This dataset requires a paid license. Please purchase access at the link below and provide your Hugging Face username. Once purchased, submit an access request on this page. Access is granted manually.

Enterprise RAG and Internal Knowledge Search Benchmark Dataset

This dataset packages a synthetic internal company workspace and an evidence-linked benchmark table into one product for teams building enterprise RAG systems, internal search assistants, knowledge-base copilots, and deep-search evaluation pipelines.

The benchmark is designed around a realistic fictional company, AsteraOps Cloud, with multiple departments, renamed projects, stale roadmaps, support escalations, compliance controls, customer success notes, engineering tickets, email decisions, and changelog entries. The point is not to provide another generic question-answer dataset. The point is to provide a corpus-grounded evaluation set where every question is tied to explicit enterprise-style artifacts and where the benchmark can punish common production failures: retrieving the wrong version of a document, ignoring email decisions, over-trusting stale FAQs, flattening contradictions, or answering confidently when the corpus does not actually contain the requested fact.

The workspace contains four coherent internal programs:

Atlas Assist for evidence-grounded internal search
Nimbus SSO for enterprise SSO and SCIM rollout
Beacon Vault for governed audit exports and legal-hold workflows
Harbor Queue for support escalation and SLA operations

Each query row includes the benchmark question, the gold answer, answerability labels, evidence document identifiers, evidence document titles, exact citation spans, retrieval notes, benchmark task tags, and a JSON bundle of the supporting source documents. The rows also include a compact manifest of the full synthetic workspace so buyers can reconstruct retrieval pools, create subsets, or feed the benchmark into their own evaluation harnesses without reverse-engineering undocumented context.

This matters because enterprise retrieval is usually judged on private corpora that cannot be shared. Buyers often need a realistic benchmark but do not want to expose real Slack exports, customer tickets, policy wikis, or production email. Synthetic benchmark work solves that confidentiality problem, but many public datasets still feel too clean, too academic, or too detached from how internal search fails in practice. This build intentionally adds the kinds of friction teams see in production: renamed programs, owner changes, stale timelines, policy overrides, customer milestones, and unresolved conflicts where the correct system behavior is to abstain or call out ambiguity instead of inventing certainty.

The dataset is useful for several practical workflows:

Compare retrievers on current-versus-stale document selection
Measure whether answer generation cites the right evidence documents
Test no-answer behavior when a likely enterprise detail is missing
Score contradiction handling when old and new artifacts disagree
Benchmark customer-history reconstruction across support and CS notes
Audit policy interpretation against current governance docs
Evaluate deep-search agents that need to gather evidence from 2-8 internal artifacts before answering

The corpus was generated locally with deterministic templates rather than scraped from a real company. Shared fictional people, projects, dates, customer accounts, and control changes are propagated across all artifacts so the workspace stays coherent. Rule-based generation was then used to produce benchmark questions tied to explicit evidence sentences. The benchmark rows were normalized into analysis-ready columns, JSON-serialized bundle fields, consistent answerability labels, and typed metadata that can be loaded directly in pandas, notebooks, BI tools, or custom evaluation services.

This is easier than building a benchmark manually from scratch. Buyers do not need to author a fictional company graph, write dozens of enterprise artifacts, create query-answer pairs, hand-link evidence docs, generate citation spans, or add no-answer and contradiction cases on their own. The free sample proves the schema, artifact style, evidence linking, and evaluation framing. The full dataset adds the complete benchmark table, the full synthetic workspace manifest, more cross-project reasoning rows, and enough coverage to test retrieval, citation grounding, and abstention behavior without starting from a blank page.

Limitations still apply. The workspace is synthetic, not a replay of a real company. The benchmark focuses on text-based internal artifacts rather than screenshots or binary documents. It is meant for evaluation and benchmarking, not for claiming guaranteed production performance. Teams should still adapt the benchmark to their own security model, data model, and retrieval stack. Even with those limits, it gives enterprise AI builders a reusable internal-knowledge benchmark that is much closer to day-to-day enterprise search than a generic public QA set.

Dataset Preview

query_id	query_text	gold_answer	answerability	reasoning_type	difficulty	evidence_doc_ids	evidence_doc_titles	citation_spans	evidence_summary	judge_rubric	temporal_scope	conflict_flag	stale_document_flag	no_answer_flag	department_scope	project_scope	source_count	expected_source_count	primary_artifact_type	supporting_artifact_types	company_persona	synthetic_company_name	created_date	updated_date	document_time_window	query_category	evaluation_split	benchmark_task	retrieval_notes	failure_mode_targeted	answer_format	source_document_titles	source_document_types	source_document_bundle_json	synthetic_workspace_bundle_json	provenance_note	license_status	synthetic_generation_method	quality_review_status	source_domain	corpus_document_count	query_intent
q-0001	If an older note still says Compass Assist, what is the current project name?	The current project name is Atlas Assist.	answerable	stale_vs_current_resolution	medium	["ATL-001", "ATL-004"]	["Atlas Assist overview wiki", "Atlas Assist rename slack thread"]	[{"char_end": 110, "char_start": 0, "doc_id": "ATL-001", "quote": "Atlas Assist is the current progr... [308 more chars]	Use Atlas Assist overview wiki, Atlas Assist rename slack thread to verify the answer.	Answer must stay grounded in cited evidence documents. Use the expected short_text format. Prefer th... [48 more chars]	historical_and_current	False	True	False	Operations, Product	Atlas Assist	2	2	wiki	["slack"]	AsteraOps Cloud is a fictional B2B workflow software company with product, engineering, support, sec... [55 more chars]	AsteraOps Cloud	2025-08-19	2025-09-02	2025-08-19 to 2025-09-02	alias_resolution	train	enterprise_rag_internal_knowledge_evaluation	Requires resolving a stale alias against the current wiki and rename thread.	stale_alias_confusion	short_text	["Atlas Assist overview wiki", "Atlas Assist rename slack thread"]	["wiki", "slack"]	[{"artifact_type": "wiki", "created_date": "2025-09-02", "department": "Product", "doc_id": "ATL-001... [1,261 more chars]	[{"artifact_type": "wiki", "created_date": "2025-09-02", "department": "Product", "doc_id": "ATL-001... [20,065 more chars]	Deterministic synthetic company workspace generated locally from templated project facts. No real co... [23 more chars]	Synthetic, rights-cleared content suitable for redistribution under CC-BY-4.0.	Rule-based corpus generation with shared fictional people, projects, stale documents, and evidence-l... [26 more chars]	rule_checked_and_type_checked	synthetic.enterprise-workspace.local	52	Resolve renamed project aliases inside enterprise search results.
q-0002	Which documents should be cited to prove that Atlas Assist replaced Compass Assist?	Cite the Atlas Assist overview wiki and the Atlas Assist rename slack thread.	answerable	evidence_citation	medium	["ATL-001", "ATL-004"]	["Atlas Assist overview wiki", "Atlas Assist rename slack thread"]	[{"char_end": 181, "char_start": 111, "doc_id": "ATL-001", "quote": "Older references to Compass Ass... [270 more chars]	Use Atlas Assist overview wiki, Atlas Assist rename slack thread to verify the answer.	Answer must stay grounded in cited evidence documents. Use the expected list format. Prefer the fres... [42 more chars]	historical_and_current	False	True	False	Operations, Product	Atlas Assist	2	2	wiki	["slack"]	AsteraOps Cloud is a fictional B2B workflow software company with product, engineering, support, sec... [55 more chars]	AsteraOps Cloud	2025-08-19	2025-09-02	2025-08-19 to 2025-09-02	evidence_citation	train	enterprise_rag_internal_knowledge_evaluation	The answer must identify both documents rather than only restating the new name.	weak_citation_grounding	list	["Atlas Assist overview wiki", "Atlas Assist rename slack thread"]	["wiki", "slack"]	[{"artifact_type": "wiki", "created_date": "2025-09-02", "department": "Product", "doc_id": "ATL-001... [1,261 more chars]	[{"artifact_type": "wiki", "created_date": "2025-09-02", "department": "Product", "doc_id": "ATL-001... [20,065 more chars]	Deterministic synthetic company workspace generated locally from templated project facts. No real co... [23 more chars]	Synthetic, rights-cleared content suitable for redistribution under CC-BY-4.0.	Rule-based corpus generation with shared fictional people, projects, stale documents, and evidence-l... [26 more chars]	rule_checked_and_type_checked	synthetic.enterprise-workspace.local	52	Test whether the system can cite the right alias-resolution evidence.
q-0003	Who currently owns Atlas Assist?	Maya Chen currently owns Atlas Assist.	answerable	single_document_lookup	easy	["ATL-001"]	["Atlas Assist overview wiki"]	[{"char_end": 230, "char_start": 182, "doc_id": "ATL-001", "quote": "Maya Chen is the current owner... [59 more chars]	Use Atlas Assist overview wiki to verify the answer.	Answer must stay grounded in cited evidence documents. Use the expected short_text format.	current	False	False	False	Product	Atlas Assist	1	1	wiki	[]	AsteraOps Cloud is a fictional B2B workflow software company with product, engineering, support, sec... [55 more chars]	AsteraOps Cloud	2025-09-02	2025-09-02	2025-09-02 to 2025-09-02	ownership_lookup	train	enterprise_rag_internal_knowledge_evaluation	Straight current-state ownership lookup.	owner_misattribution	short_text	["Atlas Assist overview wiki"]	["wiki"]	[{"artifact_type": "wiki", "created_date": "2025-09-02", "department": "Product", "doc_id": "ATL-001... [658 more chars]	[{"artifact_type": "wiki", "created_date": "2025-09-02", "department": "Product", "doc_id": "ATL-001... [20,065 more chars]	Deterministic synthetic company workspace generated locally from templated project facts. No real co... [23 more chars]	Synthetic, rights-cleared content suitable for redistribution under CC-BY-4.0.	Rule-based corpus generation with shared fictional people, projects, stale documents, and evidence-l... [26 more chars]	rule_checked_and_type_checked	synthetic.enterprise-workspace.local	52	Identify the latest accountable owner from current docs.
q-0004	What is the current target date for Atlas Assist?	The current target date for Atlas Assist is 2026-03-12.	answerable	single_document_lookup	easy	["ATL-003"]	["Atlas Assist roadmap current"]	[{"char_end": 97, "char_start": 59, "doc_id": "ATL-003", "quote": "The current target date is 2026-0... [49 more chars]	Use Atlas Assist roadmap current to verify the answer.	Answer must stay grounded in cited evidence documents. Use the expected date format.	current	False	False	False	Product	Atlas Assist	1	1	roadmap	[]	AsteraOps Cloud is a fictional B2B workflow software company with product, engineering, support, sec... [55 more chars]	AsteraOps Cloud	2026-01-15	2026-01-15	2026-01-15 to 2026-01-15	deadline_lookup	train	enterprise_rag_internal_knowledge_evaluation	Current roadmap lookup.	date_lookup_error	date	["Atlas Assist roadmap current"]	["roadmap"]	[{"artifact_type": "roadmap", "created_date": "2026-01-15", "department": "Product", "doc_id": "ATL-... [506 more chars]	[{"artifact_type": "wiki", "created_date": "2025-09-02", "department": "Product", "doc_id": "ATL-001... [20,065 more chars]	Deterministic synthetic company workspace generated locally from templated project facts. No real co... [23 more chars]	Synthetic, rights-cleared content suitable for redistribution under CC-BY-4.0.	Rule-based corpus generation with shared fictional people, projects, stale documents, and evidence-l... [26 more chars]	rule_checked_and_type_checked	synthetic.enterprise-workspace.local	52	Recover the latest committed date from a roadmap.
q-0005	Why did the target date for Atlas Assist move?	The target moved because the citation drawer and stale-document shield were not ready for broad rele... [4 more chars]	answerable	meeting_to_roadmap_reconciliation	medium	["ATL-003", "ATL-005"]	["Atlas Assist roadmap current", "Atlas Assist meeting notes"]	[{"char_end": 204, "char_start": 98, "doc_id": "ATL-003", "quote": "The schedule moved because the c... [354 more chars]	Use Atlas Assist roadmap current, Atlas Assist meeting notes to verify the answer.	Answer must stay grounded in cited evidence documents. Use the expected short_text format.	current	False	False	False	Product	Atlas Assist	2	2	roadmap	["meeting_notes"]	AsteraOps Cloud is a fictional B2B workflow software company with product, engineering, support, sec... [55 more chars]	AsteraOps Cloud	2026-01-09	2026-01-15	2026-01-09 to 2026-01-15	schedule_reasoning	train	enterprise_rag_internal_knowledge_evaluation	Requires linking current roadmap language with meeting notes that explain the slip.	multi_doc_reasoning_gap	short_text	["Atlas Assist roadmap current", "Atlas Assist meeting notes"]	["roadmap", "meeting_notes"]	[{"artifact_type": "roadmap", "created_date": "2026-01-15", "department": "Product", "doc_id": "ATL-... [1,113 more chars]	[{"artifact_type": "wiki", "created_date": "2025-09-02", "department": "Product", "doc_id": "ATL-001... [20,065 more chars]	Deterministic synthetic company workspace generated locally from templated project facts. No real co... [23 more chars]	Synthetic, rights-cleared content suitable for redistribution under CC-BY-4.0.	Rule-based corpus generation with shared fictional people, projects, stale documents, and evidence-l... [26 more chars]	rule_checked_and_type_checked	synthetic.enterprise-workspace.local	52	Explain schedule slips from enterprise project artifacts.

Access Requirements (Paid Dataset)

This dataset is behind manual gated access.

To obtain access:

Purchase the dataset here:
https://thearticulated.gumroad.com/l/enterprise-rag-internal-knowledge-search-benchmark
Provide your Hugging Face username at checkout.
Return to this Hugging Face page and click:
"Request Access"
Your access will be approved within 1-12 hours.

Once approved, you can use the Python snippet at the bottom of this README to load the dataset.

Dataset Structure

Total rows: 104

Total columns: 43

Splits

data: 104 rows

Data Files

data: data/data.parquet

Data Dictionary

The table below describes the columns included in this dataset.

column	pandas_dtype	dataset_type	description
query_id	object	string	Text column. Stable benchmark query identifier.
query_text	object	string	Text column. User-style benchmark question grounded in the synthetic enterprise corpus.
gold_answer	object	string	Text column. Auditable target answer, including no-answer or conflict language when appropriate.
answerability	object	string	Text column. Label indicating whether the query is answerable, no-answer, or conflicting.
reasoning_type	object	string	Text column. Primary reasoning pattern required to answer the query.
difficulty	object	string	Text column. Heuristic difficulty label based on evidence count and conflict handling.
evidence_doc_ids	object	string	Text column. JSON array of supporting document identifiers.
evidence_doc_titles	object	string	Text column. JSON array of supporting document titles.
citation_spans	object	long text	Free-text column containing longer text values. JSON array of evidence quote spans with character offsets.
evidence_summary	object	string	Text column. Plain-English summary of why the cited documents support the answer.
judge_rubric	object	long text	Free-text column containing longer text values. Compact grading rubric for exactness, grounding, and abstention behavior.
temporal_scope	object	string	Text column. Whether the query requires current-state, historical, or mixed temporal reasoning.
conflict_flag	boolean	boolean	True/false column. Boolean flag indicating the query includes contradictory or competing evidence.
stale_document_flag	boolean	boolean	True/false column. Boolean flag indicating stale-document handling is part of the task.
no_answer_flag	boolean	boolean	True/false column. Boolean flag indicating the correct behavior is to abstain or report missing information.
department_scope	object	string	Text column. Comma-separated departments involved in the evidence set.
project_scope	object	string	Text column. Project or cross-project scope of the query.
source_count	Int64	integer	Whole-number numeric column. Actual number of evidence documents attached to the query row.
expected_source_count	Int64	integer	Whole-number numeric column. Expected number of source documents a strong system should consult.
primary_artifact_type	object	string	Text column. Primary evidence artifact type such as wiki, roadmap, or ticket.
supporting_artifact_types	object	string	Text column. JSON array of secondary artifact types involved in the answer.
company_persona	object	long text	Free-text column containing longer text values. Short description of the synthetic company context.
synthetic_company_name	object	string	Text column. Name of the fictional company represented in the corpus.
created_date	object	datetime-like string	Stored as text, but values appear to represent dates or timestamps. Earliest evidence document date tied to this query.
updated_date	object	datetime-like string	Stored as text, but values appear to represent dates or timestamps. Latest evidence document date tied to this query.
document_time_window	object	string	Text column. Date range covered by the evidence documents.
query_category	object	string	Text column. Higher-level benchmark category used for slicing the evaluation set.
evaluation_split	object	string	Text column. Train, validation, or test label for downstream eval pipelines.
benchmark_task	object	string	Text column. Primary evaluation task family.
retrieval_notes	object	string	Text column. Notes about retrieval strategy, stale docs, or cross-document hops required.
failure_mode_targeted	object	string	Text column. Primary model failure mode this query is designed to expose.
answer_format	object	string	Text column. Expected answer format such as short_text, date, list, or abstain.
source_document_titles	object	string	Text column. JSON array of evidence document titles duplicated for convenience.
source_document_types	object	string	Text column. JSON array of evidence document artifact types.
source_document_bundle_json	object	long text	Free-text column containing longer text values. JSON payload containing the full evidence documents and metadata.
synthetic_workspace_bundle_json	object	long text	Free-text column containing longer text values. JSON payload containing a compact manifest of the entire synthetic workspace.
provenance_note	object	long text	Free-text column containing longer text values. Statement describing how the synthetic benchmark was generated.
license_status	object	string	Text column. Rights statement for the benchmark content.
synthetic_generation_method	object	long text	Free-text column containing longer text values. High-level summary of the deterministic corpus generation approach.
quality_review_status	object	string	Text column. Status of rule-based consistency and type checks applied to the row.
source_domain	object	string	Text column. Synthetic source domain used for provenance tracking.
corpus_document_count	Int64	integer	Whole-number numeric column. Number of documents in the full synthetic workspace manifest.
query_intent	object	string	Text column. Concise buyer-facing description of the retrieval task under test.

Intended Use

This dataset is intended for research, experimentation, analysis, and model prototyping.

Loading the Dataset


import os
from datasets import load_dataset

HUGGINGFACE_API_KEY_KARMANE = os.environ.get("HUGGINGFACE_API_KEY_KARMANE")

dataset = load_dataset(
    "Karmane/enterprise-rag-internal-knowledge-search-benchmark",
    token=HUGGINGFACE_API_KEY_KARMANE,
)

print(dataset)
print(dataset[list(dataset.keys())[0]][0])

# getting the DataFrame itself
# df = dataset[list(dataset.keys())[0]].to_pandas()

Karmane. (2025). Enterprise RAG and Internal Knowledge Search Benchmark Dataset. Hugging Face. https://huggingface.co/datasets/Karmane/enterprise-rag-internal-knowledge-search-benchmark

Downloads last month: -

Total file size:

150 kB