Datasetpublic

idirectships/abacus-cheat-tell-eval-v4

ABACUS Cheat-Tell Eval v4 — 1000-Row Real Prose Surgical Anachronism Why v4 Exists: Signal-Floor Argument v3 (175 rows, 140 train) was insufficient for ModernBERT fine-tuning. ModernBERT requires ≥800 train rows for a two-class surgical detection task. v4 scales to ~1000 rows (800 train / 200 eval) using real pre-modern translated prose instead of v3's synthesized templates. v4 vs v3 Changes Dimension v3 (175 rows) v4 (~1000 rows) Source… See the full description on the dataset page: https://huggingface.co/datasets/idirectships/abacus-cheat-tell-eval-v4.

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes12downloads

Dataset Card

ABACUS Cheat-Tell Eval v4 — 1000-Row Real Prose Surgical Anachronism

Why v4 Exists: Signal-Floor Argument

v3 (175 rows, 140 train) was insufficient for ModernBERT fine-tuning. ModernBERT requires ≥800 train rows for a two-class surgical detection task. v4 scales to ~1000 rows (800 train / 200 eval) using real pre-modern translated prose instead of v3's synthesized templates.

v4 vs v3 Changes

Dimension	v3 (175 rows)	v4 (~1000 rows)
Source prose	Synthesized from TEA templates	Real PD translations
Per-tradition	35 rows	~200 rows
Train split	140 rows	~800 rows
Eval split	35 rows	~200 rows
Source quality	Consistent register	Variable (scholar translation)

Sources

Tradition	Primary Sources
Greek	Perseus DL (Euclid Heath, Archimedes Heath) + Gutenberg PD translations
Islamic	archive.org al-Khwarizmi Rosen 1831 + Ball/Cajori Islamic chapters
Vedic	archive.org Aryabhatiya Clark 1930, Datta-Singh 1935-38, Colebrooke 1817
Chinese	archive.org Smith-Mikami 1914, Mikami 1913
Math	Gutenberg Cajori 1919, Ball 1908, archive.org Smith 1923-25

All sources are public-domain (pre-1928) or verified open-access.

Anachronism Pool (same 20-entry pool as v2/v3)

Gödel incompleteness (1931), Bourbaki formalism (1935/1939), Cohen forcing (1963), Grothendieck schemes (1960) / topos (1963), Langlands program (1967/1970), Wiles FLT proof (1995), Perelman Poincaré (2003) / geometrization (2003), Green-Tao theorem (2004), Tao structure/randomness (2008), ZFC axioms (1930) / choice (1930), category theory Mac Lane-Eilenberg (1945), Ramsey theory (1930).

Schema

Field	Type	Description
row_id	string	`ct-eval-v4-{0001..NNNN}`
sourceteaid	string	Bootstrap TEA ID or fetcher key
tradition	string	greek / islamic / vedic / chinese / math
tradition_full	string	Detailed tradition string
label	string	"authentic" or "anachronism"
label_id	int	0=authentic, 1=anachronism
text	string	400–800 token passage
inserted_anachronism	string or null	Exact insertion text
insertion_position	int or null	Character offset of insertion
originaltexthash	string	SHA-256[:16] of passage
source_url	string	Source URL
translator	string	Translator name
translation_year	string	Translation year
source_tier	string	A / B

Splits

Split	Authentic	Anachronism	Total
train	~400	~400	~800
eval	~100	~100	~200

Usage

python

from datasets import load_dataset
ds = load_dataset("idirectships/abacus-cheat-tell-eval-v4")
print(ds["train"][0])

License

Apache-2.0 — consistent with the ABACUS codebase license.