leeroy-jankins/Federal-Managers-Financial-Integrity-Act-of-1982

Federal Managers' Financial Integrity Act of 1982 Corpus

<img src="assets/repo-header.png" alt="Preview" width="1000"/>

Dataset Description

The Federal Managers' Financial Integrity Act of 1982 Corpus is a processed legal and federal financial management dataset derived from the Federal Managers’ Financial Integrity Act of 1982, commonly abbreviated as FMFIA.

FMFIA was enacted as Public Law 97-255 on September 8, 1982. The Act amended the Accounting and Auditing Act of 1950 and strengthened the responsibility of federal managers for establishing, evaluating, improving, and reporting on internal accounting and administrative controls.

The Act requires executive agencies to establish internal accounting and administrative controls consistent with standards prescribed by the Comptroller General. It also requires agency heads to evaluate their internal control and financial management systems and provide annual statements concerning whether those systems provide reasonable assurance that statutory objectives are being achieved.

This repository includes the original source document and several processed representations designed for natural-language processing, information retrieval, semantic search, and language-model applications.

The repository provides:

• —The original source document in PDF format
• —A chunked text dataset in CSV format
• —A token-level dataset in CSV format
• —A normalized plain-text corpus in TXT format

The processed text was divided into chunks containing approximately 1,000 tokens, with an overlap of 200 tokens between consecutive chunks.

The normalized corpus was prepared using basic preprocessing that:

• —Removes HTML
• —Removes Markdown syntax
• —Removes symbols and unsupported characters
• —Removes identified extraction errors
• —Collapses repeated whitespace
• —Converts the text to lowercase

Legal Background

The Federal Managers’ Financial Integrity Act was enacted to improve accountability and internal control throughout the executive branch.

The Act responded to long-standing concerns involving:

• —Weak internal controls
• —Ineffective accounting systems
• —Unreliable financial information
• —Waste and misuse of federal resources
• —Inadequate safeguards over assets
• —Failure to comply with laws and regulations
• —Insufficient management accountability
• —Inconsistent evaluation of agency control systems

FMFIA placed responsibility for internal control directly on federal agency managers. It established a statutory framework under which agencies must maintain systems that provide reasonable assurance that federal resources are protected and used properly.

The Act is closely associated with:

• —Internal control
• —Enterprise risk management
• —Management assurance
• —Financial management systems
• —Accounting systems
• —Operational effectiveness
• —Regulatory compliance
• —Asset protection
• —Fraud prevention
• —Waste prevention
• —Annual assurance reporting
• —Material weakness identification
• —Corrective-action planning

The principal FMFIA requirements are now codified in 31 U.S.C. § 3512(c) and § 3512(d).

Principal Requirements

FMFIA requires executive agencies to establish internal accounting and administrative controls that provide reasonable assurance that:

1. 1.Obligations and costs comply with applicable law.
2. 2.Funds, property, and other assets are safeguarded against waste, loss, unauthorized use, or misappropriation.
3. 3.Revenues and expenditures applicable to agency operations are properly recorded and accounted for.
4. 4.Accounts and reliable financial and statistical reports can be prepared.
5. 5.Accountability over agency assets can be maintained.

The Act also requires agency heads to evaluate and report on their internal control and financial management systems.

These evaluations support annual management assurance statements addressing:

• —The effectiveness of internal control
• —Compliance with applicable internal control standards
• —Material weaknesses
• —Financial management system conformance
• —Corrective actions
• —Areas requiring continued management attention

Dataset Purpose

The dataset is intended to support:

• —Retrieval-augmented generation
• —Semantic search
• —Legal and policy research
• —Federal internal control research
• —Federal financial management education
• —Document question answering
• —Regulatory text analysis
• —Embedding generation
• —Information retrieval
• —Topic classification
• —Language-model evaluation
• —Summarization
• —Knowledge-base development
• —Management assurance research
• —Risk-management research
• —Internal control training
• —Search-index development
• —Corpus linguistics

The overlapping chunk structure is particularly useful for retrieval systems because it preserves context that might otherwise be divided across chunk boundaries.

Repository Contents

The repository contains four principal artifact types.

Example filenames may include:

federal_managers_financial_integrity_act_1982.pdf
fmfia_1982_chunks.csv
fmfia_1982_tokens.csv
fmfia_1982_corpus.txt
README.md

Actual filenames may differ from these examples.

Source Document

The principal source is:

• —Title: Federal Managers’ Financial Integrity Act of 1982
• —Abbreviation: FMFIA
• —Public law: Public Law 97-255
• —Statutes at Large citation: 96 Stat. 814
• —Enactment date: September 8, 1982
• —Codification: 31 U.S.C. § 3512(c)–(d)
• —Document type: United States federal statute

The repository PDF represents the source version used to produce the dataset.

Users should inspect the PDF and repository metadata to determine:

• —Exact source filename
• —Publication source
• —Retrieval date
• —Page count
• —Document checksum
• —Whether the PDF contains only the Act or additional legislative material

Processing Methodology

Source Acquisition

The source PDF was obtained from an official or authoritative United States Government source and retained in the repository as the input document from which the processed artifacts were generated.

Text Extraction

Text was extracted from the PDF before preprocessing and tokenization.

PDF extraction may introduce artifacts involving:

• —Page headers
• —Page footers
• —Page numbers
• —Line breaks
• —Hyphenation
• —Statutory citations
• —Marginal notes
• —Multi-column layouts
• —Section headings
• —Signature blocks
• —Historical annotations
• —Scanned text
• —Character encoding

The preprocessing pipeline removes common artifacts but does not guarantee perfect reconstruction of the original document.

Corpus Preprocessing

The extracted text was normalized using a basic preprocessing pipeline.

The preprocessing operations include:

1. 1.Removing HTML tags and encoded HTML artifacts
2. 2.Removing Markdown syntax
3. 3.Removing symbols and unsupported characters
4. 4.Removing identified extraction errors
5. 5.Collapsing repeated spaces, tabs, and line breaks
6. 6.Normalizing whitespace
7. 7.Converting the text to lowercase

The resulting corpus is optimized for machine processing rather than exact visual or typographic reproduction.

Tokenization

The normalized corpus was tokenized before chunk generation.

Tokenization supports:

• —Token counting
• —Chunk generation
• —Context-window management
• —Embedding workflows
• —Model-input preparation
• —Token-frequency analysis
• —Language-model experiments

Token counts depend on the tokenizer or encoding used during dataset construction. Different tokenizers may produce different token boundaries and total token counts for the same text.

For reproducibility, the repository should identify:

• —Tokenizer name
• —Tokenizer version
• —Encoding name
• —Vocabulary version
• —Special-token behavior
• —Truncation settings
• —Padding settings

Chunking

The tokenized corpus was divided using a sliding-window chunking process.

Each new chunk begins approximately 800 tokens after the beginning of the preceding chunk.

Chunking Example

Chunk 1:
Tokens 0 through 999

Chunk 2:
Tokens 800 through 1799

Chunk 3:
Tokens 1600 through 2599

The repeated 200-token region helps preserve context where a sentence, paragraph, list, or statutory requirement crosses a chunk boundary.

Dataset Configurations

Recommended dataset configurations include:

The CSV files are the primary structured artifacts for machine-learning and retrieval workflows.

Dataset Structure

Chunks Dataset

The chunks file contains sequential text segments generated from the normalized corpus.

A recommended schema is:

The actual repository schema may contain only a subset of these fields.

Example Chunk Record

{
  "chunk_id": 1,
  "text": "the congress finds that requirements for internal accounting and administrative controls...",
  "token_count": 1000,
  "token_start": 0,
  "token_end": 999,
  "document_id": "fmfia-1982",
  "source_file": "federal_managers_financial_integrity_act_1982.pdf"
}

Tokens Dataset

The token file contains a token-level representation of the normalized corpus.

A recommended schema is:

Example Token Record

{
  "token_id": 1,
  "token": "the",
  "token_index": 0,
  "token_value": 1820,
  "chunk_id": 1,
  "document_id": "fmfia-1982"
}

Corpus Dataset

The corpus file contains the complete normalized text.

The corpus:

• —Is lowercased
• —Uses collapsed whitespace
• —Excludes HTML
• —Excludes Markdown syntax
• —Excludes unsupported symbols
• —Excludes identified extraction errors
• —Does not reliably preserve PDF pagination
• —Does not reliably preserve visual formatting

The corpus is suitable for:

• —Full-text search
• —Keyword analysis
• —Corpus statistics
• —Retokenization
• —Alternative chunking
• —Embedding generation
• —Topic modeling
• —Text classification
• —Language-model experiments

Original PDF

The original PDF should be used whenever exact statutory language or formatting is required.

The PDF is the preferred artifact for verifying:

• —Capitalization
• —Punctuation
• —Section numbers
• —Statutory citations
• —Paragraph structure
• —Page numbers
• —Headings
• —Defined terms
• —Signature information
• —Original wording

Dataset Splits

The dataset is normally provided as a single split because it represents one legal corpus.

Users may create additional splits for training and evaluation.

Possible splitting strategies include:

• —Section-based splitting
• —Topic-based splitting
• —Holdout sections
• —Random chunk splitting
• —Retrieval evaluation subsets
• —Manually curated question-answer sets

Random splitting should be used cautiously because adjacent chunks contain overlapping text.

Overlapping or neighboring chunks should remain in the same split to prevent data leakage between training, validation, and test sets.

Loading the Dataset

Load from Hugging Face

from datasets import load_dataset


dataset = load_dataset(
    "YOUR_HUGGING_FACE_USERNAME/fmfia-1982-corpus"
)

print(dataset)

Load the Chunk Dataset

from datasets import load_dataset


dataset = load_dataset(
    "csv",
    data_files={
        "train": "fmfia_1982_chunks.csv"
    }
)

print(dataset["train"][0])

Load the Token Dataset

from datasets import load_dataset


dataset = load_dataset(
    "csv",
    data_files={
        "train": "fmfia_1982_tokens.csv"
    }
)

print(dataset["train"][0])

Load the Corpus

from pathlib import Path


corpus_path = Path(
    "fmfia_1982_corpus.txt"
)

corpus = corpus_path.read_text(
    encoding="utf-8"
)

print(corpus[:1000])

Load the CSV Files with pandas

import pandas as pd


df_chunks = pd.read_csv(
    "fmfia_1982_chunks.csv"
)

df_tokens = pd.read_csv(
    "fmfia_1982_tokens.csv"
)

print(df_chunks.head())
print(df_tokens.head())

Retrieval-Augmented Generation

The chunk dataset is designed for retrieval-augmented generation workflows.

A typical workflow includes:

1. 1.Loading the chunk CSV
2. 2.Generating an embedding for each chunk
3. 3.Storing the embeddings in a vector index
4. 4.Embedding the user’s question
5. 5.Retrieving the most relevant chunks
6. 6.Supplying those chunks to a language model
7. 7.Generating a source-grounded answer
8. 8.Returning relevant source metadata

Example Embedding Preparation

from datasets import load_dataset
from sentence_transformers import SentenceTransformer


dataset = load_dataset(
    "csv",
    data_files={
        "train": "fmfia_1982_chunks.csv"
    },
    split="train"
)

model = SentenceTransformer(
    "sentence-transformers/all-MiniLM-L6-v2"
)

embeddings = model.encode(
    dataset["text"],
    normalize_embeddings=True,
    show_progress_bar=True
)

print(embeddings.shape)

Recommended Retrieval Prompt

You are a federal internal control and financial management research
assistant.

Answer the question using only the supplied Federal Managers' Financial
Integrity Act context. Distinguish explicit statutory requirements from
later implementation guidance. State when the supplied context is
insufficient and identify the relevant statutory section when available.

Context:
{retrieved_chunks}

Question:
{question}

Answer:

Core FMFIA Concepts

Internal Accounting and Administrative Controls

FMFIA requires executive agencies to establish internal accounting and administrative controls consistent with standards prescribed by the Comptroller General.

These controls are intended to provide reasonable assurance that:

• —Obligations and costs comply with applicable law
• —Assets are safeguarded
• —Revenues and expenditures are properly recorded
• —Reliable financial and statistical reports can be prepared
• —Accountability over assets is maintained

Reasonable Assurance

Reasonable assurance recognizes that internal control systems cannot eliminate every risk or guarantee perfect compliance.

Controls should provide a reasonable, rather than absolute, level of confidence that management objectives will be achieved.

Management Responsibility

FMFIA establishes internal control as a management responsibility.

Agency managers are responsible for:

• —Establishing controls
• —Evaluating controls
• —Identifying weaknesses
• —Correcting deficiencies
• —Reporting on control effectiveness
• —Maintaining accountable operations

Annual Assurance Statements

Agency heads must report annually on whether agency internal controls comply with applicable requirements and provide reasonable assurance that their intended objectives are being achieved.

Annual reporting may identify:

• —Material weaknesses
• —Nonconforming financial management systems
• —Corrective-action plans
• —Planned remediation
• —Areas requiring continued monitoring

Accounting and Financial Management Systems

FMFIA also addresses agency accounting systems and whether those systems conform to applicable principles, standards, and related requirements.

System deficiencies may affect:

• —Financial reporting
• —Funds control
• —Asset accountability
• —Transaction processing
• —Data reliability
• —Compliance
• —Auditability

Material Weaknesses

A material weakness is a significant deficiency or combination of deficiencies that warrants attention by senior management.

Material weaknesses generally require:

• —Formal identification
• —Management reporting
• —Corrective-action planning
• —Progress monitoring
• —Resolution validation

OMB Circular A-123

OMB Circular A-123 provides the principal government-wide implementation framework for management responsibility for internal control.

It addresses internal control objectives involving:

• —Effective and efficient operations
• —Reliable reporting
• —Compliance with applicable laws and regulations

The Circular should be consulted separately because it contains implementation guidance rather than the original statutory text alone.

Intended Uses

Appropriate Uses

The dataset may be used for:

• —Federal internal control research
• —Management assurance training
• —Retrieval-augmented generation
• —Semantic search
• —Document-grounded question answering
• —Legal information retrieval
• —Federal financial management education
• —Risk-management research
• —Internal control classification
• —Statutory summarization
• —Embedding-model evaluation
• —Search relevance evaluation
• —Language-model grounding
• —Knowledge-graph development
• —Regulatory comparison
• —Topic modeling
• —Keyphrase extraction
• —Named-entity recognition

Example Questions

The dataset may support questions such as:

• —What is the purpose of FMFIA?
• —What types of controls must executive agencies establish?
• —What does reasonable assurance mean under FMFIA?
• —What responsibilities are assigned to agency heads?
• —What must agencies report annually?
• —How does FMFIA address the safeguarding of assets?
• —How does the Act address compliance with applicable law?
• —What does FMFIA require regarding revenues and expenditures?
• —How are accounting systems addressed?
• —What is a material weakness?
• —What is the relationship between FMFIA and OMB Circular A-123?
• —What role does the Comptroller General have under the Act?
• —How does FMFIA support reliable financial reporting?
• —How does FMFIA promote management accountability?

Out-of-Scope Uses

The dataset should not be treated as:

• —Legal advice
• —Accounting advice
• —Audit advice
• —An official interpretation of FMFIA
• —A replacement for the enacted statute
• —A replacement for the United States Code
• —A replacement for current OMB Circular A-123 guidance
• —A replacement for current GAO internal control standards
• —A replacement for agency-specific control policies
• —A replacement for Inspector General guidance
• —A definitive source for current reporting procedures
• —A complete source for federal enterprise risk management
• —A basis for issuing an audit opinion
• —A basis for making formal management assurance determinations

The dataset should not be used by itself to make high-impact legal, financial, audit, operational, or administrative decisions.

Data Quality

The dataset was designed to provide:

• —A preserved source PDF
• —A normalized text corpus
• —Structured token records
• —Overlapping token-based chunks
• —Machine-readable CSV files
• —Context windows suitable for retrieval
• —Simplified downstream processing
• —Compatibility with common NLP libraries
• —Traceability to the source document

Potential quality limitations include:

• —PDF extraction errors
• —Missing or damaged characters
• —Broken statutory citations
• —Loss of page structure
• —Loss of capitalization
• —Loss of punctuation or symbols
• —Flattened headings
• —Merged paragraphs
• —Split words
• —Repeated headers or footers
• —Tokenizer-dependent token counts
• —Duplicate text created by overlap
• —Incomplete provenance metadata
• —Version differences between the PDF and current codified law

Effects of Preprocessing

Lowercasing

Lowercasing improves normalization but removes distinctions involving:

• —Acronyms
• —Proper nouns
• —Defined terms
• —Capitalized headings
• —Statutory titles
• —Emphasized text

Symbol Removal

Removing symbols may affect:

• —Statutory citations
• —Section identifiers
• —Parenthetical references
• —Hyphenated terms
• —Dates
• —Percentages
• —Enumerated lists
• —Legal notation

Whitespace Collapsing

Whitespace normalization may flatten:

• —Paragraph boundaries
• —Indented subsections
• —Lists
• —Tables
• —Multi-column layouts
• —Signature blocks

Error Removal

Removal of extraction errors improves general corpus quality but may inadvertently remove valid text if it resembles a known extraction artifact.

Chunk Overlap

The 200-token overlap introduces intentional duplication.

Users should account for this duplication when:

• —Calculating term frequencies
• —Measuring corpus size
• —Training language models
• —Creating train and test splits
• —Aggregating search results
• —Evaluating retrieval diversity
• —Computing document statistics

Validation Considerations

Exact-Language Verification

The original PDF or current United States Code should be consulted whenever exact statutory wording matters.

Codification

The original Act amended earlier law and its principal requirements are now codified in Title 31 of the United States Code.

A response based on the original public law should not automatically be treated as a statement of the current codified text.

Implementation Guidance

FMFIA establishes statutory requirements, while OMB Circular A-123 and related guidance provide implementation details.

Models should distinguish between:

• —Statutory requirements
• —OMB guidance
• —GAO standards
• —Agency policy
• —Audit practice
• —Management procedures

Chunk Leakage

Adjacent and overlapping chunks should remain in the same dataset split to prevent information leakage.

Numeric and Citation Accuracy

Dates, section numbers, public law references, and United States Code citations should be verified against the source PDF.

Regulatory Currency

Users should determine whether later statutes, standards, circulars, or agency policies affect the interpretation or implementation of FMFIA.

Limitations

Historical Source

The source law was enacted in 1982. Later legislation and guidance have expanded the federal internal control and financial management framework.

Loss of Formatting

The normalized corpus does not reliably preserve:

• —Page layout
• —Statutory indentation
• —Subsection hierarchy
• —Tables
• —Footnotes
• —Marginal notes
• —Signature blocks
• —Original typography

Incomplete Semantic Preservation

Removing symbols and formatting may alter the interpretation of:

• —Section references
• —Lists
• —Defined terms
• —Citations
• —Dates
• —Parenthetical qualifications
• —Cross-references

Tokenizer Dependence

The 1,000-token chunk size and 200-token overlap reflect the tokenizer used during dataset construction.

A different tokenizer may produce different:

• —Token counts
• —Chunk boundaries
• —Vocabulary identifiers
• —Decoded token values

No Guarantee of Legal Currency

The dataset is a research artifact and does not guarantee that every provision or interpretation remains current.

Limited Scope

The dataset focuses on FMFIA and does not comprehensively cover:

• —OMB Circular A-123
• —GAO Standards for Internal Control in the Federal Government
• —Federal Financial Management Improvement Act
• —Chief Financial Officers Act
• —Government Management Reform Act
• —Inspector General Act
• —Agency-specific internal control policy
• —Enterprise risk management guidance
• —Payment integrity requirements
• —Federal accounting standards

Related Authorities and Guidance

Users may benefit from reviewing FMFIA together with:

• —OMB Circular A-123, Management’s Responsibility for Enterprise Risk Management and Internal Control
• —GAO Standards for Internal Control in the Federal Government
• —Chief Financial Officers Act of 1990
• —Government Management Reform Act of 1994
• —Federal Financial Management Improvement Act of 1996
• —Inspector General Act of 1978
• —Federal Managers’ Financial Integrity Act codification in 31 U.S.C. § 3512
• —Payment Integrity Information Act of 2019
• —Federal Accounting Standards Advisory Board standards
• —Agency-specific internal control policies
• —Agency annual financial reports
• —Agency performance and accountability reports

These materials are not necessarily included in the dataset unless they appear as separate repository artifacts.

Ethical and Operational Considerations

The dataset contains public federal legal and financial management information. It is not intended to contain:

• —Classified information
• —Controlled unclassified information
• —Personally identifiable information
• —Proprietary information
• —Banking information
• —Individual financial records
• —Procurement-sensitive information

Downstream applications should nevertheless be evaluated for:

• —Incorrect statutory interpretation
• —Reliance on outdated guidance
• —Hallucinated control requirements
• —Incorrect material weakness classifications
• —Misrepresentation of agency responsibilities
• —Failure to distinguish law from guidance
• —Failure to disclose uncertainty
• —Inadequate human review
• —Failure to cite authoritative sources
• —Use in high-impact management or audit decisions

Formal assurance, audit, legal, and financial management decisions should be reviewed by qualified personnel using current authoritative sources.

Licensing

The original federal statute is a work of the United States Government and is generally not subject to copyright protection within the United States.

The repository’s derived materials may include:

• —Chunk organization
• —Token records
• —Corpus formatting
• —Metadata
• —Documentation
• —Processing scripts
• —Dataset configurations

These derived materials may be subject to the repository’s stated license.

The Hugging Face metadata uses other unless the repository owner explicitly assigns a recognized open license to the derived dataset materials.

Users are responsible for determining whether additional restrictions apply to redistribution, derivative works, or use outside the United States.

Citation

When using this dataset, cite both the dataset repository and the underlying statute.

Dataset Citation

@dataset{fmfia_1982_corpus,
  author       = {DATASET_AUTHOR},
  title        = {Federal Managers' Financial Integrity Act of 1982 Corpus},
  year         = {DATASET_PUBLICATION_YEAR},
  publisher    = {Hugging Face},
  url          = {DATASET_REPOSITORY_URL},
  note         = {Tokenized and chunked corpus derived from Public Law 97-255}
}

Statutory Citation

Federal Managers' Financial Integrity Act of 1982,
Pub. L. No. 97-255, 96 Stat. 814 (1982).

Codified Citation

31 U.S.C. § 3512(c)-(d).

Dataset Maintenance

The dataset should be reviewed periodically to:

• —Confirm the source PDF
• —Record the source retrieval date
• —Add the source URL
• —Add a source checksum
• —Correct extraction errors
• —Improve preprocessing rules
• —Validate token counts
• —Record tokenizer metadata
• —Detect missing pages
• —Remove unintended duplicate records
• —Preserve statutory section identifiers
• —Add page-level provenance
• —Document schema changes
• —Version the preprocessing pipeline
• —Rebuild chunks when the source changes
• —Compare the original Act with current codification

Material changes should be documented in repository commits or release notes.

Recommended Future Enhancements

Potential enhancements include:

• —Statutory section metadata
• —United States Code citations
• —Page numbers
• —Source character offsets
• —Source token offsets
• —Exact PDF provenance
• —Document checksums
• —Tokenizer name and version
• —Preprocessing version
• —Preserved-case corpus
• —Raw extracted text
• —Section-aware chunking
• —Paragraph-aware chunking
• —Semantic chunking
• —Heading-aware chunking
• —Manually curated question-answer pairs
• —Retrieval evaluation benchmarks
• —Material weakness classification examples
• —Internal control objective labels
• —OMB Circular A-123 cross-references
• —GAO Green Book cross-references
• —Embedding columns
• —Sparse retrieval indexes
• —Dense retrieval indexes

An expanded chunk schema could include:

Reproducibility Metadata

For reproducible dataset generation, the repository should record:

• —Source URL
• —Source filename
• —Source checksum
• —Source retrieval date
• —PDF page count
• —Extraction library
• —Extraction-library version
• —Preprocessing script
• —Preprocessing version
• —Tokenizer
• —Tokenizer version
• —Encoding
• —Chunk size
• —Chunk overlap
• —CSV schema
• —Dataset generation date
• —Python version
• —Dependency versions

Disclaimer

This dataset is provided for research, education, experimentation, and software-development purposes.

It is not an official publication of the United States Government and is not affiliated with, endorsed by, or sponsored by Congress, the Office of Management and Budget, the Government Accountability Office, or any federal agency.

The dataset does not constitute legal, accounting, audit, internal control, risk-management, or financial management advice.

Users are responsible for validating all material information against the original statute, the current United States Code, current OMB guidance, current GAO standards, and other applicable authoritative sources.