Datasets:

hahouari
/

SMedDV

Counts are measured from data/synth/v5/labels.jsonl and cross-checked against the fixed validation manifest. Rows in the category and language tables sum to 3,527,958.

Language IDs and split

There are 53 language IDs plus MRZ (mrz). The HF language: metadata uses the valid sr tag for the dataset's internal Latin-script Serbian code sr-Latn; the row table retains sr-Latn.

Code	Language	Train	Validation	Total	Share
`af`	Afrikaans	18,322	754	19,076	0.54%
`ar`	Arabic	1,178,043	49,466	1,227,509	34.79%
`az`	Azerbaijani	15,617	675	16,292	0.46%
`bs`	Bosnian	14,409	648	15,057	0.43%
`ca`	Catalan	19,299	814	20,113	0.57%
`ckb`	Central Kurdish (Sorani)	131,872	5,410	137,282	3.89%
`cs`	Czech	26,486	1,095	27,581	0.78%
`cy`	Welsh	22,349	938	23,287	0.66%
`da`	Danish	13,923	574	14,497	0.41%
`de`	German	15,415	602	16,017	0.45%
`en`	English	15,135	619	15,754	0.45%
`es`	Spanish	15,161	658	15,819	0.45%
`et`	Estonian	15,096	625	15,721	0.45%
`eu`	Basque	13,847	555	14,402	0.41%
`fa`	Persian	359,178	14,899	374,077	10.60%
`fi`	Finnish	23,711	965	24,676	0.70%
`fil`	Filipino	14,499	613	15,112	0.43%
`fr`	French	15,356	655	16,011	0.45%
`ga`	Irish	13,919	591	14,510	0.41%
`gl`	Galician	13,818	578	14,396	0.41%
`he`	Hebrew	294,949	12,295	307,244	8.71%
`hr`	Croatian	13,665	569	14,234	0.40%
`hu`	Hungarian	15,205	684	15,889	0.45%
`id`	Indonesian	13,740	592	14,332	0.41%
`is`	Icelandic	14,176	586	14,762	0.42%
`it`	Italian	13,864	541	14,405	0.41%
`ku`	Kurdish (Kurmanji)	14,015	629	14,644	0.42%
`la`	Latin	14,098	572	14,670	0.42%
`lb`	Luxembourgish	13,930	572	14,502	0.41%
`lt`	Lithuanian	20,372	790	21,162	0.60%
`lv`	Latvian	16,228	686	16,914	0.48%
`mi`	Māori	13,804	564	14,368	0.41%
`mrz`	Machine-readable zone (not a language)	163,212	6,823	170,035	4.82%
`ms`	Malay	13,842	630	14,472	0.41%
`mt`	Maltese	15,640	646	16,286	0.46%
`nl`	Dutch	14,072	579	14,651	0.42%
`no`	Norwegian	16,883	708	17,591	0.50%
`oc`	Occitan	13,850	589	14,439	0.41%
`pl`	Polish	14,000	535	14,535	0.41%
`ps`	Pashto	163,268	6,680	169,948	4.82%
`pt`	Portuguese	14,132	580	14,712	0.42%
`qu`	Quechua	14,488	565	15,053	0.43%
`rm`	Romansh	14,032	580	14,612	0.41%
`ro`	Romanian	15,582	684	16,266	0.46%
`sk`	Slovak	15,515	642	16,157	0.46%
`sl`	Slovenian	13,825	563	14,388	0.41%
`sq`	Albanian	13,908	593	14,501	0.41%
`sr-Latn`	Serbian (Latin script)	14,008	622	14,630	0.41%
`sv`	Swedish	13,975	576	14,551	0.41%
`sw`	Swahili	13,880	559	14,439	0.41%
`tr`	Turkish	14,037	577	14,614	0.41%
`ur`	Urdu	328,668	13,630	342,298	9.70%
`uz`	Uzbek	13,917	567	14,484	0.41%
`vi`	Vietnamese	58,605	2,376	60,981	1.73%

Sample categories

Category	Train	Validation	Total	Share
generic	768,580	32,301	800,881	22.70%
name	727,120	30,146	757,266	21.46%
place	440,161	18,239	458,400	12.99%
date	352,219	14,669	366,888	10.40%
number	352,015	14,643	366,658	10.39%
short	174,953	7,307	182,260	5.17%
mrz	163,212	6,823	170,035	4.82%
authority	146,126	6,080	152,206	4.31%
mixed	145,804	6,115	151,919	4.31%
coverage	116,650	4,795	121,445	3.44%

Logical text lengths

Unicode code points in text, spaces included.

Length	Rows	Share
1-10	1,562,820	44.30%
11-20	1,147,722	32.53%
21-30	357,228	10.13%
31-40	157,867	4.47%
41-50	141,211	4.00%
51-64	161,110	4.57%

long_text: 184,114 rows (5.22%). hard_negative: 60,618 rows (1.72%).

Background groups

Group	Rows	Share
flat	459,050	13.01%
gradient variants	687,899	19.50%
procedural guilloche	458,446	12.99%
noise	458,604	13.00%
CC0 pattern overlays	688,400	19.51%
ambientCG material textures	775,559	21.98%

Font families

Family	Rows	Share
Lateef	250,933	7.11%
Scheherazade New	250,860	7.11%
Amiri	250,431	7.10%
Harmattan	250,394	7.10%
IBM Plex Sans Arabic	245,316	6.95%
Almarai	233,239	6.61%
Liberation Sans	213,905	6.06%
Liberation Serif	213,894	6.06%
OCR-B	170,035	4.82%
Tajawal	163,177	4.63%
DejaVu Sans	127,561	3.62%
Noto Kufi Arabic	99,998	2.83%
Markazi Text	93,258	2.64%
Cairo	93,238	2.64%
David Libre	91,520	2.59%
Noto Sans Arabic	87,316	2.47%
Noto Sans	85,651	2.43%
Noto Naskh Arabic	83,980	2.38%
Roboto	83,309	2.36%
Readex Pro	64,484	1.83%
Noto Serif Hebrew	36,835	1.04%
Assistant	36,541	1.04%
Heebo	36,406	1.03%
Frank Ruhl Libre	36,404	1.03%
Rubik	36,373	1.03%
Patrick Hand	35,812	1.02%
Vazirmatn	31,992	0.91%
Kalam	29,414	0.83%
Aref Ruqaa	23,427	0.66%
Noto Sans Hebrew	18,543	0.53%
Noto Nastaliq Urdu	14,776	0.42%
Gveret Levin	14,622	0.41%
Gulzar	14,295	0.41%
Handlee	10,019	0.28%

Degradation tiers

Tier	Rows	Share
clean	1,058,029	29.99%
heavy	353,171	10.01%
light	1,235,438	35.02%
medium	881,320	24.98%

The separate contrast_clamped flag occurs on 214 rows; it is not a fifth tier.

Character coverage and split reuse

Dictionary characters: 503 (spaces omitted from dict.txt).
Least frequent code point: 800 total; 755 train; 24 validation.
Every one of the 503 non-space dictionary characters is present in the labels.
Character totals are in character-frequency.csv.
63,613/141,118 validation rows (45.08%) have text found in training. Unique strings: 40,406/117,675 (34.34%). This is expected; the split is not text-disjoint.