NEW: Access all datasets via MCP Server — no SDK required. Any AI agent, any MCP-compatible client.
Learn moreSEC EDGAR data your backtest can trust.
111M+ financial facts from 19,000+ entities spanning 30+ years. Point-in-time accurate. Zero survivorship bias. Query it with DuckDB, Python, or the MCP Server — then layer on the 78M-row smart-money dataset.
Built on the authoritative source for US public company financials
The SEC requires every public company to file structured XBRL financial statements. These filings are published as the EDGAR Financial Statements Data Sets — quarterly ZIP archives containing machine-readable data for every 10-K, 10-Q, 8-K, and 20-F filed with the Commission.
Each quarterly release includes five core files: num.txt (numeric values), sub.txt (submission metadata), tag.txt (XBRL tag definitions), pre.txt (presentation linkbase), and cal.txt (calculation linkbase).
Source Details
From raw SEC filings to queryable Parquet in 10 steps
Every financial fact passes through a deterministic pipeline that standardizes, enriches, and validates before export. No manual intervention. No estimation.
SEC EDGAR Ingestion
Quarterly XBRL bulk downloads from SEC EDGAR Financial Statements Data Sets are ingested automatically. Each release contains num.txt, sub.txt, tag.txt, pre.txt, and cal.txt files covering every public filing.
XBRL Submission Parsing
Every XBRL submission in the quarterly dump is parsed. Filing metadata, entity info, tagged numeric values, and calculation linkbases are extracted and validated.
Entity & Security Normalization
CIK numbers are resolved to standardized entity records. SIC codes are mapped to sectors and industries. Exchange information, ticker symbols, and CUSIP identifiers are enriched from multiple sources.
Concept Standardization
11,966 raw XBRL tags are mapped to 292 canonical standard_concept values. Revenue synonyms, debt variants, and custom extensions all resolve to a single canonical concept name; unmapped tags fall through to Other.
Point-in-Time Indexing
Every fact receives a accepted_at timestamp equal to the SEC acceptance date of the filing that introduced it. No backfilling, no estimation. What was known on any date is precisely queryable.
Amendment Reconciliation
10-K/A and 10-Q/A filings (restated financials) are tracked separately. Original values and restated values coexist in the dataset, each with their own accepted_at timestamp.
Derived Quarterly Values
Q2 and Q3 cash flow statements in 10-Qs report year-to-date totals. The pipeline computes the incremental quarterly figure and stores it as derived_quarterly_value alongside the raw YTD number.
Index Membership Enrichment
S&P500 and Russell 1000/2000/3000 membership is tracked historically in index_membership with effective_date / removal_date and [) interval semantics. JOIN references on cik = cik for any membership question (current or historical) — there is no is_sp500 flag.
Parquet Export
Column-oriented Parquet files with ZSTD compression are generated for each tier. Optimized for DuckDB, Polars, and Spark. Exported to distributed object storage after every EDGAR quarterly release.
Manifest Update
manifest.json records the snapshot date, last_updated timestamp, and row counts for every table. SDKs and integrations use this to detect fresh data automatically.
Eleven core tables — seventeen with smart-money
Each is a column-oriented Parquet file with ZSTD compression — query with DuckDB, Polars, Spark, or any engine that reads Parquet. The key core tables are below; the six FULL-tier smart-money tables live in the smart-money dataset.
Relationships (text)
securityreferencesentityviaentity_id → cik(many → 1)filingreferencesentityviaentity_id → cik(many → 1)factreferencesentityviaentity_id → cik(many → 1)factreferencesfilingviaaccession_id(many → 1)factreferencesstandard_conceptviastandard_concept(many → 1)standard_conceptreferencestaxonomy_guideviastandard_concept(many → 1)ratioreferencesentityviaentity_id → cik(many → 1)referencesreferencesentityviacik(many → 1)index_membershipreferencesreferencesviacik = cik(many → 1)
Every SEC-registered entity that has filed XBRL financial statements. Includes active, delisted, bankrupt, and acquired companies.
Show 14 columns ↓Hide columns ↑
ciknamesic_codesic_descriptionsectorindustrystate_of_incorporationfiscal_year_endbusiness_addressmailing_addressformer_namesis_foreignflagscategoryTicker symbols and exchange listings (SCD Type 2 with valid_from / valid_to). One entity may have multiple securities — filter is_primary_ticker = TRUE for one row per CIK.
Show 14 columns ↓Hide columns ↑
identity_idsymbolexchangemicvalid_fromvalid_tois_activeis_primary_tickerfigicomposite_figishare_class_figisecurity_typemarket_sectorEvery XBRL filing processed: 10-K, 10-Q, 8-K, 20-F, and their amendments since 1993. accepted_at is the SEC acceptance timestamp; superseded_by chains the amendment lineage.
Show 18 columns ↓Hide columns ↑
accession_identity_idform_typecore_typefiling_datereport_dateaccepted_atis_amendmentamendment_nosuperseded_byis_xbrlis_inline_xbrlis_xbrl_numericis_auditedprimary_documentsizefile_numberactThe core table. Every standardized financial fact extracted from every filing. Supports point-in-time queries via accepted_at, quarterly derivation via derived_quarterly_value, and Bloomberg Option-C view (value_current vs. value_as_filed) for restated values.
Show 19 columns ↓Hide columns ↑
fact_identity_idaccession_idconceptstandard_conceptnumeric_valuederived_quarterly_valuevalue_currentvalue_as_filedfirst_filed_atrestatedunitreporting_currencyfiscal_yearfiscal_periodperiod_endperiod_span_daysis_cumulativeaccepted_atPre-computed intrinsic value estimates per entity. Multiple model_type rows coexist per (entity_id, valuation_date): 'dcf', 'dcf_fcf', 'ddm'. Recomputed each pipeline run; not point-in-time.
Show 10 columns ↓Hide columns ↑
entity_idvaluation_datemodel_typeper_share_valuecurrent_pricemargin_of_safetyvaluation_labeldiscount_rategrowth_rateterminal_rateDefinitions for every standard_concept used in the fact table — human_name, definition, unit_type, balance_type, and source_reference (US-GAAP taxonomy reference).
Show 6 columns ↓Hide columns ↑
standard_concepthuman_namedefinitionunit_typebalance_typesource_referenceHistorical index constituents (SP500, RUSSELL1000, RUSSELL2000, RUSSELL3000). Keyed on cik (since migration 0015). [) interval semantics. JOIN references on cik = cik to attach company metadata.
Show 10 columns ↓Hide columns ↑
cikindex_nameeffective_dateremoval_dateannouncement_dateremoval_announcement_dateremoval_reasonsuccessor_ciksourceconfidenceDerived flat join of entity + security. One row per security. Eliminates 2-table joins for company metadata. The starting point for any cross-company analysis. For index membership (current or historical), JOIN with index_membership on cik = cik.
Show 14 columns ↓Hide columns ↑
ciksymbolnamesectorindustryexchangemicis_activevalid_fromvalid_tosic_codeentity_typefigicomposite_figiPipeline-computed financial ratios per entity per fiscal period. Append-on-restatement: accepted_at is the PIT vintage (max accepted_at of the input facts) — filter accepted_at <= as_of then take the latest vintage to avoid look-ahead bias. Filter by category for grouped screens.
Show 13 columns ↓Hide columns ↑
entity_idratio_namecategoryvalueconfidence_scoreunitperiod_endfiscal_yearfiscal_periodis_ttmaccepted_atcomputed_atingested_atKnow exactly what was known, and when
Every fact in the dataset carries a accepted_at timestamp — the exact date and time the SEC accepted the filing that introduced that fact.
This is critical for backtesting. Without point-in-time data, your 2015 backtest unknowingly uses data that was only available in 2016 (look-ahead bias). The result: inflated returns that evaporate in live trading.
Concrete Example: AAPL Q1 FY2020
Point-in-time query with the Python SDK
Fetch AAPL revenue as it was known on a specific date
Show the SDK query ↓Hide the query ↑
from valuein_sdk import ValueinClient, ValueinError sql = """SELECT r.symbol, f.filing_date, f.period_end, f.accepted_at, fa.numeric_value / 1e9 AS revenue_billions, f.form_typeFROM references rJOIN filing f ON f.entity_id = r.cikJOIN fact fa ON fa.accession_id = f.accession_idWHERE r.symbol = 'AAPL' AND fa.standard_concept = 'Revenues' AND f.form_type IN ('10-Q', '10-Q/A') AND f.accepted_at <= '2020-05-01'ORDER BY f.period_end DESC, f.accepted_at DESCLIMIT 5;""" try: with ValueinClient() as client: df = client.run_query(sql) print(df)except ValueinError as e: print(f"Valuein error: {e}")Why this matters
- Eliminates look-ahead bias in walk-forward backtests
- Supports event studies around filing dates
- Amendment tracking shows original vs. restated values
- Reproduces any historical research state exactly
Every company that ever filed. Including the ones that failed.
Most financial datasets only include companies that are still active today. That means your backtest never considers the Enrons, the Lehmans, the RadioShacks — companies that went bankrupt and dragged portfolios down.
The result? Inflated historical returns that don't replicate in live trading. Academic research estimates survivorship bias overstates annual returns by 1-2 percentage points.
Valuein includes every entity that ever filed XBRL financial statements with the SEC. Delisted, bankrupt, acquired, merged — they are all here, with their complete filing history up to the date they ceased operations.
Point-in-time and survivorship-free are the headline guarantees; the trust & security overview covers provenance, zero-retention, and reliability, and the methodology documents exactly how each is constructed.
Universe Composition
~50% of the 19,000-entity universe consists of companies that are no longer actively trading. Excluding them fundamentally distorts any historical analysis.
Notable companies in the full universe
All with complete financial statements through their final SEC filing.
11,966 XBRL tags. 292 standardized concepts.
The XBRL taxonomy is sprawling. Apple reports revenue as RevenueFromContractWithCustomerExcludingAssessedTax. Older filings use SalesRevenueNet. Some companies create custom extensions entirely. Cross-company analysis becomes impossible without standardization.
Valuein maps every raw tag to a canonical standard_concept — while preserving the original tag in the taxonomy_guide table. No black box. Full provenance.
| Raw XBRL Tag | Standardized Concept | |
|---|---|---|
us-gaap:RevenueFromContractWithCustomerExcludingAssessedTax | Revenues | |
us-gaap:SalesRevenueNet | Revenues | |
us-gaap:Revenues | Revenues | |
us-gaap:SalesRevenueGoodsNet | Revenues | |
us-gaap:RevenueFromContractWithCustomerIncludingAssessedTax | Revenues | |
custom:TotalNetRevenues | Revenues |
This is just one concept. Revenue alone has 80+ raw XBRL synonyms across the filing universe. The taxonomy_guide table documents every mapping — browse it in the Data Catalog.
Original and restated values, side by side
When a company files a 10-K/A or 10-Q/A, it is restating previously reported financial data. Most datasets silently overwrite the original values. Valuein keeps both.
The original filing and the amendment each have their own accepted_at timestamp. The is_amendment flag on the filing table distinguishes them. You can query original-only, amended-only, or compare both.
- 10-K/A: Amended annual report — restated annual financials
- 10-Q/A: Amended quarterly report — restated quarterly financials
- Both original and restated values stored with distinct accepted_at
- is_amendment flag on the filing table for easy filtering
Query both original and restated values
Compare a company's original 10-K with its amendment
Show the SDK query ↓Hide the query ↑
from valuein_sdk import ValueinClient, ValueinError sql = """SELECT r.symbol, f.form_type, f.filing_date, f.accepted_at, f.is_amendment, fa.standard_concept, fa.numeric_value / 1e9 AS value_billionsFROM references rJOIN filing f ON f.entity_id = r.cikJOIN fact fa ON fa.accession_id = f.accession_idWHERE r.symbol = 'XYZ' AND fa.standard_concept = 'Revenues' AND f.form_type IN ('10-K', '10-K/A') AND f.period_end = '2023-12-31'ORDER BY f.accepted_at ASC;""" try: with ValueinClient() as client: df = client.run_query(sql) print(df)except ValueinError as e: print(f"Valuein error: {e}")Coverage at a glance
Filing Types Covered
10-KAnnual report
10-QQuarterly report
8-KCurrent report (material events)
20-FAnnual report (foreign private issuers)
10-K/AAnnual report amendment
10-Q/AQuarterly report amendment
Tier Breakdown
S&P500 · 500+ tickers · 1993–present
60 req/min · 1,000 req/hr
Active + delisted US universe · 19,000+ entities · 15-year history (2011→present)
100 req/min · 3,000 req/hr
Two datasets: fundamentals (111M+ facts, 19,000+ entities, 1993–present) + smart-money (~78M rows · 6 tables: Forms 3/4/5/144 + 13F/13D/13G)
300 req/min · 10,000 req/hr
| Tier | Data scope | Rate limit | Price |
|---|---|---|---|
S&P500Free | S&P500 · 500+ tickers · 1993–present | 60 req/min · 1,000 req/hr | Free |
ProPopular | Active + delisted US universe · 19,000+ entities · 15-year history (2011→present) | 100 req/min · 3,000 req/hr | $49/mo |
Institutional | Two datasets: fundamentals (111M+ facts, 19,000+ entities, 1993–present) + smart-money (~78M rows · 6 tables: Forms 3/4/5/144 + 13F/13D/13G) | 300 req/min · 10,000 req/hr | $499/mo |
Start querying 111M+ facts today
Register free to access the full S&P500 universe — no credit card required. Pro full-universe + 15-year history at $49/mo. Institutional with smart-money data (insider + institutional ownership) + webhooks + redistribution at $499/mo.
Also available via direct Parquet download.