v2: charset-recalibrated 42 GB build, replace data/, update docs/notebook

DATA_DICTIONARY.md

@@ -1,6 +1,6 @@
 # code-pii POC dataset: data dictionary
 
-Span-labeled source code for training detectors of **code-related sensitive data** (passwords, API keys, tokens, private keys, emails). Not classic PII (no names/addresses/SSNs). Proof of concept; everything is generated by `generate_dataset.py` and reproducible with `--seed`.
+Version 2 (charset-recalibrated, 40 GB). Span-labeled source code for training detectors of **code-related sensitive data** (passwords, API keys, tokens, private keys, emails). Not classic PII (no names/addresses/SSNs). Proof of concept; everything is generated by `generate_dataset.py` and reproducible with `--seed`.
 
 ## 1. How the dataset is built
 
@@ -10,13 +10,13 @@ Span-labeled source code for training detectors of **code-related sensitive data
 
 **Injection.** For each kept file a coin with `--inject-frac` (0.75) decides whether anything is injected; 85% of those receive 1-5 secrets (count sampled from CredData's secrets-per-file histogram with the "10+" bucket damped) and 0-3 hard negatives. The rest (~36% of records end up with zero secret spans) are pure negatives. For each secret: a label is drawn from the CredData label distribution (plus a fixed 12% share for `email`, which CredData does not annotate); a syntactic context is drawn from CredData's per-label context distribution; a variable name from the per-label vocabulary; a value from the generators below. A language table (31 languages + generic fallback) renders the construct in the file's syntax: assignments (`const x = "..."`, `String x = "..."`, `#define X "..."`, `ENV X=...`, `set(X "...")`...), dictionary/object keys, `.env`/shell exports, comments with the language's comment leader, URL credentials (`postgres://user:PASS@host/db`, span = password only), `Authorization: Bearer/Basic` headers in the language's HTTP idiom, and PEM blocks as heredocs/template literals/concatenated string literals. The rendered line is inserted at a uniformly random line boundary, indented like the following line. Offsets are computed during assembly and asserted (`text[start:end] == value`) for every span before writing.
 
-**Value generators.** `api_key`: AWS access key id (`AKIA`+16), AWS secret (40 base64), Google (`AIza`+35), Stripe (`sk_live_`/`sk_test_`/`rk_live_`+24), SendGrid, Twilio, Mailgun, OpenAI, Anthropic, Azure client secret, and generic hex (32-64), base64 (24-64 raw bytes) and alphanumeric keys (weighted toward generic, as in CredData). `token`: GitHub classic (`ghp_/gho_/ghs_`+36) and fine-grained PATs, Slack tokens and webhooks, JWTs (real base64url header.payload.signature), GitLab, npm, Discord, DigitalOcean, Hugging Face, opaque bearer tokens, Basic-auth base64 of `user:password`. `password`: Faker passwords, word+digits+symbol patterns, short lowercase alphanumerics (length distribution follows CredData: median 9, p90 21). `private_key`: PEM blocks (RSA/EC/DSA/OPENSSH/PKCS8 headers, 64-char base64 lines, realistic lengths). `email`: Faker free/company/domain emails, never `example.*`.
+**Value generators (v2, charset-calibrated).** Values are drawn in two steps: first a target *charset class* (which of lower/upper/digit/symbol occur in the value) is sampled from CredData's per-label charset distribution (`priors.json: charset_by_label`), then a generator is chosen proportionally to its base weight times the empirically profiled probability that it produces that class, and rejection-sampled (<=12 tries) until the value matches. This makes the synthetic charset mix track CredData within ~2 points per class (v1 had 25% plain-hex api keys vs 70% real, and <1% pure-lowercase passwords vs 49% real). `api_key`: AWS access key id (`AKIA`+16), AWS secret (40 base64), Google (`AIza`+35), Stripe (`sk_live_`/`sk_test_`/`rk_live_`+24), SendGrid, Twilio, Mailgun, OpenAI, Anthropic, Azure client secret, and generic hex (32-64, lower or upper case), base64 (24-64 raw bytes), alphanumeric, lowercase-only, uppercase/digit, digit-only and dashed-hex keys. Vendor-prefixed formats end up at ~14% of api keys and ~35% of tokens (CredData's own vendor-specific categories are <10% of its true positives; the synthetic share is kept a little higher on purpose so each format is seen tens of thousands of times). `token`: GitHub classic (`ghp_/gho_/ghs_`+36) and fine-grained PATs, Slack tokens and webhooks, JWTs (real base64url header.payload.signature), GitLab, npm, Discord, DigitalOcean, Hugging Face, opaque bearer tokens, Basic-auth base64 of `user:password`, plus generic hex, alphanumeric, dashed-lowercase and uppercase tokens. `password`: lowercase dictionary words and word pairs (the largest real class), word+digits, word+symbol, Faker strong passwords, capitalised word+digits, digit-only (length distribution follows CredData: median 9, p90 21). `private_key`: PEM blocks (RSA/EC/DSA/OPENSSH/PKCS8 headers, 64-char base64 lines, realistic lengths). `email`: Faker free/company/domain emails, never `example.*`.
 
 **Hard negatives** (label `not_secret`): placeholders (`CHANGEME`, `<YOUR_API_KEY>`, AWS's documented `AKIAIOSFODNN7EXAMPLE`, `ghp_xxx...`, `********`), `example.com`/`test.com`/`localhost` emails, UUIDs, MD5/SHA-1/SHA-256 hex digests, environment-variable references (`os.environ["API_KEY"]`, `${{ secrets.TOKEN }}`...), public keys (PEM `PUBLIC KEY`, `ssh-rsa ...`), base64 of plain text, and version/IP/path/package identifiers. Negatives are rendered with the same constructs but with names that fit the value (`checksum`, `commit`, `request_id`, `public_key`, `version`, ...), except placeholders, example emails and env references, which deliberately use secret-like names (`password = "<YOUR_API_KEY>"`).
 
 **Validation.** `validate.py` re-checks offsets on every span (0 failures on 151k records), tabulates labels/value types/contexts/languages, runs a gitleaks-style regex scanner for 15 value types (100% of injected spans of those types are found exactly at the labeled offsets, i.e. quoting/escaping/indentation did not mangle them), and prints 20 random records for eyeballing.
 
-**Known limitations and biases.** (1) Injected lines are syntactically plausible but semantically disconnected from the surrounding file (a Java constant may land inside a method body, a PEM block inside HTML); a model can learn "line that does not fit" as a cue. (2) Placement is at random line boundaries, not where a human would put config; CredData priors drive *what* and *in which construct*, not *where in the file*. (3) The base corpus is "clean-ish": github-code-clean removed auto-generated files and near-duplicates but was never scrubbed for secrets or PII, so unlabeled real emails, keys or tokens exist in the base text and are implicitly negatives (label noise; eval on `eval_creddata.jsonl` instead of on synthetic splits). (4) Variable-name vocabulary inherits CredData quirks (e.g. `sharedsecret`, `nonce`, `salt` are frequent because a few repos dominate). (5) `generic_hex_key` positives and hash negatives are the same character class; they are disambiguated only by the variable name, which is intentional but means a 40-hex value is not labelable in isolation. (6) Scope is secrets + emails only; names, addresses, phone numbers, SSNs are neither injected nor labeled. (7) Language mix mirrors github-code-clean (Java/JS/HTML/PHP/C heavy; little YAML/JSON/.env, which are where CredData sees most real secrets). (8) CredData values in the eval set are obfuscated by CredData's pipeline (format-preserving), not the originals. (9) Value charsets are not yet calibrated: synthetic `api_key` values are ~25% plain hex versus 70% in CredData, and synthetic passwords are rarely pure-lowercase words versus 49% in CredData (see `notebooks/eda.ipynb`, section 6).
+**Known limitations and biases.** (1) Injected lines are syntactically plausible but semantically disconnected from the surrounding file (a Java constant may land inside a method body, a PEM block inside HTML); a model can learn "line that does not fit" as a cue. (2) Placement is at random line boundaries, not where a human would put config; CredData priors drive *what* and *in which construct*, not *where in the file*. (3) The base corpus is "clean-ish": github-code-clean removed auto-generated files and near-duplicates but was never scrubbed for secrets or PII, so unlabeled real emails, keys or tokens exist in the base text and are implicitly negatives (label noise; eval on `eval_creddata.jsonl` instead of on synthetic splits). (4) Variable-name vocabulary inherits CredData quirks (e.g. `sharedsecret`, `nonce`, `salt` are frequent because a few repos dominate). (5) `generic_hex_key` positives and hash negatives are the same character class; they are disambiguated only by the variable name, which is intentional but means a 40-hex value is not labelable in isolation. (6) Scope is secrets + emails only; names, addresses, phone numbers, SSNs are neither injected nor labeled. (7) Language mix mirrors github-code-clean (Java/JS/HTML/PHP/C heavy; little YAML/JSON/.env, which are where CredData sees most real secrets). (8) CredData values in the eval set are obfuscated by CredData's pipeline (format-preserving), not the originals. (9) Charset calibration (v2) matches CredData's per-label charset classes to within ~2 points but not its exact lengths per class or its vendor mix; classes CredData reaches with <1% mass (`upper`-only keys, `digit+symbol`) are not generated.
 
 ## 2. Record schema (`data/*.parquet`, `sample.jsonl`)
 
@@ -92,6 +92,9 @@ Example record (abridged `text`):
 | api_key | anthropic_api_key | `sk-ant-api03-<93>AA` | `sk-ant-api03-...AA` | same |
 | api_key | azure_client_secret | 40 chars `[A-Za-z0-9~._-]` | `YJYGkd2XGEbq0xHBUYUN9wY7-ts1F1dO0raZXm1s` | same |
 | api_key | generic_hex_key | 32/40/48/64 lowercase hex | `b4a9fa3ca185cc40da7c9edce315ac17f112fbd4` | CredData "Key"/"Secret"/"Salt"/"Nonce" (hex is the dominant charset: 70%) |
+| api_key | generic_hex_key_upper | 32/40/64 uppercase hex | `0A4396E2885F2D6F9965B81B44051234` | CredData `digit+upper` class (8%) |
+| api_key | generic_lower_key / generic_upper_key / generic_digit_key | 16-32 lowercase letters / uppercase+digits / digits | `qzkmwvbtxaplrfhsncyd`, `7KQ2M9XWBP4RZ1TY6LC3`, `4981234509871234` | CredData `lower` (3%), `digit+upper`, `digit` classes |
+| api_key | generic_hex_dashed_key | 3-5 groups of 8 hex joined by `-` | `3fa85f64-5717b4d2-9c1e8a7f` | CredData `digit+lower+symbol` class (2%) |
 | api_key | generic_base64_secret | base64 of 24-64 bytes | `r/Y+kPmu4sQFtSoBF4iSX7XKFgdpVk4QiiAx2H5Lwk8=` | same |
 | api_key | generic_alnum_key | 20-40 alphanumerics | `PU5lRtqn363qL3CA2bH1jYRJ` | same |
 | token | github_pat | `gh[pos]_[A-Za-z0-9]{36}` | `ghp_Ig187tXPStYwqtbfh7hArK9jNHCrkejKHHRL` | CredData "Token"/"Bearer Authorization" (auth header 37%, json 24%) |
@@ -101,8 +104,9 @@ Example record (abridged `text`):
 | token | jwt | `eyJ..\.eyJ..\.<sig>` | `eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOi...` | same |
 | token | gitlab_pat / npm_token / hf_token / digitalocean_pat / discord_bot_token | `glpat-<20>`, `npm_<36>`, `hf_<34>`, `dop_v1_<64hex>`, `<24>.<6>.<27>` | `hf_Lk...` | same |
 | token | bearer_token_opaque | 32-64 base64url | `YJYGkd2XGEbq0xHBUYUN9wY7...` | same |
+| token | generic_hex_token / generic_alnum_token / generic_dashed_token / generic_upper_token | 32-64 hex / 24-48 alnum / `abcd-ef12-...` lowercase groups / 20-32 uppercase+digits | `9f1c...`, `Lk3P...`, `k9x2-7hqa-m1zt-04pd`, `Q7MZ2...` | CredData token charset classes `digit+lower` 17%, `digit+lower+upper` 20%, `digit+lower+symbol` 15%, `digit+upper` 4% |
 | token | basic_auth_b64 | base64(`user:password`) | `ZGVwbG95OkhkNyFrcTJ6` | CredData "Basic Authorization" |
-| password | password | 5-20 chars, Faker/word-based | `dbComputer@1601`, `pecn95` | CredData "Password"/"URL Credentials" (assignment 33%, yaml 22%, comment 11%, url 8%) |
+| password | password | 5-20 chars; charset classes follow CredData (lower 49%, digit+lower 16%, mixed+symbol 10%, ...) | `secretgarden`, `summer2019`, `dbComputer@1601` | CredData "Password"/"URL Credentials" (assignment 33%, yaml 22%, comment 11%, url 8%) |
 | private_key | pem_private_key | `-----BEGIN (RSA\|EC\|DSA\|OPENSSH\|) PRIVATE KEY-----` + base64 lines | see example in `sample.jsonl` | CredData "PEM Private Key" (89% standalone block) |
 | email | email | RFC-ish `local@domain.tld`, never example.* | `stacie16@gardner.com` | not in CredData; fixed mix: assignment 40%, comment 35%, json/yaml 20% |
 | not_secret | placeholder | known dummy values | `AKIAIOSFODNN7EXAMPLE`, `<YOUR_API_KEY>` | secret-like names |
@@ -114,14 +118,14 @@ Example record (abridged `text`):
 | not_secret | base64_non_secret | base64 of an English sentence | `QXVkaWVuY2UgYXJndWUg...` | payload/body names |
 | not_secret | version_or_id | IPv4, semver, package id, path, numeric id | `v7.20.44`, `151.134.43.138` | version/host names |
 
-Label counts for this release (24 shards, 1,810,060 records): api_key 1.01M, password 482k, token 364k, email 249k, private_key 216k secret spans; 1.70M hard negatives split evenly over 11 types. See `stats.md`.
+Label shares are stable across runs (api_key ~43%, password ~21%, token ~16%, email ~11%, private_key ~9%); exact counts for the current release are in `stats.md`.
 
 ## 4. Files
 
 | file | what | size / count |
 |---|---|---|
-| `data/part-000NN.parquet` | generated records, one Parquet file per base-corpus shard, zstd; ~75k records / ~600 MB JSONL / ~160 MB Parquet each | 24 files, shards 0-23: 1,810,060 records, 14.4 GB JSONL, 3.84 GB Parquet |
-| `sample.jsonl` | 500 records reservoir-sampled from shards 0-1, same schema | 500 |
+| `data/train-AAAAA-BBBBB.parquet` | generated records; each Parquet file (zstd) merges base-corpus shards AAAAA..BBBBB (3 shards, ~225k records, ~1.8 GB JSONL, ~480 MB Parquet) | v2: see `stats.md` for the exact file count, records and bytes |
+| `sample.jsonl` | 500 records reservoir-sampled from shards 0-1 of the v2 run, same schema | 500 |
 | `eval_creddata.jsonl` | real CredData lines (+-3 lines context), **not synthetic**; use only for evaluation / threshold tuning, never for training if you want an honest number. `T` rows = positives, `F` rows = hard negatives, `X` rows = ignore. Offsets are code-point based like the synthetic data | 67,896 rows; 15,753 T / 48,061 F / 4,082 X; ~90 MB |
 | `priors.json` | CredData-derived statistics that drive the generator | 12 KB |
 | `stats.md` | counts by label/value_type/context/language, size histogram, throughput | |

README.md

@@ -23,6 +23,7 @@ tags:
 pretty_name: code-pii (POC) - span-labeled secrets in source code
 size_categories:
   - 1M<n<10M
+version: "2.0"
 configs:
   - config_name: default
     data_files:
@@ -30,7 +31,7 @@ configs:
         path: data/*.parquet
 ---
 
-# code-pii (proof of concept)
+# code-pii (v2, charset-recalibrated)
 
 Span-labeled source files for fine-tuning an LLM / token classifier to detect **code-related sensitive data**: passwords, API keys, tokens, private keys and emails. Out of scope: names, addresses, phone numbers, SSNs.
 
@@ -42,7 +43,7 @@ Full field-by-field documentation: [`DATA_DICTIONARY.md`](DATA_DICTIONARY.md). C
 
 | path | description |
 |---|---|
-| `data/part-000NN.parquet` | generated training records (one file per base-corpus shard, ~75k records / ~160 MB each, zstd) |
+| `data/train-AAAAA-BBBBB.parquet` | generated training records; each file merges 3 base-corpus shards (~225k records, ~480 MB, zstd) |
 | `sample.jsonl` | 500 records, same schema, for a quick look |
 | `eval_creddata.jsonl` | 67,896 **real** labeled lines from CredData (+-3 lines of context): 15,753 true secrets, 48,061 annotated false positives, 4,082 ambiguous. Evaluation only. |
 | `priors.json` | CredData-derived placement statistics that drive the generator |
@@ -69,7 +70,7 @@ Offsets are **0-based Unicode code-point offsets into `text`, end-exclusive**: `
 
 `label` in `spans`: `api_key`, `password`, `token`, `private_key`, `email`. In `hard_negatives` it is always `not_secret`.
 
-`value_type` (positives): `aws_access_key_id`, `aws_secret_access_key`, `google_api_key`, `stripe_secret_key`, `sendgrid_api_key`, `twilio_api_key`, `mailgun_api_key`, `openai_api_key`, `anthropic_api_key`, `azure_client_secret`, `generic_hex_key`, `generic_base64_secret`, `generic_alnum_key`, `github_pat`, `github_fine_grained_pat`, `slack_token`, `slack_webhook`, `jwt`, `gitlab_pat`, `npm_token`, `discord_bot_token`, `digitalocean_pat`, `hf_token`, `bearer_token_opaque`, `basic_auth_b64`, `password`, `pem_private_key`, `email`.
+`value_type` (positives): `aws_access_key_id`, `aws_secret_access_key`, `google_api_key`, `stripe_secret_key`, `sendgrid_api_key`, `twilio_api_key`, `mailgun_api_key`, `openai_api_key`, `anthropic_api_key`, `azure_client_secret`, `generic_hex_key`, `generic_hex_key_upper`, `generic_base64_secret`, `generic_alnum_key`, `generic_lower_key`, `generic_upper_key`, `generic_digit_key`, `generic_hex_dashed_key`, `github_pat`, `github_fine_grained_pat`, `slack_token`, `slack_webhook`, `jwt`, `gitlab_pat`, `npm_token`, `discord_bot_token`, `digitalocean_pat`, `hf_token`, `bearer_token_opaque`, `basic_auth_b64`, `generic_hex_token`, `generic_alnum_token`, `generic_dashed_token`, `generic_upper_token`, `password`, `pem_private_key`, `email`.
 `value_type` (hard negatives): `placeholder`, `example_email`, `uuid`, `sha256_hash`, `sha1_git_sha`, `md5_hash`, `env_var_reference`, `public_key_pem`, `ssh_public_key`, `version_or_id`, `base64_non_secret`.
 
 `context` (how the secret was rendered): `assignment`, `json_key`, `yaml_key`, `env_assignment`, `comment`, `url_credential` (span = the password inside `scheme://user:PASS@host`), `auth_header` (span = the token after `Bearer `/`Basic `), `function_arg`, `pem_block`.
@@ -88,7 +89,7 @@ Example (abridged):
 
 1. **Priors** (`build_priors.py`): parse CredData's `meta/*.csv` (67,896 rows with line + column offsets, category, GroundTruth) and the rebuilt, CredData-obfuscated files. Collapse ~200 categories to the label set, classify each true secret's syntactic context by regex on the text before the value, collect variable-name vocabularies, value length/charset stats, quote style, extension frequencies and secrets-per-file histogram into `priors.json`. Emit every labeled line as `eval_creddata.jsonl`.
 2. **Base corpus**: stream `codeparrot/github-code-clean` parquet shards over HTTPS (ungated, no click-through), keep rows whose `license` is MIT/Apache-2.0/BSD/ISC/Unlicense/CC0 (~63%) and 200 B-200 KB.
-3. **Injection** (`generate_dataset.py`): 75% of files get 1-5 secrets (CredData histogram) and 0-3 hard negatives; label, construct and variable name are sampled from the priors; the value comes from a per-type generator; a 31-language syntax table renders the line (assignment, dict/object key, `.env`/export, comment, URL credential, `Authorization` header in the language's HTTP idiom, PEM as heredoc/template literal/string concatenation); the line is inserted at a random line boundary with the neighbor's indentation; offsets are asserted.
+3. **Injection** (`generate_dataset.py`): 75% of files get 1-5 secrets (CredData histogram) and 0-3 hard negatives; label, construct and variable name are sampled from the priors; the value's *charset class* is sampled from CredData's per-label charset distribution and a matching generator is chosen (v2 recalibration, see Changelog); a 31-language syntax table renders the line (assignment, dict/object key, `.env`/export, comment, URL credential, `Authorization` header in the language's HTTP idiom, PEM as heredoc/template literal/string concatenation); the line is inserted at a random line boundary with the neighbor's indentation; offsets are asserted.
 4. **Validation** (`validate.py`): offsets re-checked (0 failures), label tables, regex scanner with gitleaks-style patterns finds 100% of injected spans of 15 pattern-bearing types at the labeled offsets, 20 random records printed for review.
 
 ## Statistics (this release: shards 0-23)
@@ -122,9 +123,16 @@ Real code, real annotations, obfuscated secret values (CredData's pipeline repla
 - Hex key positives vs hash negatives are the same character class and are disambiguated only by variable name, by design.
 - Variable-name vocabulary inherits CredData's repo skew (`sharedsecret`, `nonce`, `salt` are over-represented).
 - Language mix is code-heavy; YAML/JSON/.env/properties files, where CredData sees most real secrets, are under-represented because github-code-clean has few of them.
-- Value charsets are not yet calibrated to CredData: synthetic `api_key` values are ~25% plain hex (CredData: 70%) and synthetic passwords are rarely pure lowercase words (CredData: 49%). See `notebooks/eda.ipynb` section 6; a re-weighting of `GENERATORS`/`gen_password` is the recommended first fix.
+- Charset classes are calibrated to CredData (v2) but value *lengths within a class* and the vendor-prefix mix are generator choices; CredData classes with <1% mass (`upper`-only, `digit+symbol`) are not produced.
 - Secrets + emails only; no names/addresses/SSNs.
 
+## Changelog
+
+| version | date | change |
+|---|---|---|
+| v1 (PoC) | 2026-11-05 | 24 shards, 1.81M records, 14.4 GB JSONL / 3.84 GB Parquet (one 160 MB file per shard). EDA found value charsets diverged from CredData: `api_key` 25% plain hex vs 70% real; passwords <1% pure lowercase vs 49% real; tokens 93% mixed-case+symbol vs 40% real. |
+| v2 | 2026-11-05 | **Charset recalibration**: `gen_value` now samples a target charset class from `priors.json: charset_by_label` and picks a generator by `weight x P(class | generator)` with rejection sampling; new generic generators (`generic_hex_key_upper`, `generic_lower_key`, `generic_upper_key`, `generic_digit_key`, `generic_hex_dashed_key`, `generic_hex_token`, `generic_alnum_token`, `generic_dashed_token`, `generic_upper_token`) and a rewritten `gen_password` with lowercase-word classes. Resulting api_key: 71% hex (prior 70%); password: 51% lowercase (prior 49%); token: 43% mixed+symbol (prior 40%). Full 40 GB build (`--seed 2`), Parquet merged to ~480 MB files (`train-AAAAA-BBBBB.parquet`), `data/` replaced. Docs, notebook and sample regenerated. |
+
 ## Licenses
 
 - Base corpus: [codeparrot/github-code-clean](https://huggingface.co/datasets/codeparrot/github-code-clean) (dataset compilation Apache-2.0; ungated). Only files whose per-file `license` is MIT, Apache-2.0, BSD-2/3-Clause, ISC, Unlicense or CC0-1.0 are included and each record carries that license; attribution requirements of those licenses apply to the `repo`/`path` listed in each record.

data/part-00004.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:fe15ffb281085b66c2723ae3d0ffd71d7300fda93118da937149fa29ca8cddcc
-size 157775245

data/part-00005.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:d448e0966ce90f542b1bf5164b6371ab684a1f8d02e00bb09771fd59f06f157b
-size 162030330

data/part-00006.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:97985a427cac74fa6376fb19cc7fbede08e295510383deda3befe40fbc7b9bb0
-size 156398382

data/part-00007.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:204096160f04334cda5ceb6436cf715c65c5158ac9bd1c98d7d2e1ed4cd5696b
-size 161852997

data/part-00008.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:af55b0c723673aee3770b2037ac76921c00f699202cc56eff716b2ccce944861
-size 156932668

data/part-00009.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:4376745b32430a411c08213094708e92feeee1cca8ffc6e87ab843a23afffb6c
-size 163511052

data/part-00010.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:accf01d2cece6a6b2088b0e67015aa19dcea25fbf7abfb8c42ce19e908b4b02b
-size 157608460

data/part-00011.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a5ca30450827c6a50cbdb0e52145ccd12aadc312f9d0d61e5fd4f16712df8b27
-size 160831338

data/part-00012.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:9024ed29e41c806511f71c2cc09b8751016da2a6e28d8682d9bd03859ba4ab94
-size 157231805

data/part-00013.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:b3329b842594e437f055fae850a4656b2c5b6062f1c3b1411f2dde611779aa91
-size 162674672

data/part-00014.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:96ee1f12e0c6cde22cfd01e9a4abe9a3062b6a795ac828738b49000c942f3874
-size 158412882

data/part-00015.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:1250f2e88e4f6b93eee8a79b74093e3fdb76bec1a58156b16f2e449ee29d0df3
-size 161960948

data/part-00016.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:d9ca64fe280c8cbde1d840568c890d094ee31750101a5d741949095a19e77b91
-size 157745434

data/part-00017.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:b47cfaf0c82988b58c2809d1d80b513b1dc8bb3330361f8fdebaa8a77908075d
-size 161972403

data/part-00018.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:668956b678c15e4ec290c356b40bbf49acb82ef4f4af3974ca81c5f8b60ee864
-size 155522962

data/part-00019.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:78b27fee3e92decf94cc6fd1af49bebfe93aa8542d1c9f677396fda5e04d16fe
-size 162810732

data/part-00020.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:cc986ba36137aa1c044eed2d8a78799cd484371f07f0f2049343ee51411864f7
-size 157559410

data/part-00021.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:e39380bd2fa952bd064281caeb521187bc5f9300729c3707542106c5f7ea754f
-size 162164664

data/part-00022.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:bfb7aac86dc47df16d471ffdf98407f151868a7fd21caa4b1bbbe9762f4d96af
-size 159624230

data/part-00023.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:8e28bff5b27d4cdada138659f5b9e672b20f2d67bd4559fb59d927c5fac2239c
-size 162676071

data/{part-00000.parquet → train-00000-00002.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c75593310d23f042071faa510bd8fc5190680c72cb709d544671054ac9d01e5
-size 156930226
+oid sha256:1a24dd0c86de207b340b70a2119de017e1fec0934331f0f88bc81b7d524977b4
+size 474202995

data/{part-00001.parquet → train-00003-00005.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd5811de8dfc1f3cdabef07dd84935b3660bfb622ee6776af73b5a5b043d6b3f
-size 161619930
+oid sha256:c560128f7996a2a71d974bf65efab156646f8a5ba474720e2c0ae98b81f3ee8b
+size 480551376

data/{part-00002.parquet → train-00006-00008.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64aeac32fe5d40cba4e548283aea68df33ad820edd98b7d2475a4902878fab2c
-size 156427434
+oid sha256:a4fa5e9bd055f64e224c007025a4e3e5d802945d2f6a62de1905e38debe2d233
+size 473983847

data/{part-00003.parquet → train-00009-00011.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79999ff761b92f4c35418a580c48f59e752711199203889a65218f6c53165345
-size 162741946
+oid sha256:5ef279fe0726eb1a99f0c6a0aa3802b017736bb793b1d37a6eee675322c6519e
+size 480044467

data/train-00012-00014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ee75059c354dee80f2130c3d9d4cc0352362f234eda23c9a4f3063c0f87bd6f
+size 476766732

data/train-00015-00017.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e276953a09c0b2dbfa111d27e726a4788411e3956a67976041a73a0afd248945
+size 480375176

data/train-00018-00020.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:210b10190aebd9d998ad524dba796e191be19045db050676876736b4ab421a09
+size 475610117

data/train-00021-00023.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d0ea10c272a7f17e5a2712b1274d2651a1f9eb3f936c9c1462e432348e61c8e
+size 483569061

data/train-00024-00026.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e38d3902e277ee3228ecba6008f1c5d8fe7255c3df6f805d3181e20cf22e0d6
+size 479051338

data/train-00027-00029.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:baebe3064382dbfcc24403000f86507cc58bd8f65b3afba9ea5e53e52b8729f3
+size 473025807

data/train-00030-00032.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b9670a216ecf104b68f11bb23d6dd71d1ea84f78d93763b7471368e06ddcd84
+size 480399119

data/train-00033-00035.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:860ebf4029f6f3c2db3697e74d371e418ce85bf42bca85ec40ad50b20e56ab96
+size 478168950

data/train-00036-00038.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf55bf569ff38a0879b2f28ac960c1f1f2a6b958eb73b402d0c31a1d182933c5
+size 482926481

data/train-00039-00041.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e471c3589233dcdcad4b19c526cb129503ec3c7fb683609ae834d83b125451fa
+size 473504996

data/train-00042-00044.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f516eded3ddc95fbfb32d188bd06469c0746156da04763cc4981c24ead9a37ab
+size 483011399

data/train-00045-00047.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71c7592f1d3d3c2ba4364b6778502948a531b93704c2d03a73850dd3e8507c4b
+size 478048648

data/train-00048-00050.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c25248b6059f5cca1067ceabcbb7a99a53a3f1f3fec82e1a472921d11dd6b65
+size 478676578

data/train-00051-00053.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:021b16041c3887687560fd568b4217ef6765c905d16c8ceb265c7a8145b7bec5
+size 484807633

data/train-00054-00056.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bccb32fc4e95824377bd31b220b9029a6e48696b72ae4bbbb6cee2b4566c2ba
+size 478119930

data/train-00057-00059.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d417240d9a52ae4f01373401246681129d5351806df4c47d4de1ddcffc61e4fd
+size 475373339

data/train-00060-00062.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab4fbea52bc5764bbcc0f6b1f8bf71fa8302b4473c03081aedbfd470fa558643
+size 479144687

data/train-00063-00065.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c434d6a6afbe6ab4001ed9a74bec84692a04669574924f6fdb0e1ce37553c873
+size 474893789

data/train-00066-00068.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36d7b472a02216fdbb6bfd7bf33f7581bda651562d71bf684e476bad399f9370
+size 484232757

data/train-00069-00069.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0015c31c00e9f8a622111b89fb47d6ac31c563e1d79838843af091017084e4b
+size 158574328

generate_dataset.py

@@ -99,14 +99,24 @@ def gen_jwt():
     return f"{hdr}.{pl}.{b64url(32)}"
 
 def gen_password():
+    """Password sub-generators spanning CredData's charset buckets (lower 49%, digit+lower 16%, mixed 10%, ...).
+    The final mix is set by charset-targeted sampling (see gen_value), not by the branch probabilities here."""
     k = R.random()
-    if k < 0.45:
-        return fake.password(length=R.randint(8, 16), special_chars=R.random() < 0.4)
-    if k < 0.70:
-        return R.choice(["", "", "admin", "root", "dev", "test", "db", "qa"]) + fake.word().capitalize() + R.choice(["", "!", "#", "@"]) + str(R.randint(1, 2024))
-    if k < 0.85:
-        return rs(string.ascii_lowercase + string.digits, R.randint(6, 12))
-    return fake.word() + "_" + fake.word() + str(R.randint(10, 99))
+    if k < 0.30:   # pure lowercase: dictionary words, as in most real leaked passwords
+        return R.choice([fake.word() + fake.word(), fake.word() + R.choice(["pass", "pwd", "admin", "secret", "test", "root"]), fake.user_name().replace(".", "").replace("_", ""), rs(string.ascii_lowercase, R.randint(5, 12))])
+    if k < 0.45:   # lowercase + digits
+        return R.choice([fake.word() + str(R.randint(1, 9999)), rs(string.ascii_lowercase + string.digits, R.randint(6, 12)), fake.word() + str(R.randint(10, 99)) + fake.word()])
+    if k < 0.55:   # lowercase + symbol
+        return R.choice([fake.word() + R.choice(["_", "-", ".", "!", "@"]) + fake.word(), fake.word() + R.choice(["!", "#", "$"])])
+    if k < 0.70:   # mixed case, digits, symbols (Faker-style strong passwords)
+        return fake.password(length=R.randint(8, 16), special_chars=R.random() < 0.6)
+    if k < 0.85:   # capitalised word + digits (+ symbol)
+        return R.choice(["", "", "admin", "root", "dev", "test", "db", "qa"]) + fake.word().capitalize() + R.choice(["", "", "!", "#", "@"]) + str(R.randint(1, 2024))
+    if k < 0.92:
+        return fake.word() + "_" + fake.word() + str(R.randint(10, 99))
+    if k < 0.96:
+        return str(R.randint(10 ** 5, 10 ** 9))                       # digits only
+    return rs(string.ascii_uppercase, R.randint(6, 10)) if R.random() < 0.5 else fake.word().capitalize() + fake.word().capitalize()
 
 def gen_pem():
     kind = R.choice(["RSA PRIVATE KEY", "RSA PRIVATE KEY", "PRIVATE KEY", "EC PRIVATE KEY", "OPENSSH PRIVATE KEY", "DSA PRIVATE KEY"])
@@ -139,8 +149,13 @@ GENERATORS = {
         (1, "anthropic_api_key", lambda: "sk-ant-api03-" + rs(B64URL, 93) + "AA"),
         (1, "azure_client_secret", lambda: rs(B62 + "~._-", 40)),
         (6, "generic_hex_key", lambda: rs(HEX, R.choice([32, 32, 40, 48, 64, 64]))),
+        (2, "generic_hex_key_upper", lambda: rs(HEX.upper(), R.choice([32, 40, 64]))),
         (4, "generic_base64_secret", lambda: b64(R.choice([24, 32, 32, 48, 64]))),
         (2, "generic_alnum_key", lambda: rs(B62, R.choice([20, 24, 32, 40]))),
+        (1, "generic_lower_key", lambda: rs(string.ascii_lowercase, R.choice([16, 20, 24, 32]))),
+        (1, "generic_upper_key", lambda: rs(string.ascii_uppercase + string.digits, R.choice([16, 20, 25, 32]))),
+        (1, "generic_digit_key", lambda: rs(string.digits, R.choice([12, 16, 20, 32]))),
+        (1, "generic_hex_dashed_key", lambda: "-".join(rs(HEX, 8) for _ in range(R.choice([3, 4, 5])))),
     ],
     "token": [
         (3, "github_pat", lambda: R.choice(["ghp_", "gho_", "ghs_"]) + rs(B62, 36)),
@@ -155,6 +170,10 @@ GENERATORS = {
         (1, "hf_token", lambda: "hf_" + rs(B62, 34)),
         (3, "bearer_token_opaque", lambda: rs(B64URL, R.choice([32, 40, 43, 64]))),
         (2, "basic_auth_b64", lambda: base64.b64encode(f"{fake.user_name()}:{gen_password()}".encode()).decode()),
+        (3, "generic_hex_token", lambda: rs(HEX, R.choice([32, 40, 48, 64]))),
+        (2, "generic_alnum_token", lambda: rs(B62, R.choice([24, 32, 40, 48]))),
+        (2, "generic_dashed_token", lambda: "-".join(rs(string.ascii_lowercase + string.digits, R.choice([4, 6, 8])) for _ in range(R.choice([3, 4, 5, 6])))),
+        (1, "generic_upper_token", lambda: rs(string.ascii_uppercase + string.digits, R.choice([20, 24, 32]))),
     ],
     "password": [(1, "password", gen_password)],
     "private_key": [(1, "pem_private_key", gen_pem)],
@@ -170,9 +189,52 @@ def pick_weighted(items):
             return it
     return items[-1]
 
+def charset(v):
+    s = set()
+    if re.search(r"[a-z]", v): s.add("lower")
+    if re.search(r"[A-Z]", v): s.add("upper")
+    if re.search(r"[0-9]", v): s.add("digit")
+    if re.search(r"[^A-Za-z0-9]", v): s.add("symbol")
+    return "+".join(sorted(s))
+
+def _profile_generators(n=300):
+    """Empirical P(charset | generator), used to steer sampling toward CredData's charset mix."""
+    prof = {}
+    for label, gens in GENERATORS.items():
+        for w, vt, fn in gens:
+            c = {}
+            for _ in range(n):
+                cs = charset(fn()); c[cs] = c.get(cs, 0) + 1
+            prof[vt] = {k: v / n for k, v in c.items()}
+    return prof
+
+GEN_PROFILE = None
+CHARSET_TARGET = {}   # label -> {charset: weight}; filled from priors.json by Priors()
+
 def gen_value(label):
-    w, vt, fn = pick_weighted(GENERATORS[label])
-    return fn(), vt
+    """Pick a generator and value.  When a charset target exists for the label (api_key, password, token),
+    first draw a target charset from CredData's distribution, then draw a generator proportionally to
+    base_weight * P(charset | generator) and rejection-sample (<=12 tries) until the value matches."""
+    global GEN_PROFILE
+    gens = GENERATORS[label]
+    tgt = CHARSET_TARGET.get(label)
+    if not tgt:
+        w, vt, fn = pick_weighted(gens)
+        return fn(), vt
+    if GEN_PROFILE is None:
+        GEN_PROFILE = _profile_generators()
+    cs = R.choices(list(tgt), list(tgt.values()))[0]
+    cands = [(w * GEN_PROFILE[vt].get(cs, 0), vt, fn) for w, vt, fn in gens if GEN_PROFILE[vt].get(cs, 0) > 0]
+    if not cands:
+        w, vt, fn = pick_weighted(gens)
+        return fn(), vt
+    w, vt, fn = pick_weighted(cands)
+    v = fn()
+    for _ in range(12):
+        if charset(v) == cs:
+            break
+        v = fn()
+    return v, vt
 
 # Hard negatives: (value_type, fn, label_of_similar_secret)
 HARD_NEGATIVES = [
@@ -208,6 +270,13 @@ class Priors:
         self.names["token"] = dict(self.names["token"]); self.names["token"].update({"token": 50, "access_token": 30, "auth_token": 20, "api_token": 15, "github_token": 10, "slack_token": 8, "bearer": 5})
         self.names["api_key"] = dict(self.names["api_key"]); self.names["api_key"].update({"api_key": 60, "apikey": 30, "secret_key": 30, "aws_secret_access_key": 20, "aws_access_key_id": 20, "client_secret": 20, "private_key": 5, "stripe_key": 5})
         self.names["private_key"] = {"private_key": 30, "key": 20, "rsa_key": 10, "ssh_key": 10, "pem": 5, "signing_key": 5, "server_key": 5}
+        global GEN_PROFILE
+        if GEN_PROFILE is None:
+            GEN_PROFILE = _profile_generators()
+        for l in ("api_key", "password", "token"):
+            d = p.get("charset_by_label", {}).get(l, {})
+            reachable = set().union(*(GEN_PROFILE[vt].keys() for _, vt, _ in GENERATORS[l]))
+            CHARSET_TARGET[l] = {cs: n for cs, n in d.items() if cs in reachable and n > 0}
         h = p.get("secrets_per_file_hist", {"1": 60, "2": 20, "3": 10, "4": 5})
         # CredData's last bucket is "10+", dominated by a few credential-dump files; damp it
         self.nsec = [min(int(k), 5) for k in h]; self.nsec_w = [h[k] / (4 if int(k) >= 10 else 1) for k in h]

notebooks/make_stats_summary.py

@@ -0,0 +1,16 @@
+#!/usr/bin/env python3
+"""Aggregate part-*.stats.json + summary.json of a generated dataset dir into notebooks/stats_summary.json."""
+import json, glob, collections, sys, os
+d = sys.argv[1]; out = sys.argv[2] if len(sys.argv) > 2 else os.path.join(os.path.dirname(__file__), "stats_summary.json")
+C = collections.Counter
+agg = {k: C() for k in ("label", "value_type", "lang", "context")}; tot = C(); per = []
+for p in sorted(glob.glob(os.path.join(d, "part-*.stats.json"))):
+    s = json.load(open(p))
+    for k in agg: agg[k].update(s[k])
+    for k in ("files", "injected_files", "clean_files", "bytes", "spans", "negs", "skipped_license", "skipped_size"): tot[k] += s[k]
+    per.append({k: s[k] for k in ("shard", "files", "bytes", "spans", "negs", "seconds")})
+summ = json.load(open(os.path.join(d, "summary.json")))
+res = {"note": f"Aggregated from per-shard stats emitted by generate_dataset.py for the full run ({summ['shards']} shards).", "summary": summ, "totals": dict(tot),
+       "label": dict(agg["label"]), "value_type": {k: v for k, v in agg["value_type"].items() if not k.startswith("neg:")},
+       "hard_negative_type": {k[4:]: v for k, v in agg["value_type"].items() if k.startswith("neg:")}, "language": dict(agg["lang"]), "context": dict(agg["context"]), "per_shard": per}
+json.dump(res, open(out, "w"), indent=1); print(out, tot["files"], tot["bytes"])

notebooks/stats_summary.json

@@ -1,310 +1,687 @@
 {
- "note": "Aggregated from per-shard stats emitted by generate_dataset.py for the full 24-shard run (shards 0-23).",
+ "note": "Aggregated from per-shard stats emitted by generate_dataset.py for the full run (70 shards).",
  "summary": {
-  "shards": 24,
-  "files": 1810060,
-  "bytes": 14426362622,
-  "spans": 2319190,
-  "hard_negatives": 1696239,
-  "seconds": 217.29999999999998,
-  "workers": 4,
-  "MB_per_s": 66.39
+  "shards": 70,
+  "files": 5278400,
+  "bytes": 42079749385,
+  "spans": 6765485,
+  "hard_negatives": 4947748,
+  "seconds": 709.7,
+  "MB_per_s": 59.3,
+  "workers": 4
  },
  "totals": {
-  "files": 1810060,
-  "injected_files": 1153616,
-  "clean_files": 656444,
-  "bytes": 14426362622,
-  "spans": 2319190,
-  "negs": 1696239,
-  "skipped_license": 1075029,
-  "skipped_size": 161111
+  "files": 5278400,
+  "injected_files": 3365090,
+  "clean_files": 1913310,
+  "bytes": 42079749385,
+  "spans": 6765485,
+  "negs": 4947748,
+  "skipped_license": 3136065,
+  "skipped_size": 470285
  },
  "label": {
-  "private_key": 215959,
-  "password": 482135,
-  "api_key": 1008714,
-  "token": 363752,
-  "email": 248630
+  "api_key": 2941426,
+  "private_key": 630711,
+  "token": 1057949,
+  "password": 1411282,
+  "email": 724117
  },
  "value_type": {
-  "pem_private_key": 215959,
-  "password": 482135,
-  "stripe_secret_key": 72231,
-  "openai_api_key": 35976,
-  "generic_base64_secret": 143999,
-  "generic_hex_key": 215759,
-  "bearer_token_opaque": 51927,
-  "email": 248630,
-  "jwt": 69314,
-  "mailgun_api_key": 36176,
-  "aws_access_key_id": 108440,
-  "twilio_api_key": 36039,
-  "google_api_key": 72035,
-  "generic_alnum_key": 72287,
-  "npm_token": 17265,
-  "basic_auth_b64": 34582,
-  "aws_secret_access_key": 107670,
-  "github_pat": 52267,
-  "sendgrid_api_key": 36048,
-  "gitlab_pat": 17297,
-  "slack_token": 34404,
-  "discord_bot_token": 17510,
-  "slack_webhook": 17266,
-  "github_fine_grained_pat": 17252,
-  "digitalocean_pat": 17313,
-  "azure_client_secret": 35964,
-  "anthropic_api_key": 36090,
-  "hf_token": 17355
+  "generic_hex_key": 2088403,
+  "pem_private_key": 630711,
+  "hf_token": 23999,
+  "generic_hex_key_upper": 84257,
+  "generic_alnum_token": 125593,
+  "generic_hex_token": 188962,
+  "generic_upper_token": 42387,
+  "generic_dashed_token": 108217,
+  "password": 1411282,
+  "google_api_key": 46522,
+  "github_pat": 72223,
+  "stripe_secret_key": 45652,
+  "generic_base64_secret": 84345,
+  "twilio_api_key": 28982,
+  "email": 724117,
+  "slack_token": 47429,
+  "digitalocean_pat": 52983,
+  "jwt": 94858,
+  "basic_auth_b64": 78316,
+  "generic_alnum_key": 70933,
+  "aws_access_key_id": 140211,
+  "generic_upper_key": 42447,
+  "discord_bot_token": 23750,
+  "bearer_token_opaque": 100417,
+  "openai_api_key": 20557,
+  "anthropic_api_key": 20045,
+  "aws_secret_access_key": 68075,
+  "gitlab_pat": 26876,
+  "generic_lower_key": 84658,
+  "slack_webhook": 23980,
+  "generic_digit_key": 14650,
+  "generic_hex_dashed_key": 30003,
+  "mailgun_api_key": 30009,
+  "sendgrid_api_key": 20259,
+  "npm_token": 24231,
+  "github_fine_grained_pat": 23728,
+  "azure_client_secret": 21418
  },
  "hard_negative_type": {
-  "uuid": 154284,
-  "ssh_public_key": 154826,
-  "sha1_git_sha": 153941,
-  "version_or_id": 154375,
-  "public_key_pem": 154348,
-  "example_email": 153763,
-  "base64_non_secret": 153819,
-  "env_var_reference": 153704,
-  "md5_hash": 154226,
-  "sha256_hash": 154820,
-  "placeholder": 154133
+  "public_key_pem": 450917,
+  "example_email": 449706,
+  "sha256_hash": 448904,
+  "base64_non_secret": 449929,
+  "uuid": 450581,
+  "sha1_git_sha": 449891,
+  "env_var_reference": 449976,
+  "ssh_public_key": 449066,
+  "placeholder": 449463,
+  "version_or_id": 448970,
+  "md5_hash": 450345
  },
  "language": {
-  "JavaScript": 210187,
-  "GO": 48747,
-  "C": 144024,
-  "CSS": 26863,
-  "Markdown": 128952,
-  "C++": 94112,
-  "Batchfile": 1844,
-  "Java": 347001,
-  "SQL": 6795,
-  "Python": 116954,
-  "PHP": 154047,
-  "HTML": 195124,
-  "C#": 128866,
-  "Ruby": 83580,
-  "TypeScript": 40648,
-  "Rust": 6296,
-  "Shell": 17885,
-  "Scala": 18369,
-  "CMake": 2478,
-  "TeX": 2930,
-  "Lua": 5003,
-  "Julia": 1242,
-  "Makefile": 6405,
-  "FORTRAN": 1316,
-  "PowerShell": 3105,
-  "Assembly": 720,
-  "Perl": 4947,
-  "Visual Basic": 1969,
-  "Dockerfile": 3153,
-  "Haskell": 6498
+  "JavaScript": 611284,
+  "GO": 142183,
+  "C": 418772,
+  "CSS": 77104,
+  "Markdown": 377703,
+  "C++": 276426,
+  "Batchfile": 5395,
+  "Java": 1013525,
+  "SQL": 19549,
+  "Python": 342346,
+  "PHP": 447622,
+  "HTML": 569348,
+  "C#": 374456,
+  "Ruby": 244789,
+  "TypeScript": 118371,
+  "Rust": 18294,
+  "Shell": 52401,
+  "Scala": 53472,
+  "CMake": 7311,
+  "TeX": 8167,
+  "Lua": 14602,
+  "Julia": 3596,
+  "Makefile": 18329,
+  "FORTRAN": 3693,
+  "PowerShell": 8941,
+  "Assembly": 2144,
+  "Perl": 14562,
+  "Visual Basic": 5753,
+  "Dockerfile": 9249,
+  "Haskell": 19013
  },
  "context": {
-  "pem_block": 216577,
-  "yaml_key": 239092,
-  "assignment": 1091580,
-  "url_credential": 72576,
-  "comment": 191427,
-  "json_key": 258561,
-  "env_assignment": 42691,
-  "auth_header": 159434,
-  "function_arg": 47252
+  "json_key": 754676,
+  "pem_block": 632717,
+  "auth_header": 462377,
+  "assignment": 3184852,
+  "yaml_key": 698655,
+  "comment": 558213,
+  "url_credential": 211787,
+  "function_arg": 137084,
+  "env_assignment": 125124
  },
  "per_shard": [
   {
    "shard": 0,
    "files": 74815,
-   "bytes": 590694516,
-   "spans": 95653,
-   "negs": 70354,
-   "seconds": 26.2
+   "bytes": 590132293,
+   "spans": 95461,
+   "negs": 70511,
+   "seconds": 32.1
   },
   {
    "shard": 1,
    "files": 76521,
-   "bytes": 607625703,
-   "spans": 98151,
-   "negs": 71293,
-   "seconds": 26.5
+   "bytes": 607328364,
+   "spans": 98481,
+   "negs": 71264,
+   "seconds": 33.0
   },
   {
    "shard": 2,
    "files": 74273,
-   "bytes": 588171619,
-   "spans": 95466,
-   "negs": 69760,
-   "seconds": 25.9
+   "bytes": 587502343,
+   "spans": 94997,
+   "negs": 69573,
+   "seconds": 31.8
   },
   {
    "shard": 3,
    "files": 76117,
-   "bytes": 612627702,
-   "spans": 97511,
-   "negs": 71403,
-   "seconds": 26.8
+   "bytes": 611344041,
+   "spans": 97950,
+   "negs": 71405,
+   "seconds": 33.4
   },
   {
    "shard": 4,
    "files": 75057,
-   "bytes": 595213026,
-   "spans": 96511,
-   "negs": 70390,
-   "seconds": 26.1
+   "bytes": 594260184,
+   "spans": 96401,
+   "negs": 70334,
+   "seconds": 31.6
   },
   {
    "shard": 5,
    "files": 76333,
-   "bytes": 610242813,
-   "spans": 98239,
-   "negs": 71773,
-   "seconds": 26.5
+   "bytes": 609450038,
+   "spans": 98172,
+   "negs": 71875,
+   "seconds": 32.6
   },
   {
    "shard": 6,
    "files": 74175,
-   "bytes": 587985954,
-   "spans": 94366,
-   "negs": 69551,
-   "seconds": 25.7
+   "bytes": 587380148,
+   "spans": 95239,
+   "negs": 69560,
+   "seconds": 31.9
   },
   {
    "shard": 7,
    "files": 76218,
-   "bytes": 605199682,
-   "spans": 97763,
-   "negs": 71536,
-   "seconds": 26.3
+   "bytes": 604513085,
+   "spans": 97164,
+   "negs": 71421,
+   "seconds": 32.2
   },
   {
    "shard": 8,
    "files": 74476,
-   "bytes": 591982426,
-   "spans": 95276,
-   "negs": 69713,
-   "seconds": 25.9
+   "bytes": 591230541,
+   "spans": 95493,
+   "negs": 70012,
+   "seconds": 30.7
   },
   {
    "shard": 9,
    "files": 76612,
-   "bytes": 617047081,
-   "spans": 97947,
-   "negs": 71645,
-   "seconds": 27.0
+   "bytes": 616167696,
+   "spans": 98466,
+   "negs": 71443,
+   "seconds": 31.5
   },
   {
    "shard": 10,
    "files": 74578,
-   "bytes": 593557485,
-   "spans": 95816,
-   "negs": 69759,
-   "seconds": 25.9
+   "bytes": 592730222,
+   "spans": 94585,
+   "negs": 70091,
+   "seconds": 30.8
   },
   {
    "shard": 11,
    "files": 76314,
-   "bytes": 599733027,
-   "spans": 97762,
-   "negs": 71858,
-   "seconds": 26.9
+   "bytes": 598231423,
+   "spans": 97160,
+   "negs": 71253,
+   "seconds": 31.2
   },
   {
    "shard": 12,
    "files": 74366,
-   "bytes": 590979194,
-   "spans": 95345,
-   "negs": 69367,
-   "seconds": 26.4
+   "bytes": 590322850,
+   "spans": 95512,
+   "negs": 69961,
+   "seconds": 30.4
   },
   {
    "shard": 13,
    "files": 76371,
-   "bytes": 610640246,
-   "spans": 98079,
-   "negs": 71815,
-   "seconds": 27.3
+   "bytes": 609610505,
+   "spans": 97516,
+   "negs": 71776,
+   "seconds": 31.3
   },
   {
    "shard": 14,
    "files": 74739,
-   "bytes": 596045071,
-   "spans": 95807,
-   "negs": 69986,
-   "seconds": 26.8
+   "bytes": 595168614,
+   "spans": 95780,
+   "negs": 69996,
+   "seconds": 30.8
   },
   {
    "shard": 15,
    "files": 76680,
-   "bytes": 609048185,
-   "spans": 98107,
-   "negs": 71782,
-   "seconds": 27.3
+   "bytes": 608035568,
+   "spans": 98368,
+   "negs": 71774,
+   "seconds": 31.3
   },
   {
    "shard": 16,
    "files": 74304,
-   "bytes": 593641813,
-   "spans": 95509,
-   "negs": 69405,
-   "seconds": 26.3
+   "bytes": 593074785,
+   "spans": 95677,
+   "negs": 69458,
+   "seconds": 31.0
   },
   {
    "shard": 17,
    "files": 76491,
-   "bytes": 608303848,
-   "spans": 98348,
-   "negs": 71545,
-   "seconds": 27.2
+   "bytes": 607856415,
+   "spans": 98786,
+   "negs": 71772,
+   "seconds": 31.8
   },
   {
    "shard": 18,
    "files": 73595,
-   "bytes": 587978743,
-   "spans": 93698,
-   "negs": 68802,
-   "seconds": 26.0
+   "bytes": 587845861,
+   "spans": 94105,
+   "negs": 69206,
+   "seconds": 30.8
   },
   {
    "shard": 19,
    "files": 77065,
-   "bytes": 611322163,
-   "spans": 98188,
-   "negs": 72281,
-   "seconds": 27.6
+   "bytes": 611444809,
+   "spans": 99267,
+   "negs": 72463,
+   "seconds": 31.7
   },
   {
    "shard": 20,
    "files": 73881,
-   "bytes": 592887155,
-   "spans": 94528,
-   "negs": 69262,
-   "seconds": 26.4
+   "bytes": 592401284,
+   "spans": 94943,
+   "negs": 69107,
+   "seconds": 30.4
   },
   {
    "shard": 21,
    "files": 75915,
-   "bytes": 609462360,
-   "spans": 97166,
-   "negs": 71082,
-   "seconds": 26.3
+   "bytes": 609008949,
+   "spans": 97873,
+   "negs": 71138,
+   "seconds": 31.5
   },
   {
    "shard": 22,
    "files": 75145,
-   "bytes": 602208787,
-   "spans": 96099,
-   "negs": 70374,
-   "seconds": 26.4
+   "bytes": 602087055,
+   "spans": 96249,
+   "negs": 70554,
+   "seconds": 30.9
   },
   {
    "shard": 23,
    "files": 76019,
-   "bytes": 613764023,
-   "spans": 97855,
-   "negs": 71503,
-   "seconds": 26.5
+   "bytes": 612602995,
+   "spans": 97029,
+   "negs": 70791,
+   "seconds": 31.3
+  },
+  {
+   "shard": 24,
+   "files": 74954,
+   "bytes": 597697343,
+   "spans": 96405,
+   "negs": 70447,
+   "seconds": 31.0
+  },
+  {
+   "shard": 25,
+   "files": 75824,
+   "bytes": 598667598,
+   "spans": 96853,
+   "negs": 70800,
+   "seconds": 31.2
+  },
+  {
+   "shard": 26,
+   "files": 75079,
+   "bytes": 602335974,
+   "spans": 95583,
+   "negs": 70613,
+   "seconds": 30.5
+  },
+  {
+   "shard": 27,
+   "files": 74615,
+   "bytes": 591009169,
+   "spans": 95506,
+   "negs": 69708,
+   "seconds": 30.7
+  },
+  {
+   "shard": 28,
+   "files": 75692,
+   "bytes": 605563624,
+   "spans": 97319,
+   "negs": 71072,
+   "seconds": 30.9
+  },
+  {
+   "shard": 29,
+   "files": 74409,
+   "bytes": 585883347,
+   "spans": 95215,
+   "negs": 69672,
+   "seconds": 30.1
+  },
+  {
+   "shard": 30,
+   "files": 76437,
+   "bytes": 615154167,
+   "spans": 97399,
+   "negs": 71308,
+   "seconds": 31.1
+  },
+  {
+   "shard": 31,
+   "files": 74451,
+   "bytes": 583150261,
+   "spans": 95144,
+   "negs": 69710,
+   "seconds": 30.5
+  },
+  {
+   "shard": 32,
+   "files": 76588,
+   "bytes": 612105938,
+   "spans": 97796,
+   "negs": 71890,
+   "seconds": 31.7
+  },
+  {
+   "shard": 33,
+   "files": 74324,
+   "bytes": 593802594,
+   "spans": 95171,
+   "negs": 69476,
+   "seconds": 30.7
+  },
+  {
+   "shard": 34,
+   "files": 76786,
+   "bytes": 613622932,
+   "spans": 97776,
+   "negs": 71684,
+   "seconds": 32.0
+  },
+  {
+   "shard": 35,
+   "files": 74677,
+   "bytes": 597110242,
+   "spans": 95188,
+   "negs": 70120,
+   "seconds": 30.5
+  },
+  {
+   "shard": 36,
+   "files": 76331,
+   "bytes": 609250642,
+   "spans": 98390,
+   "negs": 71716,
+   "seconds": 31.8
+  },
+  {
+   "shard": 37,
+   "files": 74762,
+   "bytes": 594955850,
+   "spans": 95836,
+   "negs": 70273,
+   "seconds": 31.1
+  },
+  {
+   "shard": 38,
+   "files": 76982,
+   "bytes": 610517355,
+   "spans": 98714,
+   "negs": 72509,
+   "seconds": 31.3
+  },
+  {
+   "shard": 39,
+   "files": 73696,
+   "bytes": 590028866,
+   "spans": 94736,
+   "negs": 69389,
+   "seconds": 30.6
+  },
+  {
+   "shard": 40,
+   "files": 76805,
+   "bytes": 609629232,
+   "spans": 98720,
+   "negs": 72030,
+   "seconds": 31.3
+  },
+  {
+   "shard": 41,
+   "files": 73550,
+   "bytes": 585326200,
+   "spans": 94365,
+   "negs": 69266,
+   "seconds": 29.9
+  },
+  {
+   "shard": 42,
+   "files": 76799,
+   "bytes": 614933045,
+   "spans": 97931,
+   "negs": 71813,
+   "seconds": 31.4
+  },
+  {
+   "shard": 43,
+   "files": 74288,
+   "bytes": 602217237,
+   "spans": 95507,
+   "negs": 69735,
+   "seconds": 30.8
+  },
+  {
+   "shard": 44,
+   "files": 75757,
+   "bytes": 601638184,
+   "spans": 96573,
+   "negs": 70663,
+   "seconds": 31.0
+  },
+  {
+   "shard": 45,
+   "files": 74949,
+   "bytes": 601525609,
+   "spans": 95777,
+   "negs": 70890,
+   "seconds": 30.9
+  },
+  {
+   "shard": 46,
+   "files": 75268,
+   "bytes": 599280505,
+   "spans": 96724,
+   "negs": 70905,
+   "seconds": 30.7
+  },
+  {
+   "shard": 47,
+   "files": 75489,
+   "bytes": 602931158,
+   "spans": 96879,
+   "negs": 71273,
+   "seconds": 31.5
+  },
+  {
+   "shard": 48,
+   "files": 74391,
+   "bytes": 594166535,
+   "spans": 95069,
+   "negs": 69208,
+   "seconds": 30.7
+  },
+  {
+   "shard": 49,
+   "files": 76698,
+   "bytes": 614015269,
+   "spans": 98694,
+   "negs": 71601,
+   "seconds": 31.9
+  },
+  {
+   "shard": 50,
+   "files": 74692,
+   "bytes": 598207114,
+   "spans": 95501,
+   "negs": 69766,
+   "seconds": 30.6
+  },
+  {
+   "shard": 51,
+   "files": 76585,
+   "bytes": 611624465,
+   "spans": 98134,
+   "negs": 72022,
+   "seconds": 31.8
+  },
+  {
+   "shard": 52,
+   "files": 74806,
+   "bytes": 600973913,
+   "spans": 96834,
+   "negs": 70250,
+   "seconds": 31.4
+  },
+  {
+   "shard": 53,
+   "files": 76631,
+   "bytes": 612841824,
+   "spans": 98559,
+   "negs": 71658,
+   "seconds": 34.5
+  },
+  {
+   "shard": 54,
+   "files": 74826,
+   "bytes": 601986791,
+   "spans": 96578,
+   "negs": 70475,
+   "seconds": 34.8
+  },
+  {
+   "shard": 55,
+   "files": 76376,
+   "bytes": 617386650,
+   "spans": 97124,
+   "negs": 70941,
+   "seconds": 34.7
+  },
+  {
+   "shard": 56,
+   "files": 74130,
+   "bytes": 588567521,
+   "spans": 94804,
+   "negs": 69353,
+   "seconds": 34.3
+  },
+  {
+   "shard": 57,
+   "files": 76842,
+   "bytes": 607987857,
+   "spans": 98898,
+   "negs": 72404,
+   "seconds": 33.4
+  },
+  {
+   "shard": 58,
+   "files": 73896,
+   "bytes": 582690634,
+   "spans": 95103,
+   "negs": 69722,
+   "seconds": 31.6
+  },
+  {
+   "shard": 59,
+   "files": 75786,
+   "bytes": 598450116,
+   "spans": 96643,
+   "negs": 70497,
+   "seconds": 32.3
+  },
+  {
+   "shard": 60,
+   "files": 74908,
+   "bytes": 602315944,
+   "spans": 95900,
+   "negs": 70140,
+   "seconds": 31.8
+  },
+  {
+   "shard": 61,
+   "files": 75366,
+   "bytes": 595927576,
+   "spans": 96952,
+   "negs": 70353,
+   "seconds": 31.8
+  },
+  {
+   "shard": 62,
+   "files": 75704,
+   "bytes": 611601766,
+   "spans": 96670,
+   "negs": 71212,
+   "seconds": 32.6
+  },
+  {
+   "shard": 63,
+   "files": 74327,
+   "bytes": 585714976,
+   "spans": 95412,
+   "negs": 69829,
+   "seconds": 31.7
+  },
+  {
+   "shard": 64,
+   "files": 76282,
+   "bytes": 606543535,
+   "spans": 98099,
+   "negs": 71088,
+   "seconds": 32.6
+  },
+  {
+   "shard": 65,
+   "files": 74785,
+   "bytes": 595159622,
+   "spans": 96001,
+   "negs": 70460,
+   "seconds": 31.9
+  },
+  {
+   "shard": 66,
+   "files": 76346,
+   "bytes": 608772572,
+   "spans": 98414,
+   "negs": 71761,
+   "seconds": 32.6
+  },
+  {
+   "shard": 67,
+   "files": 74886,
+   "bytes": 599251778,
+   "spans": 95727,
+   "negs": 69488,
+   "seconds": 31.8
+  },
+  {
+   "shard": 68,
+   "files": 76642,
+   "bytes": 618080826,
+   "spans": 98144,
+   "negs": 72136,
+   "seconds": 31.9
+  },
+  {
+   "shard": 69,
+   "files": 74923,
+   "bytes": 599414961,
+   "spans": 96044,
+   "negs": 69684,
+   "seconds": 30.8
   }
  ]
 }