A high-volume, portfolio-manager decision dataset containing economic, industry-relevant, market-context, SEC, and single-stock-moving information for use in the FENRIX PM decision game.

The dataset is organized around the question a portfolio manager would face:

Given the information available at the time, should a PM buy, sell, reallocate, or do nothing?

This repo is not just a raw news dump. It includes a professor-facing product tree, a teaching-ready compact subset, a full technical archive, official SEC/EDGAR add-ons, Apify news expansion, structured/event data, documentation, validation reports, and a private restricted-data quarantine workflow.

Latest Verified Release — 2026-07-04

Official-body product-path merge completed at commit a482b7b.

Added:

210 official-body enrichment rows
61 Parquet files across 5 product paths
107 SEC filings rows → Data/SEC/Filings/
64 structured economic rows → Data/Structured/Economic/
36 legal/regulatory news rows → Data/News/Legal_Regulatory/
2 SEC exhibit rows → Data/SEC/Exhibits/
1 rates/yields/credit row → Data/News/Rates_Yields_Credit/

Validation:

0 collisions · 0 Data/News/General fallback · 0 pre-2019 rows
0 path leaks · 0 secrets · 0 license-risk rows
compileall clean · pytest 31/31 pass

See Reports/FINAL_PM_PRODUCT_QA_20260704.md and Reports/Product_Merge_Changelog_20260704.md.

Executive summary

Component	Status
Final public repo	`Scottswi/fenrix-pm-decision-news`
Target window	July 2019 through July 2026
Full technical archive copied into final repo	3,249 Parquet files
Technical archive size	52.746 GB
Final repo audit after archive copy	105,373 total files / 105,327 Parquet files
Product Parquet files under `Data/` after archive audit	102,078
Teaching-ready compact subset	51,326 rows / 33 columns
Official SEC/EDGAR expansion	15,033 rows
Final 32-company Apify/news sprint	1,402 rows across all 32 companies
Apify rows in PM window	1,382 / 1,402
Restricted local CSV quarantine	229 CSVs / 1,764,593 rows, private only
Final verification	See `Reports/Final_Product_Closeout_20260703.md`

The Hugging Face browser viewer may understate total file size or disable preview because this repo is large. Use the verification reports and direct file checks as the source of truth.

Start here

Path	Purpose
`Docs/START_HERE_FOR_PROFESSOR.md`	Non-technical overview and recommended entry point
`Docs/QUICKSTART.md`	Minimal loading examples
`Data/Teaching_Ready_Compact/`	Small, clean subset for immediate use
`Data/_SAMPLES/`	Tiny sample files for fast inspection
`Reports/Final_Product_Closeout_20260703.md`	Final delivery report
`Reports/final_product_closeout_20260703.json`	Machine-readable final verification
`Archive/Technical_Warehouse_Full_Source_Lake/`	Full high-volume source-lake archive

Recommended first load:

from datasets import load_dataset

repo = "Scottswi/fenrix-pm-decision-news"

ds = load_dataset(
    "parquet",
    data_files={
        "train": f"hf://datasets/{repo}/Data/Teaching_Ready_Compact/*.parquet"
    },
    split="train",
)

print(ds)
print(ds.column_names)
print(ds[0])

Why this dataset exists

The dataset was built to support a portfolio-management classroom/game workflow requiring:

economic news,
industry-relevant news,
market-moving macro context,
company-specific but not trivially reverse-engineerable events,
SEC filings and official company disclosures,
structured data that can support market decisions,
news spread across the last seven years rather than clustered in one period.

The intent is to provide enough information to create realistic PM decisions: buy, sell, reallocate, or do nothing.

Public data structure

Data/
  News/
    Economic_Macro/
    Sector_Industry/
    Single_Stock/
    Multi_Company/
    Market_Context/
  Structured/
  SEC/
  Teaching_Ready_Compact/
  _SAMPLES/

Docs/
Reports/
Archive/

News categories

Folder	Example content
`Data/News/Economic_Macro/`	inflation, rates, growth, recession, labor, consumer activity
`Data/News/Sector_Industry/`	retail, energy, semiconductors, banking, healthcare, consumer, media
`Data/News/Single_Stock/`	earnings, guidance, product, legal/regulatory, analyst, SEC-linked company events
`Data/News/Multi_Company/`	peer comparisons, supply-chain events, industry readthroughs
`Data/News/Market_Context/`	yields, volatility, sentiment, geopolitical risk, commodity context
`Data/Structured/`	official macro, event calendar, CFTC, GDELT metadata, structured indicators
`Data/SEC/`	target-company filings, 8-K, EX-99 detection, insider/filing-related records

Source coverage

The dataset combines multiple source families rather than relying on a single feed.

1. High-volume financial-news source lake

Source family	Role in dataset
FNSPID / financial news + stock-price integration source	Large-scale financial news backbone and historical company-news coverage
Kaggle massive stock news	Broad ticker/company headline coverage
Hugging Face financial-news datasets	Additional financial headline/source diversity
GitHub-hosted financial-news datasets	Supplemental public/community financial news
Existing technical warehouse	Preserved under `Archive/Technical_Warehouse_Full_Source_Lake/`

FNSPID is especially important because the public FNSPID paper describes 15.7 million time-aligned financial-news records and 29.7 million stock-price records across 4,775 S&P 500 companies from 1999 to 2023.

2. Official and semi-official market-moving data

Source	Type	PM-decision relevance
SEC/EDGAR	filings, 8-K, EX-99-detected events	company-specific material disclosures
CFTC	positioning / commitments-style structured market data	commodity, rate, and futures market context
Federal Reserve / macro-policy sources	policy text and macro context	rates, liquidity, recession/growth regime
GDELT BigQuery cached metadata	event/news metadata, tone, entities, geography	global macro, geopolitical, and market context
Guardian metadata	metadata/link-level news context	public discovery layer where full bodies were withheld

3. Final 32-company Apify/news sprint

A final targeted Apify sprint added fresh, query-targeted coverage for all 32 target companies.

Metric	Value
Rows collected/uploaded	1,402
Companies covered	32 / 32
Rows in PM window	1,382
Path	`Data/News/Single_Stock/Apify_32_Company_Sprint/`
Teaching slice	`Data/Teaching_Ready_Compact/teaching_apify_32_company_20260703.parquet`

The sprint focused on earnings, guidance, lawsuits, regulation, layoffs, restructuring, mergers/acquisitions, supply chain, pricing/margins, analyst ratings, stock news, and SEC filing references.

4. Restricted/private source preservation

Local Bloomberg-like CSVs were preserved in a private companion repo, not redistributed publicly.

Component	Value
Restricted CSV files preserved	229
Restricted rows preserved	1,764,593
Public redistribution	No
Status	Private restricted quarantine only

These records are available for internal/private project review but are excluded from the public repo because redistribution status is unclear.

Target company universe

The final company universe includes 32 names:

Ticker	Company
CL	Colgate-Palmolive
PEP	PepsiCo
TJX	TJX Companies
PM	Philip Morris International
AMZN	Amazon
HBAN	Huntington Bancshares / Huntington Bank
BLK	BlackRock
GOOGL	Alphabet / Google
AAPL	Apple
SBUX	Starbucks
NRG	NRG Energy
DIS	Disney / Walt Disney
ETSY	Etsy
RL	Ralph Lauren
TSLA	Tesla
NKE	Nike
AMD	Advanced Micro Devices
SNDK	SanDisk / Sandisk
APP	AppLovin
MCK	McKesson
NFLX	Netflix
PLTR	Palantir
GM	General Motors
KR	Kroger
TGT	Target Corporation
BAC	Bank of America
ORCL	Oracle
GIS	General Mills
AXP	American Express
JNJ	Johnson & Johnson
ADBE	Adobe
FDX	FedEx

Ambiguous tickers such as APP, GM, KR, GIS, DIS, RL, NRG, CL, and PM should be interpreted with company-name context rather than ticker-only matching.

What a row represents

Rows may represent one of several information types:

Row type	Examples
News article/headline	company news, macro news, sector news, market-moving events
SEC filing/disclosure	8-K, EX-99-detected filing, company disclosure
Structured market/event data	CFTC, GDELT metadata, event calendar
Metadata-only news reference	URL/title/publisher/date without full redistributed article body
Teaching-ready normalized record	33-column standardized subset for classroom use

Common normalized columns are available in Data/Teaching_Ready_Compact/ and include source_id, source_name, canonical_url, published_at, publisher, title, description, tickers, companies, article_group, topic, and more.

Loading examples

Load the teaching-ready subset

from datasets import load_dataset

repo = "Scottswi/fenrix-pm-decision-news"

ds = load_dataset(
    "parquet",
    data_files={"train": f"hf://datasets/{repo}/Data/Teaching_Ready_Compact/*.parquet"},
    split="train",
)

print(ds)

Load only Apify 32-company sprint rows

from datasets import load_dataset

repo = "Scottswi/fenrix-pm-decision-news"

apify = load_dataset(
    "parquet",
    data_files={
        "train": f"hf://datasets/{repo}/Data/News/Single_Stock/Apify_32_Company_Sprint/*.parquet"
    },
    split="train",
)

print(apify)

Load SEC/EDGAR rows

from datasets import load_dataset

repo = "Scottswi/fenrix-pm-decision-news"

sec = load_dataset(
    "parquet",
    data_files={"train": f"hf://datasets/{repo}/Data/SEC/**/*.parquet"},
    split="train",
)

print(sec)

Load the full product tree

from datasets import load_dataset

repo = "Scottswi/fenrix-pm-decision-news"

news = load_dataset(
    "parquet",
    data_files={"train": f"hf://datasets/{repo}/Data/News/**/*.parquet"},
    split="train",
)

print(news)

For most users, start with Data/Teaching_Ready_Compact/ rather than the full tree.

Verification and reports

Important reports:

Report	Purpose
`Reports/FINAL_PM_PRODUCT_QA_20260704.md`	Latest QA report (2026-07-04)
`Reports/Product_Merge_Changelog_20260704.md`	Official-body merge changelog
`Reports/Final_Product_Closeout_20260703.md`	Final delivery summary
`Reports/final_product_closeout_20260703.json`	Machine-readable closeout
`Reports/final_repo_tree_path_audit_after_archive_20260703.json`	Final tree audit after archive copy
`Reports/Technical_Warehouse_Archive_Copy_20260703.md`	Archive copy report
`Reports/final_dataset_inventory_compact.csv`	Compact inventory
`Reports/usability_remote_validation_20260703.json`	Remote loading validation
`Reports/Apify_32_Company_Sprint_20260703.md`	Apify sprint summary
`Reports/Final_One_Hour_Expansion_Status_20260703.md`	SEC/quarantine expansion summary

Validation performed during build included:

remote Hugging Face file-existence checks,
local and remote Parquet loading,
teaching subset schema consistency checks,
local path leak scans on samples,
SEC/EDGAR row validation,
Apify normalized Parquet remote load,
private quarantine privacy check,
Hugging Face upload lock acquisition/release during write operations.

Use notes and limitations

Viewer metadata may lag. The Hugging Face web UI may understate file size or disable preview because the repo is large. Use reports and direct file checks as the source of truth.
Not every row has full article body text. Some sources provide headline/metadata/URL-level records. Guardian full bodies and other redistribution-sensitive full texts were withheld where needed.
Restricted local CSVs are not public. The private quarantine preserves them for internal review but avoids public redistribution.
Some source families have different licenses. Official sources, community datasets, scraped metadata, and private restricted files have different use conditions. Use row-level source and redistribution fields.
Company-specific rows are not intended to make the game trivially identifiable. Later game construction should anonymize company identity and use sector/macro context carefully.
The technical archive is large. Use compact/teaching subsets for classroom prototyping and full archive paths for deeper retrieval.

Why this answers the assignment

The assignment asked for a foundation of economic, industry-relevant news and market/single-stock-moving information over the last seven years, spread across time and relevant to PM decisions.

This dataset answers that by combining:

high-volume financial news,
macroeconomic and market-context news,
sector/industry news,
single-stock and multi-company event coverage,
official SEC/EDGAR filings,
structured event/positioning/macro records,
a 32-company targeted news sprint,
a teaching-ready compact subset,
a full source-lake archive for deeper retrieval,
reports and documentation for reproducible use.

The public dataset is designed for the game workflow. The private quarantine keeps restricted local data available without exposing it publicly.

Downloads last month: 2,184

Total file size:

60.2 GB