stats.md

code-pii-preview: v3 statistics

Cluster-preserving sample: probability 0.36, seed 42, decision by cluster_id over all 70 source shards.

Totals

metric	preview	full v3	ratio
records	1,900,308	5,278,400	0.360
clusters	856,252	2,379,311	0.360
files with >=1 span	1,256,231	3,491,739	0.360
labeled spans	2,853,437	7,936,370	0.360
hard-negative spans	1,781,642	4,950,684	0.360
JSONL bytes (estimated retained records)	17,114,634,270	47,856,733,695	0.358
Parquet files / bytes	8 / 4,284,943,933 (4.28 GB)	24 / 11,915,030,320 (11.92 GB)	0.360

Files

file	records	bytes
data/train-00000-00008.parquet	245,177	551,780,864
data/train-00009-00017.parquet	244,395	551,026,794
data/train-00018-00026.parquet	244,167	550,873,989
data/train-00027-00035.parquet	242,454	545,765,081
data/train-00036-00044.parquet	245,582	553,852,160
data/train-00045-00053.parquet	245,369	554,751,153
data/train-00054-00062.parquet	243,884	551,396,894
data/train-00063-00069.parquet	189,280	425,496,998

Labeled spans by label

label	count	%
api_key	1,058,528	37.10
password	511,694	17.93
token	374,605	13.13
email	338,109	11.85
private_key	227,111	7.96
person_name	208,247	7.30
username	135,143	4.74

Records by language

lang	count	%
Java	363,576	19.13
JavaScript	220,285	11.59
HTML	203,753	10.72
PHP	161,344	8.49
C	151,248	7.96
Markdown	135,663	7.14
C#	135,638	7.14
Python	123,285	6.49
C++	99,979	5.26
Ruby	88,371	4.65
GO	51,280	2.70
TypeScript	42,506	2.24
CSS	27,837	1.46
Scala	19,239	1.01
Shell	19,104	1.01
SQL	7,019	0.37
Haskell	7,004	0.37
Makefile	6,587	0.35
Rust	6,567	0.35
Perl	5,280	0.28
Lua	5,245	0.28
Dockerfile	3,396	0.18
PowerShell	3,172	0.17
TeX	3,028	0.16
CMake	2,621	0.14
Visual Basic	2,051	0.11
Batchfile	1,880	0.10
FORTRAN	1,332	0.07
Julia	1,218	0.06
Assembly	800	0.04

Spans by syntactic context

context	count	%
assignment	1,284,301	45.01
comment	379,575	13.30
json_key	308,980	10.83
yaml_key	290,972	10.20
pem_block	232,895	8.16
auth_header	164,119	5.75
url_credential	99,533	3.49
function_arg	47,986	1.68
env_assignment	45,076	1.58

Spans by value type

value_type	count	%
generic_hex_key	751,741	26.35
password	511,694	17.93
email	338,109	11.85
pem_private_key	227,111	7.96
full_name	157,320	5.51
username	127,385	4.46
generic_hex_token	64,599	2.26
aws_access_key_id	50,570	1.77
generic_alnum_token	42,847	1.50
basic_auth_b64	39,224	1.37
first_name	39,134	1.37
generic_dashed_token	36,850	1.29
bearer_token_opaque	34,277	1.20
jwt	32,711	1.15
generic_lower_key	30,900	1.08
generic_hex_key_upper	30,556	1.07
generic_base64_secret	30,529	1.07
github_pat	24,845	0.87
generic_alnum_key	24,598	0.86
aws_secret_access_key	24,338	0.85
digitalocean_pat	18,206	0.64
google_api_key	16,334	0.57
slack_token	16,328	0.57
stripe_secret_key	16,309	0.57
generic_upper_key	15,636	0.55
generic_upper_token	14,328	0.50
last_name	11,793	0.41
mailgun_api_key	10,846	0.38
twilio_api_key	10,560	0.37
generic_hex_dashed_key	10,559	0.37
gitlab_pat	9,407	0.33
slack_webhook	8,300	0.29
github_fine_grained_pat	8,259	0.29
npm_token	8,144	0.29
hf_token	8,141	0.29
discord_bot_token	8,139	0.29
handle	7,758	0.27
azure_client_secret	7,702	0.27
sendgrid_api_key	7,674	0.27
openai_api_key	7,329	0.26
anthropic_api_key	7,127	0.25
generic_digit_key	5,220	0.18

Validation

• Sampling is by cluster_id; no cluster is split.
• Parent validate.py reports zero offset/entity/replacement/format problems on the regenerated v3 data.
• The label, language and context distributions track the full set in expectation under uniform cluster sampling.