (agi axis)

Training + MOOC data for the safety axis specialist — feeds the Council of AI agentic specialist hive and the GSPC Human-in-the-Loop academy.

Every item carries per-model outcomes from real measured runs: a 19-model fleet (6 base models — deepseek-r1:8b, gemma3:12b, llama3.2:3b, mistral:7b, qwen2.5:0.5b-instruct, qwen2.5:3b — and 13 CSOAI sov6-* prompt-overlay specialists), deterministic grading, 2026-08-12 measurement date, zero transport errors.

Stats

metric value
items 36
arms 19
hard cases (majority_wrong) 4
split 0
majority-correct 32
measurement rows behind it 684

The 4 majority_wrong items are the hard cases — the training gold for the safety specialist.

Per-arm accuracy (measured)

arm runs correct accuracy
deepseek-r1:8b 36 28 0.778
gemma3:12b 36 34 0.944
llama3.2:3b 36 16 0.444
mistral:7b 36 29 0.806
qwen2.5:0.5b-instruct 36 15 0.417
qwen2.5:3b 36 32 0.889
sov6-abstraction-v3-light 36 19 0.528
sov6-aesthetics-v3-light 36 30 0.833
sov6-agency-v3-light 36 30 0.833
sov6-creation-v3-light 36 31 0.861
sov6-destruction-v3-light 36 30 0.833
sov6-embodiment-v3-light 36 30 0.833
sov6-ethics-v3-light 36 29 0.806
sov6-identity-v3-light 36 21 0.583
sov6-logic-v3-light 36 29 0.806
sov6-preservation-v3-light 36 33 0.917
sov6-relationality-v3-light 36 23 0.639
sov6-synthesis-v3-light 36 24 0.667
sov6-temporality-v3-light 36 18 0.500

Fleet total: 501/684 correct (0.732).

Files

  • items.jsonl — one record per unique item: {item_id, scenario, gold_label, arms, discrimination, item_sha256}. Each arms entry is {prediction, correct} (plus runs/correct_runs where an arm was measured more than once on the item).

  • specialist-prompt.md — the lane's specialist prompt. Not yet statutory-pinned — honest placeholder, no rubric invented.

  • mooc-map.json — curriculum map: Observer (up to 30 easiest items), Steward (medium discrimination), Sentinel (the majority_wrong hard cases).

  • board-card.json — this axis's entry from the live GSPC board API, verbatim.

  • 37 arm outcomes were unparseable; they are recorded with prediction "UNPARSED" and correct: false — excluded nothing, hidden nothing.

Board row

bench task n separation fleet_mean dataset colour
DefBench calibrated refusal on paired requests 36 TIE 0.732 csoai/gspc-agi #f87171

Live board: councilof.ai/board · live API serves the same axis entries (the board card here is a verbatim snapshot, measured_on behavioural axes 2026-08-12).

MOOC ranks

mooc-map.json assigns every item to exactly one tier: Observer (up to 30 easiest, largest correct-arm margin), Steward (medium discrimination: narrow majorities and split items), Sentinel (every majority_wrong hard case). Signing/review flows through the EAT review queue.

Upstream attribution & licence

Items are CSOAI's own GSPC bank for this axis (no external upstream). The raw per-item measurement rows are published at csoai/gspc-peritem-rows-2026-08-12.

Licence: CC-BY-4.0.

The rule

Measurement, not certification. Items are public; board changes require human-reviewed measurement runs.

Part of the GSPC Academy series — one dataset per measurement axis, patterned after the topic-partitioned JSONL + high-quality card format, with per-model measured outcomes on every item.
Downloads last month
73
Total file size:
69.8 kB