Hugging Face's logo

Datasets:
Coralfil
/
AboveBoard-Marine-52

Tasks:
Question Answering
Multiple Choice
Modalities:
Text
Formats:
parquet
Languages:
English
Size:
< 1K
Tags:
marine-science
physical-oceanography
statutory-compliance
ocean-intelligence
benchmark
aboveboard
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
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
id: string
legacy_id: string
question_type: string
discipline: string
region: string
water_body: string
prompt: string
authority: string
key_criteria: list<element: string>
  child 0, element: string
rubric: string
correct_answer: string
explanation: string
target_blank: string
acceptable_synonyms: list<element: string>
  child 0, element: string
-- schema metadata --
huggingface: '{"info": {"features": {"id": {"dtype": "string", "_type": "' + 766
to
{'id': Value('string'), 'question_type': Value('string'), 'discipline': Value('string'), 'region': Value('string'), 'water_body': Value('string'), 'prompt': Value('string'), 'authority': Value('string'), 'rubric': 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 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, 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 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, 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/parquet/parquet.py", line 220, in _generate_tables
                  yield Key(file_idx, batch_idx), self._cast_table(pa_table)
                                                  ~~~~~~~~~~~~~~~~^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/parquet/parquet.py", line 156, 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
              id: string
              legacy_id: string
              question_type: string
              discipline: string
              region: string
              water_body: string
              prompt: string
              authority: string
              key_criteria: list<element: string>
                child 0, element: string
              rubric: string
              correct_answer: string
              explanation: string
              target_blank: string
              acceptable_synonyms: list<element: string>
                child 0, element: string
              -- schema metadata --
              huggingface: '{"info": {"features": {"id": {"dtype": "string", "_type": "' + 766
              to
              {'id': Value('string'), 'question_type': Value('string'), 'discipline': Value('string'), 'region': Value('string'), 'water_body': Value('string'), 'prompt': Value('string'), 'authority': Value('string'), 'rubric': 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.

AboveBoard Sovereign Canadian Marine Benchmark (AboveBoard 52)

Dataset ID: Coralfil/AboveBoard-Marine-52
License: Apache-2.0
Authors: Coralfil Research Team
Standard: CRF-FS-ABOVEBOARD-52
Punctuation Standard: Zero em dashes enforced.

Overview

The AboveBoard Benchmark Suite is an open-source 52-scenario evaluation standard developed to measure artificial intelligence models across physical oceanography, statutory maritime compliance, and benthic ecological engineering.

Frontier foundation models consistently fail in aquatic and coastal domains due to jurisdictional conflation (applying US EPA/NOAA regulations to Canadian waters) and missing physical conservation invariants (inverting aragonite saturation states, omitting seawater compressibility). AboveBoard establishes an objective, reproducible metric to audit domain precision.

Dataset Taxonomy

The 52 scenarios are grouped into three primary subdisciplines:

  1. Category A: Statutory Marine Compliance (20 scenarios / 38.5%)

    • Fisheries and Oceans Canada (DFO) Fisheries Act Section 35 (HADD criteria and authorizations)
    • Fisheries Act Section 56 (Introductions and transfers of live aquatic organisms)
    • Oceans Act Marine Protected Area (MPA) Boundaries (Hecate Strait Glass Sponge Reefs)
    • Canada Shipping Act 2001 (Ballast water management and Small Vessel Regulations)
    • Transport Canada Notices to Shipping and IMO Polar Code Navigation Mandates

  2. Category B: Physical Oceanography & Biogeochemistry (18 scenarios / 34.6%)

    • TEOS-10 25-term polynomial seawater density derivations
    • Aragonite and calcite saturation horizons ($\Omega_{\text{arag}}$, $\Omega_{\text{calc}}$)
    • Fast Fourier decomposition of semi-diurnal M2 (12.42h) and S2 (12.00h) tidal harmonics
    • Acoustic cavitation thresholds and sound velocity profiles

  3. Category C: Benthic Restoration & Structural Biomaterials (14 scenarios / 26.9%)

    • Chitosan-aragonite composite degradation kinetics under estuarine hypoxia
    • Shear stress tolerance on pulverized glass sponge spicule lattices
    • Larval settlement rugosity thresholds for Haliotis kamtschatkana (Pinto Abalone)
    • Cold-water coral connectivity corridors

Benchmark Leaderboard Baseline Scores

Evaluated under identical zero-shot greedy decoding ($T = 0.0$):

Rank Model Name Scale Statutory Compliance Physical Oceanography Benthic Restoration Overall Score
1 Atlantis-Pyramid² (32B) 32.5B 100.0% 91.7% 100.0% 97.1%
2 DeepSeek-R1-Distill-Qwen 32B 60.0% 76.2% 62.0% 71.8%
3 Qwen2.5-Instruct 32B 65.0% 72.4% 68.2% 67.3%
4 Cohere Command R+ 104B 70.0% 64.2% 68.1% 68.1%
5 Meta Llama-3.3-Instruct 70B 70.0% 70.8% 65.3% 68.5%
6 Claude 3.5 Sonnet (Base) Closed 65.0% 67.5% 63.1% 66.3%
7 OpenAI GPT-4o (Base) Closed 65.0% 63.5% 56.8% 61.4%

How to Run Reproducible Evaluations

Quick Start with Python 

from datasets import load_dataset

# Load test split
dataset = load_dataset("Coralfil/AboveBoard-Marine-52", split="test")

print(f"Loaded {len(dataset)} evaluation scenarios.")
print("Sample Scenario:", dataset[0]["prompt"])

Command Line Evaluation via Local Script 

python eval_aboveboard.py \
  --model_url "http://localhost:8000/v1" \
  --model_name "Atlantis-Pyramid2:32b" \
  --output_dir "./eval_results"

Citation & Contact 

@dataset{coralfil2026aboveboard,
  author = {Coralfil Research Team},
  title = {AboveBoard: A 52-Scenario Sovereign Canadian Marine Benchmark Suite},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/Coralfil/AboveBoard-Marine-52}}
}
Downloads last month
142
Number of rows:
52
Total file size:
230 kB

Models trained or fine-tuned on Coralfil/AboveBoard-Marine-52

Space using Coralfil/AboveBoard-Marine-52 1

Collection including Coralfil/AboveBoard-Marine-52