v1.0.0 camera-ready: keyed HMAC-SHA-256 author pseudonyms (corpus and sample), de-anonymized and validated Croissant, mixed per-source licenses, current datasheet

README.md

@@ -3,7 +3,9 @@ language:
 - en
 - zh
 - ru
-license: cc-by-4.0
+license: other
+license_name: hackersignal-mixed
+license_link: https://huggingface.co/datasets/BenAmpel/HackerSignal/blob/main/data/release/source_release_manifest.json
 size_categories:
 - 1M<n<10M
 task_categories:
@@ -358,13 +360,13 @@ Each record in the main corpus contains:
 
 | Field | Type | Description |
 |-------|------|-------------|
-| `unified_id` | string | SHA-256-derived release identifier |
+| `unified_id` | string | Release identifier (truncated SHA-256 of provenance fields and the keyed author pseudonym) |
 | `source_dataset` | string | Input source file or dataset name |
 | `source_layer` | string | Normalized layer (see above) |
 | `text` | string | UTF-8 post/advisory text (max 8000 chars) |
 | `timestamp` | string | ISO 8601 publication datetime (UTC) |
 | `forum_id` | string | Source identifier |
-| `author_hash` | string | SHA-256 pseudonymized author |
+| `author_hash` | string | Keyed HMAC-SHA-256 author pseudonym (unpublished key, rotated across major versions) |
 | `release_mode` | string | Governance mode (see below) |
 
 ## Release Governance
@@ -419,4 +421,6 @@ Paper: [arXiv:2605.03158](https://arxiv.org/abs/2605.03158) · Code: [github.com
 
 ## License
 
-Code and metadata: CC BY 4.0. Source text retains source-specific terms documented in the governance addendum and release manifest.
+Mixed. Code and metadata: CC BY 4.0. Source text keeps its source's terms, listed per source in `data/release/source_release_manifest.json`: US government public domain (NVD, CISA KEV), CC BY 4.0 (CVEfixes, GitHub Advisory), CC BY-SA 4.0 (ExploitDB-derived records), and dataset-specific or site terms for the rest. Where a source's terms conflict with this tag, the per-source terms govern.
+
+Author pseudonyms changed in version 1.0.0 (keyed HMAC-SHA-256). `author_hash` and `unified_id` values from earlier revisions do not match this release.

data/release/source_release_manifest.json

@@ -1,12 +1,12 @@
 {
   "dataset": "ETG: Exploit Text Graph",
-  "version": "1.0.0-rc1",
+  "version": "1.0.0",
   "doi": "TBD after immutable archive deposit",
-  "maintainer": "Anonymous (see dataset repository issue tracker)",
+  "maintainer": "Benjamin M. Ampel (bampel@gsu.edu); issues via the dataset repository",
   "issue_tracker": "GitHub issues for the accompanying repository",
   "source_manifest": "data/unified_hacker_communities_neurips_manifest.json",
-  "release_file": "data/unified_hacker_communities_neurips_public.jsonl",
-  "release_file_sha256": "bd409acd7ffc37ced54ac8d4808cb0a1339cb7cb8c7dd229e73b32d706805e4f",
+  "release_file": "unified_hacker_communities_neurips_public.jsonl",
+  "release_file_sha256": "f2fd23989993a6b435767c760c0bc11b0351d13bfdbf64de02be79d3e7610576",
   "croissant": "data/etg_croissant.json",
   "datasheet": "docs/datasheet.md",
   "governance": "docs/release_governance.md",
@@ -313,5 +313,6 @@
     "metadata_or_pointer_only": 13,
     "redistributable_text": 5,
     "research_text_with_terms": 9
-  }
-}
+  },
+  "author_pseudonymization": "keyed HMAC-SHA-256, key_id v1 (unpublished build-time key)"
+}

docs/datasheet.md

@@ -7,7 +7,7 @@
 ## Motivation
 
 **For what purpose was the dataset created?**
-HackerSignal was created to support NLP and machine learning research on cybersecurity threat intelligence. Specifically, it enables: (1) classifying text as exploit-relevant or not, (2) linking unstructured forum or advisory text to CVE records, and (3) predicting vulnerability severity from text. No prior *public* dataset links hacker community discourse, exploit databases, vulnerability advisories, and fix commits through a shared CVE identifier space at this scale.
+HackerSignal was created to support NLP and machine learning research on cybersecurity threat intelligence. Specifically, it enables: (1) CVE linkage retrieval — given exploit/advisory evidence text, retrieve the correct CVE description from a 340K-entry NVD corpus; (2) exploit type classification — 8-class vulnerability type prediction (injection, XSS, memory corruption, DoS, file inclusion, auth/access control, RCE, information disclosure) with temporal OOD evaluation; and (3) temporal CVE-disjoint generalization — prospective retrieval where the gold CVE sets in train and test are strictly disjoint. No prior *public* dataset links hacker community discourse, exploit databases, vulnerability advisories, and fix commits through a shared CVE identifier space at this scale.
 
 **Who created the dataset, and on behalf of which entity?**
 Anonymous (redacted for double-blind review). The dataset is an independent research contribution and is not created on behalf of any commercial entity.
@@ -36,16 +36,14 @@ It is a near-complete collection for sources with public APIs or structured feed
 - `text_raw`: source text before release cleaning, where available
 - `timestamp`: ISO 8601 publication datetime (UTC)
 - `forum_id`: source identifier string
-- `author_hash`: SHA-256 hash of source + author identifier (pseudonymised)
+- `author_hash`: keyed HMAC-SHA-256 pseudonym of the source-namespaced author identifier (unpublished key; consistent within a release version)
 - `thread_url`, `thread_title`, `section`: preserved when exposed by the source
 
 **Is there a label or target associated with each instance?**
-The base corpus has no labels. The five benchmark splits add labels:
-- Task 1: binary exploit-relevance label (0/1)
-- Task 2: CVE ID (for quality-controlled retrieval evaluation)
-- Task 3: severity class (0=low, 1=medium, 2=high, 3=critical)
-- Task 4: ternary hacker exploit label (0=non-exploit/noise, 1=vulnerability discussion, 2=actionable exploit)
-- Task 5: actionable exploit-signal label plus CVE ID when applicable
+The base corpus has no labels. The three benchmark splits add labels:
+- Task 1 (CVE-R): gold CVE identifier for retrieval evaluation (metadata-derived linkage to 340K NVD corpus; train <2022, val 2022–2023, test 2024+)
+- Task 2 (ETC): 8-class exploit type label (injection, xss, memory_corruption, dos, file_inclusion, auth_access, rce, info_disclosure) derived from ExploitDB, 0day.today, and HackerOne metadata; train <2020, val 2020–2022, test ≥2023
+- Task 3 (TG): gold CVE identifier for CVE-disjoint retrieval evaluation; splits defined by CVE publication year so that train and test gold CVE sets are strictly disjoint
 
 **Is any information missing from individual instances?**
 All retained rows in the audited release have parseable timestamps, but some timestamps are placeholders or scrape dates when the original publication date was unavailable. Author information is pseudonymised and the original author string is not recoverable. Some imported sources contain garbled characters from source encoding errors.
@@ -58,10 +56,10 @@ Yes. The CVE index creates explicit linkages between instances and CVE records.
 - **Date noise**: Some timestamps reflect scraping date rather than original publication date (flagged in `meta.json` per source).
 - **Language noise**: The corpus is predominantly English, but `deepdarkcti` and several community sources contain multilingual posts (including Russian, Turkish, Persian, Arabic, and Chinese).
 - **Short rows**: The release contains 1,569,263 rows under eight tokens. These are useful for interaction analysis but should usually be filtered for text-only classification and retrieval tasks.
-- **Task-label caveats**: Task 1 labels are weak CVE-linkage labels sampled within source/split cells to remove source-membership confounding; the release includes a source-only diagnostic baseline and a 240-row manual audit packet for human adjudication. Task 2 is quality-controlled to remove unretrievable empty/high-risk rows, but its CVE labels are still metadata-derived rather than manually adjudicated. Task 3 strips explicit severity and CVSS label strings from benchmark input after label extraction; the automated leakage audit reports zero strict sanitized leakage hits. Task 4 uses weak ternary labels derived from CVE metadata and actionability cues, not human exploit-semantics labels. Task 5 positives are the actionable CVE-linked subset of Task 4 and should be interpreted as conservative actionable-signal metadata.
+- **Task-label caveats**: Task 1 (CVE-R) and Task 3 (TG) labels are metadata-derived CVE linkages (source URL or CVE regex extraction), not manually adjudicated ground truth; CVE-index absence is not a manual negative annotation; a 240-row stratified manual audit packet is provided in `data/audits/` for transparency. Task 2 (ETC) labels are derived from structured metadata fields (ExploitDB exploit descriptions and tags, 0day.today exploit type fields, HackerOne vulnerability type prefixes) mapped to an 8-class taxonomy via deterministic rules; they are not human-validated annotations. All three tasks use temporal splits so that evaluation reflects prospective performance on data postdating training.
 
 **Is the dataset self-contained, or does it link to external resources?**
-The governed public release is self-contained for sources whose terms permit text redistribution. Sources with ambiguous or restrictive terms are released as metadata/pointer-only records following `docs/release_governance.md`: provenance, source IDs, timestamps, CVE references, text hashes, and text lengths are retained, while `text` and `text_raw` are withheld. The CVE linkage task (Task 2) uses NVD CVE descriptions as the corpus, which are included in the `nvd_posts.jsonl` file. No live external resources are required at inference time for redistributable-text records.
+The governed public release is self-contained for sources whose terms permit text redistribution. Sources with ambiguous or restrictive terms are released as metadata/pointer-only records following `docs/release_governance.md`: provenance, source IDs, timestamps, CVE references, text hashes, and text lengths are retained, while `text` and `text_raw` are withheld. The CVE linkage retrieval tasks (Tasks 1 and 3) use NVD CVE descriptions as the retrieval corpus, which are included in the `nvd_posts.jsonl` file. No live external resources are required at inference time for redistributable-text records.
 
 **Does the dataset contain data that might be considered confidential?**
 No private communications, medical records, financial records, or government classified information is intentionally included. All text was collected from publicly accessible sources and no authentication was bypassed. Public availability does not eliminate privacy risk, so the release pseudonymises authors, documents source modes, supports takedown/correction requests, and can downgrade terms-ambiguous sources to metadata/pointer-only release.
@@ -87,7 +85,7 @@ Scripts were written and executed by the dataset authors. Collection occurred be
 Collection scripts ran between January 2026 and April 2026. The audited release contains timestamped records spanning 1988 to April 2026, with the bulk of structured NVD records beginning in 2002.
 
 **Were any ethical review processes conducted?**
-The dataset consists of publicly accessible OSINT data, but the release treats public accessibility as insufficient by itself. No personally identifiable information was deliberately collected. Author strings are pseudonymised via SHA-256 hashing. No IRB review was required for collection of public OSINT data under the authors' institution's guidelines; however, researchers who build on HackerSignal for studies involving inferences about individuals should consult their own IRB.
+The dataset consists of publicly accessible OSINT data, but the release treats public accessibility as insufficient by itself. No personally identifiable information was deliberately collected. Author strings are pseudonymised with keyed HMAC-SHA-256 under an unpublished build-time key, rotated across major versions, so hashing candidate usernames does not reproduce a released pseudonym. Residual risk: pseudonyms stay consistent within a source, so one author's posts remain linkable to each other, and post text can still contain self-identifying content (signatures, handles, contact details). No IRB review was required for collection of public OSINT data under the authors' institution's guidelines; however, researchers who build on HackerSignal for studies involving inferences about individuals should consult their own IRB.
 
 **Did the individuals whose data is collected consent to collection?**
 Individuals did not provide study-specific consent. Participants posted to public internet forums or public reporting/advisory channels, and all sources were publicly indexed or exposed through public APIs. The release therefore minimizes identity signals, preserves provenance, and applies source-specific release modes rather than assuming that public visibility alone justifies unrestricted redistribution.
@@ -103,7 +101,7 @@ Individuals did not provide study-specific consent. Participants posted to publi
 - Encoding errors replaced with the Unicode replacement character (U+FFFD)
 - Source provenance normalized into `source_dataset`, `source_layer`, `forum_id`, optional thread/category fields, and release-mode manifests
 - Exact duplicates removed across sources by normalized-text SHA-256 fingerprint
-- Task 3 benchmark inputs strip explicit severity labels, CVSS score lines, and CVSS vectors after labels are extracted
+- Near-duplicates estimated, not removed: 64-bit SimHash over 5-token shingles on a deterministic 3% sample (Hamming distance <= 3) puts about 8.2% of retained rows (~609K) in near-duplicate clusters; users can apply their own threshold
 - Reviewer-facing audit packets and leakage/overlap/quality reports are generated under `data/audits/`
 
 **Was the raw data saved?**
@@ -114,14 +112,12 @@ Yes — the collection and preprocessing scripts are available in the `src/etg/c
 
 **Were any labels created by human annotators?**
 No. All labels are derived automatically:
-- Task 1 labels: within-source CVE linkage (positive if the source CVE index maps the post ID to a CVE; negative if the same source/split has no CVE-index link and no explicit CVE token)
-- Task 2 labels: CVE linkage from existing metadata in quality-controlled source datasets
-- Task 3 labels: severity strings extracted from structured advisory fields via regex
-- Task 4 labels: weak ternary labels from source CVE metadata and exploit/actionability cues
-- Task 5 labels: actionable-signal labels from Task 4 actionable CVE-linked records, with same-split non-actionable negatives
+- Task 1 (CVE-R) labels: metadata-derived CVE linkages from source URL or CVE regex extraction; only low/medium CVE-linkage audit risk sources included (CISA KEV, CVEfixes, HackerOne)
+- Task 2 (ETC) labels: exploit type from structured metadata fields (ExploitDB exploit descriptions and tags, 0day.today exploit type fields, HackerOne vulnerability type prefixes) mapped to an 8-class taxonomy via deterministic rules
+- Task 3 (TG) labels: same CVE linkage derivation as Task 1; splits defined by CVE publication year to enforce strict gold-set disjointness
 
 **What mechanisms were used to validate labels?**
-Task 1: positives inherit the quality of each source's CVE index, and negatives are weak absence labels rather than manual non-exploit judgments. The benchmark balances positives and negatives within retained source/split cells; the source-only diagnostic now falls to chance AUC, which supports removal of the earlier source-membership shortcut. A 240-row stratified packet is prepared for manual adjudication and remains annotation-pending until human labels are filled. A separate LLM-assisted triage pass used two independent judge prompts and reached consensus on 231/240 rows, with 75.32% consensus agreement with the weak benchmark label; these are automated triage labels, not human validation. The two LLM judges had 96.25% raw agreement, Cohen's kappa 0.9188, Krippendorff's nominal alpha 0.9188, and Gwet's AC1 0.9304. Task 2: CVE linkages are provided by source metadata and retained only for source families with low or medium CVE-linkage audit risk (CISA KEV, CVEfixes, DTL Exploits, and HackerOne). The split requires at least eight query tokens and uses titles as fallback query evidence when a body snippet is missing. A 240-row CVE-link packet is prepared for manual precision auditing. The LLM-assisted triage pass reached consensus on 233/240 Task 2 rows and marked 115 consensus rows as correct CVE links; this automated audit is used to filter high-risk source families and prioritize human review rather than as a precision claim. The two LLM judges had 97.08% raw agreement, Cohen's kappa 0.9417, Krippendorff's nominal alpha 0.9418, and Gwet's AC1 0.9417. Task 3: severity strings were extracted from structured advisory headers; records without a parseable severity field were excluded, and explicit severity/CVSS tokens are stripped from model input after extraction. The leakage audit finds 29,836 strict raw hits and zero strict sanitized hits, with 10 warning phrases left for manual review. Task 4: labels are regenerated from transparent weak rules using CVE linkage and actionability cues, and should be treated as benchmark supervision rather than human exploit labels. Task 5: positives are Task 4 actionable records with CVE metadata; retrieval labels are source CVE metadata and inherit the same CVE-linkage caveats as Task 2.
+Task 1 (CVE-R) and Task 3 (TG): CVE linkages are provided by source metadata and retained only for source families with low or medium CVE-linkage audit risk (CISA KEV, CVEfixes, and HackerOne). The split requires at least eight query tokens and uses titles as fallback query evidence when a body snippet is missing. A 240-row CVE-link stratified packet is prepared for manual precision auditing. A LLM-assisted triage pass reached consensus on 233/240 CVE-R rows and marked 115 consensus rows as correct CVE links; this automated audit is used to filter high-risk source families and prioritize human review rather than as a precision claim. The two LLM judges had 97.08% raw agreement, Cohen's kappa 0.9417, Krippendorff's nominal alpha 0.9418, and Gwet's AC1 0.9417. Task 3 (TG) additionally verifies programmatically that |C_train ∩ C_test| = 0 (strict CVE disjointness). Task 2 (ETC): exploit type labels are mapped from structured metadata to an 8-class taxonomy via deterministic rules and validated against a 240-row stratified manual audit packet in `data/audits/`. Source families (ExploitDB, 0day.today, HackerOne) were chosen for their structured metadata quality; records without a parseable exploit type are excluded.
 
 ---
 
@@ -143,9 +139,8 @@ Earlier, smaller versions of the hacker forum data were used in prior work (ISI
 - The forum sources skew toward English-language, Western cybersecurity communities. Threats emerging from non-English communities may be underrepresented.
 - High-volume historical forum and dark-market imports skew toward 2015-2017 or earlier. Newer public sources such as DeepDarkCTI, 0x00sec, HackerSploit, ParrotSec, and Hack The Box extend into 2026 but are smaller production slices.
 - CISA KEV covers only actively exploited vulnerabilities; using KEV alone as a "critical" proxy will over-represent critical labels relative to the true severity distribution.
-- Task 1 should not be interpreted as a fully manually validated exploit-semantics benchmark because labels are weak CVE-linkage labels and CVE-index absence is not a manual negative annotation.
-- Task 2 is easier and higher-quality than an all-source noisy linkage benchmark because empty GitHub Advisory rows and high-risk source families are filtered out; it should be interpreted as a quality-controlled metadata-linkage benchmark, not a manually validated CVE-link ground truth set.
-- Task 4 and Task 5 lean deliberately on hacker-community signals, but labels are weak and actionability is rare; macro-F1, per-class F1, and joint retrieval metrics are more informative than accuracy alone.
+- Task 1 (CVE-R) and Task 3 (TG) should not be interpreted as fully manually validated CVE-linkage benchmarks; labels are metadata-derived and CVE-index absence is not a manual negative annotation. Both tasks are quality-controlled by filtering to low/medium audit-risk source families (CISA KEV, CVEfixes, HackerOne).
+- Task 2 (ETC) labels are derived from structured metadata rather than human annotation; the 8-class taxonomy covers the most common vulnerability types but may not capture rare or ambiguous classes equally well. Macro-F1 is the primary metric to surface per-class difficulty rather than accuracy alone.
 - Some public forum/community sources may be released as metadata/pointer-only records depending on source terms and takedown requests.
 
 **Are there tasks for which the dataset should not be used?**

etg_croissant.json

@@ -2,20 +2,64 @@
   "@context": {
     "@language": "en",
     "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
     "cr": "http://mlcommons.org/croissant/",
-    "sc": "https://schema.org/",
-    "dct": "http://purl.org/dc/terms/",
     "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "md5": "cr:md5",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "regex": "cr:regex",
+    "repeated": "cr:repeated",
+    "replace": "cr:replace",
+    "sc": "https://schema.org/",
+    "separator": "cr:separator",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform",
+    "wd": "https://www.wikidata.org/wiki/",
     "prov": "http://www.w3.org/ns/prov#"
   },
   "@type": "sc:Dataset",
-  "dct:conformsTo": "http://mlcommons.org/croissant/RAI/1.0",
+  "dct:conformsTo": [
+    "http://mlcommons.org/croissant/1.0",
+    "http://mlcommons.org/croissant/RAI/1.0"
+  ],
   "name": "HackerSignal",
-  "description": "HackerSignal is a large-scale, multi-source dataset linking hacker community discourse, exploit databases, vulnerability advisories, and fix commits through a shared CVE identifier space. It spans 64 public forum/source identifiers across eight source layers, covers 1988\u20132026, and supports three benchmark tasks: CVE linkage retrieval, hacker signal detection with CVE grounding, and CVE-disjoint temporal generalization.",
+  "description": "HackerSignal is a large-scale, multi-source dataset linking hacker community discourse, exploit databases, vulnerability advisories, and fix commits through a shared CVE identifier space. It spans 64 public forum/source identifiers across eight source layers, covers 1988–2026, and supports three benchmark tasks: CVE linkage retrieval, hacker signal detection with CVE grounding, and CVE-disjoint temporal generalization.",
   "url": "https://huggingface.co/datasets/BenAmpel/HackerSignal",
   "version": "1.0.0",
-  "license": "https://creativecommons.org/licenses/by/4.0/",
-  "usageInfo": "Code and metadata are CC BY 4.0 unless otherwise noted. Source text follows source-specific release modes documented in docs/release_governance.md; terms-ambiguous sources may be metadata_or_pointer_only.",
+  "license": [
+    "https://creativecommons.org/licenses/by/4.0/",
+    "https://creativecommons.org/licenses/by-sa/4.0/"
+  ],
+  "usageInfo": "Mixed licensing. Code and metadata: CC BY 4.0. Source text keeps its source's terms, recorded per source in data/release/source_release_manifest.json (license_or_terms, release_mode): US government public domain (NVD, CISA KEV); CC BY 4.0 (CVEfixes, GitHub Advisory); CC BY-SA 4.0 (ExploitDB-derived records); dataset-specific or site terms for the remaining sources, which are released as research text with attribution or as metadata/pointer only. Where a source's terms conflict with the repository-level tag, the per-source terms govern.",
   "isAccessibleForFree": true,
   "datePublished": "2026",
   "inLanguage": [
@@ -23,13 +67,19 @@
     "zh",
     "ru"
   ],
-  "citation": "Anonymous. (2026). HackerSignal \u2014 A Multi-Source Dataset for Hacker Community Analysis and CVE Linkage. Proceedings of NeurIPS Evaluations & Datasets Track.",
+  "citation": "Ampel, B. M., & Samtani, S. (2026). HackerSignal: A Large-Scale Multi-Source Dataset Linking Hacker Community Discourse to the CVE Vulnerability Lifecycle. Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track.",
   "creator": [
     {
       "@type": "sc:Person",
-      "name": "Anonymous",
-      "email": "anonymous@example.com",
-      "affiliation": "Anonymous"
+      "name": "Benjamin M. Ampel",
+      "email": "bampel@gsu.edu",
+      "affiliation": "J. Mack Robinson College of Business, Georgia State University"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Sagar Samtani",
+      "email": "ssamtani@iu.edu",
+      "affiliation": "Kelley School of Business, Indiana University"
     }
   ],
   "keywords": [
@@ -95,7 +145,7 @@
     "prov:startedAtTime": "2026-01-01",
     "prov:endedAtTime": "2026-04-30"
   },
-  "rai:dataCollection": "Data was collected through four mechanisms: (1) Public APIs \u2014 NVD (NIST REST), GitHub Advisory Database (GraphQL), CISA KEV (JSON feed), HackerOne (HuggingFace Hub); (2) Public dataset ingestion \u2014 Gayanku (Kaggle), HackForums (Kaggle), DTL Exploits (GitHub/AZSecure), CVEfixes (Zenodo), DeepDarkCTI (GitHub), Kaeli hacker forums (Kaggle), CrackingArena (Kaggle), ZeroDay (GitHub); (3) Web scraping and public forum APIs \u2014 ExploitDB (CSV), Vulnerability-Lab, ZeroScience Lab, 0x00sec, HackerSploit, ParrotSec, Hack The Box, Go4Expert, Full Disclosure mailing list archive, Seebug; (4) Pre-processed datasets from HuggingFace Hub. All sources were publicly accessible at collection time; no authentication was bypassed.",
+  "rai:dataCollection": "Data was collected through four mechanisms: (1) Public APIs — NVD (NIST REST), GitHub Advisory Database (GraphQL), CISA KEV (JSON feed), HackerOne (HuggingFace Hub); (2) Public dataset ingestion — Gayanku (Kaggle), HackForums (Kaggle), DTL Exploits (GitHub/AZSecure), CVEfixes (Zenodo), DeepDarkCTI (GitHub), Kaeli hacker forums (Kaggle), CrackingArena (Kaggle), ZeroDay (GitHub); (3) Web scraping and public forum APIs — ExploitDB (CSV), Vulnerability-Lab, ZeroScience Lab, 0x00sec, HackerSploit, ParrotSec, Hack The Box, Go4Expert, Full Disclosure mailing list archive, Seebug; (4) Pre-processed datasets from HuggingFace Hub. All sources were publicly accessible at collection time; no authentication was bypassed.",
   "rai:dataCollectionType": "Web scraping, API, Surveys/Interviews",
   "rai:dataCollectionTimeframe": {
     "@type": "sc:DateTime",
@@ -105,37 +155,95 @@
   "rai:dataCollectionRawData": "Raw scraped HTML is not distributed. The processed JSONL files are the primary artifact. Collection and preprocessing scripts are available in the src/etg/collectors/ directory of the code repository.",
   "rai:dataCollectionMissingData": "Coverage is limited by public API availability, rate limits, and pagination depth for forum sources. Posts behind authentication walls are excluded. Some timestamps reflect scraping date rather than original publication date. Author information is pseudonymized and not recoverable. Some legacy forum imports do not preserve thread structure.",
   "rai:dataPreprocessingProtocol": "HTML tags stripped from forum post content; HTML entities unescaped; text truncated at 8000 characters; encoding errors replaced with Unicode replacement character (U+FFFD); source provenance normalized into source_dataset, source_layer, forum_id, and optional thread/category fields; exact duplicates removed across sources by normalized-text SHA-256 fingerprint; Task 3 benchmark inputs strip explicit severity labels, CVSS score lines, and CVSS vectors after label extraction; author strings pseudonymized via SHA-256 hashing of source-namespaced author identifiers.",
-  "rai:dataLimitations": "Forum sources skew toward English-language Western cybersecurity communities; threats emerging from non-English communities may be underrepresented. High-volume historical forum imports skew toward 2015\u20132017 or earlier. Task 1 labels are weak CVE-linkage labels, not human exploit-semantics annotations; CVE-index absence is not a manual negative annotation. Task 2 signal detection uses weak labels derived from CVE metadata and actionability cues, not human annotations. Task 3 shares the same label methodology as Task 1 under a stricter CVE-disjoint constraint. Approximately 8.18% near-duplicate content remains after exact deduplication. The corpus contains 1,569,263 rows under eight tokens useful for interaction analysis but not text classification.",
-  "rai:dataBiases": "English-language bias: the corpus is predominantly English with multilingual content (Russian, Turkish, Persian, Arabic, Chinese) limited to deepdarkcti and several community sources. Temporal bias: bulk of structured NVD records begin in 2002; high-volume forum imports cluster around 2015\u20132017. Source selection bias: only publicly indexed or API-accessible sources are included, potentially underrepresenting closed or invitation-only communities. CISA KEV covers only actively exploited vulnerabilities, over-representing critical severity relative to true distribution. Western-centric community selection may underrepresent threat actors operating primarily in non-English forums.",
-  "rai:personalSensitiveInformation": "No private communications, medical records, financial records, or government classified information is intentionally included. All text was collected from publicly accessible sources. Author strings are pseudonymized via SHA-256 hashing of source-namespaced author identifiers; original author strings are not recoverable from the released dataset. Public forum usernames are not distributed. The release supports takedown and correction requests through the repository issue tracker.",
-  "rai:dataUseCases": "Intended uses: defensive cybersecurity research, temporal retrieval, provenance-aware benchmark design, vulnerability lifecycle analysis, exploit prediction and early warning systems, cross-lingual threat intelligence, graph neural network research on exploit\u2013CVE\u2013fix commit knowledge graphs, and pre-training domain-adapted language models for cybersecurity. Prohibited uses: training systems for automated exploit code generation or malware; de-anonymizing forum participants or linking pseudonymous identities outside approved research protocols; operational blocking, law-enforcement, or employment decisions based solely on model outputs; republishing source text from metadata_or_pointer_only sources.",
+  "rai:dataLimitations": "Forum sources skew toward English-language Western cybersecurity communities; threats emerging from non-English communities may be underrepresented. High-volume historical forum imports skew toward 2015–2017 or earlier. Task 1 labels are weak CVE-linkage labels, not human exploit-semantics annotations; CVE-index absence is not a manual negative annotation. Task 2 signal detection uses weak labels derived from CVE metadata and actionability cues, not human annotations. Task 3 shares the same label methodology as Task 1 under a stricter CVE-disjoint constraint. Approximately 8.18% near-duplicate content remains after exact deduplication. The corpus contains 1,569,263 rows under eight tokens useful for interaction analysis but not text classification.",
+  "rai:dataBiases": "English-language bias: the corpus is predominantly English with multilingual content (Russian, Turkish, Persian, Arabic, Chinese) limited to deepdarkcti and several community sources. Temporal bias: bulk of structured NVD records begin in 2002; high-volume forum imports cluster around 2015–2017. Source selection bias: only publicly indexed or API-accessible sources are included, potentially underrepresenting closed or invitation-only communities. CISA KEV covers only actively exploited vulnerabilities, over-representing critical severity relative to true distribution. Western-centric community selection may underrepresent threat actors operating primarily in non-English forums.",
+  "rai:personalSensitiveInformation": "No private communications, medical records, financial records, or government classified information is intentionally included. All text was collected from publicly accessible sources. Author strings are pseudonymized with keyed HMAC-SHA-256 of source-namespaced author identifiers under an unpublished build-time key, rotated across major versions; without the key, hashing candidate usernames does not reproduce a released pseudonym. Pseudonyms stay consistent within a source, so one author's posts remain linkable to each other. Public forum usernames are not distributed. The release supports takedown and correction requests through the repository issue tracker.",
+  "rai:dataUseCases": "Intended uses: defensive cybersecurity research, temporal retrieval, provenance-aware benchmark design, vulnerability lifecycle analysis, exploit prediction and early warning systems, cross-lingual threat intelligence, graph neural network research on exploit–CVE–fix commit knowledge graphs, and pre-training domain-adapted language models for cybersecurity. Prohibited uses: training systems for automated exploit code generation or malware; de-anonymizing forum participants or linking pseudonymous identities outside approved research protocols; operational blocking, law-enforcement, or employment decisions based solely on model outputs; republishing source text from metadata_or_pointer_only sources.",
   "rai:dataSocialImpact": "HackerSignal enables earlier detection of vulnerability exploitation activity by connecting hacker community discourse to the CVE lifecycle. Positive impacts include faster defensive response times and better understanding of exploit emergence patterns. Potential negative impacts include misuse for automated attack generation or targeting of individuals; these are mitigated through responsible-use restrictions, author pseudonymization, source-level release governance, and explicit prohibition of offensive applications in the dataset license.",
   "rai:dataReleaseMaintenancePlan": "Maintained by the dataset authors. Annual updates planned to incorporate new CVE records, CISA KEV entries, and newly published advisories. Versioning follows semantic versioning (MAJOR.MINOR.PATCH). All tagged versions remain available on HuggingFace Hub. Takedown and correction requests acknowledged within 10 business days; disputed text removed in next patch release. Sources that change terms post-release are downgraded to metadata_or_pointer_only until permissions are clarified.",
   "rai:hasSyntheticData": "none",
   "distribution": [
     {
       "@type": "cr:FileObject",
-      "@id": "hf-repo",
-      "name": "HuggingFace Repository",
+      "@id": "repo",
+      "name": "repo",
+      "description": "The HackerSignal Hugging Face dataset repository.",
       "contentUrl": "https://huggingface.co/datasets/BenAmpel/HackerSignal",
-      "encodingFormat": "application/jsonlines",
-      "sha256": "See versioned manifest checksums in the HuggingFace repository."
+      "encodingFormat": "git+https",
+      "sha256": "main"
     },
     {
       "@type": "cr:FileObject",
       "@id": "public-jsonl",
       "name": "unified_hacker_communities_neurips_public.jsonl",
-      "description": "Governed public release file enforcing source-level release modes from data/release/source_release_manifest.json.",
-      "contentUrl": "https://huggingface.co/datasets/BenAmpel/HackerSignal/resolve/main/data/unified_hacker_communities_neurips_public.jsonl",
-      "encodingFormat": "application/jsonlines"
+      "description": "Governed public corpus, one JSON record per line.",
+      "contentUrl": "https://huggingface.co/datasets/BenAmpel/HackerSignal/resolve/main/unified_hacker_communities_neurips_public.jsonl",
+      "encodingFormat": "application/jsonlines",
+      "sha256": "f2fd23989993a6b435767c760c0bc11b0351d13bfdbf64de02be79d3e7610576"
     },
     {
       "@type": "cr:FileObject",
       "@id": "source-manifest",
       "name": "source_release_manifest.json",
-      "description": "Machine-readable source provenance and release-mode manifest with checksums and row counts.",
+      "description": "Per-source license terms, release modes and checksums.",
       "contentUrl": "https://huggingface.co/datasets/BenAmpel/HackerSignal/resolve/main/data/release/source_release_manifest.json",
-      "encodingFormat": "application/json"
+      "encodingFormat": "application/json",
+      "sha256": "d09c94ed2dc45d16848c3442c9539878de4661992b466a77aaf89b920b4c5a17"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "task1-corpus",
+      "name": "task1_corpus.jsonl",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "contentUrl": "benchmark_v2/task1_cve_linkage/corpus.jsonl",
+      "encodingFormat": "application/jsonlines",
+      "sha256": "95c2449b77c2ff019bcef044ab2c77ae42f35aadff5751a5e15ab4e7c174736a"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "task3-corpus",
+      "name": "task3_corpus.jsonl",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "contentUrl": "benchmark_v2/task3_temporal_generalization/corpus.jsonl",
+      "encodingFormat": "application/jsonlines",
+      "sha256": "875266e9e23a9e7862fe31c0d96ef1ed27828390ab34fa31a4f679feca5d60f8"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "task1-splits",
+      "name": "task1-splits",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "encodingFormat": "application/jsonlines",
+      "includes": "benchmark_v2/task1_cve_linkage/[tv]*.jsonl",
+      "description": "task1_cve_linkage train.jsonl, val.jsonl and test.jsonl."
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "task2-splits",
+      "name": "task2-splits",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "encodingFormat": "application/jsonlines",
+      "includes": "benchmark_v2/task2_exploit_type/[tv]*.jsonl",
+      "description": "task2_exploit_type train.jsonl, val.jsonl and test.jsonl."
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "task3-splits",
+      "name": "task3-splits",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "encodingFormat": "application/jsonlines",
+      "includes": "benchmark_v2/task3_temporal_generalization/[tv]*.jsonl",
+      "description": "task3_temporal_generalization train.jsonl, val.jsonl and test.jsonl."
     }
   ],
   "recordSet": [
@@ -143,288 +251,569 @@
       "@type": "cr:RecordSet",
       "@id": "posts",
       "name": "posts",
-      "description": "Forum posts, exploit advisories, vulnerability descriptions, and fix-commit messages from 64 public forum/source identifiers. Each record is one document. 7,447,646 exact-deduplicated rows spanning 1988\u20132026.",
+      "description": "Forum posts, exploit advisories, vulnerability descriptions, and fix-commit messages from 64 public forum/source identifiers. Each record is one document. 7,447,646 exact-deduplicated rows spanning 1988–2026. Each row's governing terms follow its source (source_dataset) in data/release/source_release_manifest.json (license_or_terms).",
       "field": [
         {
           "@type": "cr:Field",
           "@id": "posts/unified_id",
           "name": "unified_id",
-          "description": "SHA-256-derived release identifier.",
-          "dataType": "sc:Text"
+          "description": "Release identifier: truncated SHA-256 of provenance fields and the keyed author pseudonym.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "unified_id"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/source_dataset",
           "name": "source_dataset",
           "description": "Input source file or dataset name.",
-          "dataType": "sc:Text"
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "source_dataset"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/source_layer",
           "name": "source_layer",
           "description": "Normalized layer: hacker_community, exploit_archive, vulnerability_reference, fix_commit_reference, exploit_qa_reference, or exploitation_reference.",
-          "dataType": "sc:Text"
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "source_layer"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/text",
           "name": "text",
           "description": "Full text of the post or advisory (UTF-8, max 8000 chars). Withheld for metadata_or_pointer_only sources.",
-          "dataType": "sc:Text"
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "text"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/text_raw",
           "name": "text_raw",
           "description": "Source text before release cleaning, where available. Withheld for metadata_or_pointer_only sources.",
-          "dataType": "sc:Text"
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "text_raw"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/timestamp",
           "name": "timestamp",
           "description": "Publication date-time (ISO 8601, UTC).",
-          "dataType": "sc:DateTime"
+          "dataType": "sc:DateTime",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "timestamp"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/forum_id",
           "name": "forum_id",
           "description": "Source identifier (e.g. 'exploitdb', 'nvd', 'hackforums').",
-          "dataType": "sc:Text"
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "forum_id"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/author_hash",
           "name": "author_hash",
-          "description": "SHA-256 hash of the source-namespaced author string (pseudonymised).",
-          "dataType": "sc:Text"
+          "description": "Keyed HMAC-SHA-256 pseudonym of the source-namespaced author string (unpublished key, rotated across major versions).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "author_hash"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/release_mode",
           "name": "release_mode",
           "description": "Release-governance mode: redistributable_text, research_text_with_terms, metadata_or_pointer_only, or excluded.",
-          "dataType": "sc:Text"
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "release_mode"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/thread_url",
           "name": "thread_url",
           "description": "Source thread URL when exposed by the source.",
-          "dataType": "sc:URL"
+          "dataType": "sc:URL",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "thread_url"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/thread_title",
           "name": "thread_title",
           "description": "Thread or advisory title when available.",
-          "dataType": "sc:Text"
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "thread_title"
+            }
+          }
         },
         {
           "@type": "cr:Field",
           "@id": "posts/section",
           "name": "section",
           "description": "Forum section or category when available.",
-          "dataType": "sc:Text"
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "public-jsonl"
+            },
+            "extract": {
+              "column": "section"
+            }
+          }
         }
       ]
     },
     {
       "@type": "cr:RecordSet",
-      "@id": "cve_index",
-      "name": "cve_index",
-      "description": "Cross-source CVE linkage index mapping post IDs to CVE identifiers. One row per (post, CVE) pair. 360,004 rows contain CVE identifiers.",
+      "@id": "task1_queries",
+      "name": "task1_queries",
+      "description": "Task 1 (CVE linkage retrieval) queries with their gold CVE.",
       "field": [
         {
           "@type": "cr:Field",
-          "@id": "cve_index/exploit_id",
-          "name": "exploit_id",
-          "description": "Foreign key to posts/unified_id.",
-          "dataType": "sc:Text"
+          "@id": "task1_queries/id",
+          "name": "id",
+          "description": "Query identifier.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task1-splits"
+            },
+            "extract": {
+              "column": "id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "task1_queries/text",
+          "name": "text",
+          "description": "Query text.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task1-splits"
+            },
+            "extract": {
+              "column": "text"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "cve_index/cve_id",
+          "@id": "task1_queries/cve_id",
           "name": "cve_id",
-          "description": "CVE identifier in the form CVE-YYYY-NNNNN.",
-          "dataType": "sc:Text"
+          "description": "Gold CVE identifier.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task1-splits"
+            },
+            "extract": {
+              "column": "cve_id"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "cve_index/exploit_text",
-          "name": "exploit_text",
-          "description": "First 500 characters of the source post text.",
-          "dataType": "sc:Text"
+          "@id": "task1_queries/source",
+          "name": "source",
+          "description": "Source the query was drawn from.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task1-splits"
+            },
+            "extract": {
+              "column": "source"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "cve_index/title",
-          "name": "title",
-          "description": "Advisory or post title.",
-          "dataType": "sc:Text"
+          "@id": "task1_queries/timestamp",
+          "name": "timestamp",
+          "description": "Publication timestamp (ISO 8601).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task1-splits"
+            },
+            "extract": {
+              "column": "timestamp"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "cve_index/published",
-          "name": "published",
-          "description": "Publication date-time (ISO 8601, UTC).",
-          "dataType": "sc:DateTime"
+          "@id": "task1_queries/split",
+          "name": "split",
+          "description": "Split, from the file name (train, val or test).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task1-splits"
+            },
+            "extract": {
+              "fileProperty": "filename"
+            },
+            "transform": {
+              "regex": "^(train|val|test)\\.jsonl$"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "task1_corpus",
+      "name": "task1_corpus",
+      "description": "Task 1 retrieval corpus: 340,536 NVD CVE descriptions.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "task1_corpus/id",
+          "name": "id",
+          "description": "CVE identifier.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task1-corpus"
+            },
+            "extract": {
+              "column": "id"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "cve_index/source",
-          "name": "source",
-          "description": "Source identifier matching posts/forum_id.",
-          "dataType": "sc:Text"
+          "@id": "task1_corpus/text",
+          "name": "text",
+          "description": "NVD description.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "task1-corpus"
+            },
+            "extract": {
+              "column": "text"
+            }
+          }
         }
       ]
     },
     {
       "@type": "cr:RecordSet",
-      "@id": "task1_cve_linkage",
-      "name": "task1_cve_linkage",
-      "description": "Task 1 \u2014 CVE Linkage Retrieval (CVE-R). Cross-source temporally OOD entity grounding: given exploit/advisory evidence text, retrieve the correct NVD CVE entry from a corpus of 340,536 descriptions. Quality-controlled to remove unretrievable rows. 40,698 train / 2,551 val / 550 test.",
+      "@id": "task2_exploit_type",
+      "name": "task2_exploit_type",
+      "description": "Task 2 (exploit type classification, eight classes).",
       "field": [
         {
           "@type": "cr:Field",
-          "@id": "task1/id",
+          "@id": "task2_exploit_type/id",
           "name": "id",
-          "dataType": "sc:Text"
+          "description": "Query identifier.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task2-splits"
+            },
+            "extract": {
+              "column": "id"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task1/text",
+          "@id": "task2_exploit_type/text",
           "name": "text",
-          "dataType": "sc:Text"
+          "description": "Query text.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task2-splits"
+            },
+            "extract": {
+              "column": "text"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task1/cve_id",
-          "name": "cve_id",
+          "@id": "task2_exploit_type/label",
+          "name": "label",
+          "description": "Exploit type label (one of eight classes).",
           "dataType": "sc:Text",
-          "description": "Target CVE identifier in the corpus."
+          "source": {
+            "fileSet": {
+              "@id": "task2-splits"
+            },
+            "extract": {
+              "column": "label"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task1/source",
+          "@id": "task2_exploit_type/source",
           "name": "source",
-          "dataType": "sc:Text"
+          "description": "Source the query was drawn from.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task2-splits"
+            },
+            "extract": {
+              "column": "source"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task1/timestamp",
+          "@id": "task2_exploit_type/timestamp",
           "name": "timestamp",
-          "dataType": "sc:DateTime"
+          "description": "Publication timestamp (ISO 8601).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task2-splits"
+            },
+            "extract": {
+              "column": "timestamp"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task1/split",
+          "@id": "task2_exploit_type/split",
           "name": "split",
+          "description": "Split, from the file name (train, val or test).",
           "dataType": "sc:Text",
-          "description": "train / val / test"
+          "source": {
+            "fileSet": {
+              "@id": "task2-splits"
+            },
+            "extract": {
+              "fileProperty": "filename"
+            },
+            "transform": {
+              "regex": "^(train|val|test)\\.jsonl$"
+            }
+          }
         }
       ]
     },
     {
       "@type": "cr:RecordSet",
-      "@id": "task2_signal_detection",
-      "name": "task2_signal_detection",
-      "description": "Task 2 \u2014 Hacker Signal Detection (HSD). Two-stage: detect actionable exploit signals from hacker community noise (10:1 negative ratio), then ground positives to NVD CVE. 4,202 train / 1,606 val / 407 test.",
+      "@id": "task3_queries",
+      "name": "task3_queries",
+      "description": "Task 3 (CVE-disjoint temporal generalization) queries with their gold CVE.",
       "field": [
         {
           "@type": "cr:Field",
-          "@id": "task2/id",
+          "@id": "task3_queries/id",
           "name": "id",
-          "dataType": "sc:Text"
+          "description": "Query identifier.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task3-splits"
+            },
+            "extract": {
+              "column": "id"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task2/text",
+          "@id": "task3_queries/text",
           "name": "text",
-          "dataType": "sc:Text"
-        },
-        {
-          "@type": "cr:Field",
-          "@id": "task2/actionable_label",
-          "name": "actionable_label",
-          "dataType": "sc:Integer",
-          "description": "0=not_actionable, 1=actionable_exploit_signal"
+          "description": "Query text.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task3-splits"
+            },
+            "extract": {
+              "column": "text"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task2/cve_id",
+          "@id": "task3_queries/cve_id",
           "name": "cve_id",
+          "description": "Gold CVE identifier.",
           "dataType": "sc:Text",
-          "description": "Primary CVE target for actionable records; empty for non-actionable."
+          "source": {
+            "fileSet": {
+              "@id": "task3-splits"
+            },
+            "extract": {
+              "column": "cve_id"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task2/source",
+          "@id": "task3_queries/source",
           "name": "source",
-          "dataType": "sc:Text"
+          "description": "Source the query was drawn from.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task3-splits"
+            },
+            "extract": {
+              "column": "source"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task2/timestamp",
+          "@id": "task3_queries/timestamp",
           "name": "timestamp",
-          "dataType": "sc:DateTime"
+          "description": "Publication timestamp (ISO 8601).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "task3-splits"
+            },
+            "extract": {
+              "column": "timestamp"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task2/split",
+          "@id": "task3_queries/split",
           "name": "split",
+          "description": "Split, from the file name (train, val or test).",
           "dataType": "sc:Text",
-          "description": "train / val / test"
+          "source": {
+            "fileSet": {
+              "@id": "task3-splits"
+            },
+            "extract": {
+              "fileProperty": "filename"
+            },
+            "transform": {
+              "regex": "^(train|val|test)\\.jsonl$"
+            }
+          }
         }
       ]
     },
     {
       "@type": "cr:RecordSet",
-      "@id": "task3_temporal_generalization",
-      "name": "task3_temporal_generalization",
-      "description": "Task 3 \u2014 Temporal Generalization (TG). Identical retrieval formulation to Task 1 but with strict CVE-disjoint constraint: C_train \u2229 C_test = \u2205. Split by CVE publication year (train <2022, test 2024+). 41,630 train / 1,776 val / 393 test.",
+      "@id": "task3_corpus",
+      "name": "task3_corpus",
+      "description": "Task 3 retrieval corpus: NVD CVE descriptions.",
       "field": [
         {
           "@type": "cr:Field",
-          "@id": "task3/id",
-          "name": "id",
-          "dataType": "sc:Text"
-        },
-        {
-          "@type": "cr:Field",
-          "@id": "task3/text",
-          "name": "text",
-          "dataType": "sc:Text"
-        },
-        {
-          "@type": "cr:Field",
-          "@id": "task3/cve_id",
+          "@id": "task3_corpus/cve_id",
           "name": "cve_id",
+          "description": "CVE identifier.",
           "dataType": "sc:Text",
-          "description": "Target CVE identifier (from 2024+ CVE set, disjoint from training CVEs)."
-        },
-        {
-          "@type": "cr:Field",
-          "@id": "task3/source",
-          "name": "source",
-          "dataType": "sc:Text"
-        },
-        {
-          "@type": "cr:Field",
-          "@id": "task3/timestamp",
-          "name": "timestamp",
-          "dataType": "sc:DateTime"
+          "source": {
+            "fileObject": {
+              "@id": "task3-corpus"
+            },
+            "extract": {
+              "column": "cve_id"
+            }
+          }
         },
         {
           "@type": "cr:Field",
-          "@id": "task3/split",
-          "name": "split",
+          "@id": "task3_corpus/text",
+          "name": "text",
+          "description": "NVD description.",
           "dataType": "sc:Text",
-          "description": "train / val / test"
+          "source": {
+            "fileObject": {
+              "@id": "task3-corpus"
+            },
+            "extract": {
+              "column": "text"
+            }
+          }
         }
       ]
     }
-  ]
-}
+  ],
+  "citeAs": "@inproceedings{ampel2026hackersignal, title={{HackerSignal}: A Large-Scale Multi-Source Dataset Linking Hacker Community Discourse to the {CVE} Vulnerability Lifecycle}, author={Ampel, Benjamin M. and Samtani, Sagar}, booktitle={Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track}, year={2026}}"
+}

hackersignal_sample_10k.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9cad9cdbdfdbc67ee4330f00029dc02a6a8cd4db04af815e2aca2de145b76c21
-size 24982654
+oid sha256:d9192a3c67add44c444b258e1de1a7cf2fff73475199dd3c0847d363839f6b17
+size 24989918

unified_hacker_communities_neurips_public.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd409acd7ffc37ced54ac8d4808cb0a1339cb7cb8c7dd229e73b32d706805e4f
-size 12539636055
+oid sha256:f2fd23989993a6b435767c760c0bc11b0351d13bfdbf64de02be79d3e7610576
+size 12564492359