Upload folder using huggingface_hub

LICENSE

@@ -0,0 +1,33 @@
+Tradevo Data — free sample licensing
+
+DATA (data/ directory): CC0 1.0 Universal
+The dataset files are dedicated to the public domain under CC0 1.0
+(https://creativecommons.org/publicdomain/zero/1.0/). They are derived
+from U.S. SEC EDGAR filings, which are public domain. Use them for
+anything, no attribution required (though a link is appreciated).
+
+CODE (everything else, including query_asof.py): MIT License
+
+Copyright (c) 2026 Tradevo Technologies Inc.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+Note: the data is provided as-is, without warranty of accuracy or fitness.
+It is not investment advice. Verify against primary SEC filings before
+relying on it.

METHODOLOGY.md

@@ -0,0 +1,95 @@
+# Tradevo Data — Methodology
+
+*Point-in-time US equity fundamentals, built honestly from SEC EDGAR.*
+
+A small, honest fundamentals dataset built entirely from free SEC EDGAR filings.
+Every value is stamped with the date it *first became public*, so a backtest can only
+ever use what was actually knowable at each point in time.
+
+This document exists because, with data, **transparency is the product.** If you can't
+see how it was built and validated, you can't trust it — so here's exactly how it's built.
+
+---
+
+## Problem 1 — Lookahead bias (the silent backtest killer)
+
+A company's fiscal year has a *period end* (e.g. Apple's FY2024 ended **2024-09-28**) but
+the numbers aren't *public* until the 10-K is filed weeks later (Apple's was filed
+**2024-11-01**). A backtest that joins fundamentals on the period-end date is using
+information that didn't exist yet — it's peeking into the future.
+
+In this sample, on rows with a reliable filing date, fundamentals became public on
+average **43 days after** the period ended (max 61). That's the future-peek a naive
+join silently grants you on *every data point*. It makes strategies look better in
+testing than they are in life.
+
+**Live example, straight from the data:**
+
+| As of date | Newest AAPL annual revenue you could *honestly* know | Filed |
+|---|---|---|
+| 2024-10-15 | FY2023 — $383.3B | 2023-11-03 |
+| 2024-11-15 | FY2024 — $391.0B | 2024-11-01 |
+
+Same company, one month apart, a different "latest known" number. This dataset encodes
+that; a naive one hands you FY2024 too early.
+
+## Problem 2 — Dirty / inconsistent raw data
+
+EDGAR is free but messy: companies tag the same concept under different XBRL labels and
+switch them over time (e.g. revenue under `Revenues` in older years, then
+`RevenueFromContractWithCustomerExcludingAssessedTax`). A naive pull locks onto whichever
+tag it sees first and can return a *stale year* or the *wrong number*. We resolve tags
+to the most recent reporting and validate every row (see below).
+
+---
+
+## How it's built
+
+1. **Source** — SEC EDGAR `companyfacts` API (`data.sec.gov`). Public domain, free, no key.
+2. **Annual figures** — only `10-K` filings; for flow items (revenue, net income) only
+   ~full-year durations (345–385 days) are kept, so quarters and stub periods can't leak in.
+3. **Tag resolution** — among candidate XBRL tags for each concept, we choose the one whose
+   data extends to the *most recent* period (ties broken by priority). This kills the
+   stale-tag bug.
+4. **Point-in-time stamping** — `first_filed` = the *earliest* filing that reported a given
+   period. That is the first date the number was knowable. `lag_days` = first_filed − period_end.
+5. **Restatement detection** — if a later filing revised a period's value by >0.5%, the row is
+   marked `restated = True`, and we keep both `original_value` (first knowable) and
+   `latest_value` (most recent).
+6. **QA + reliability** — every row is checked (filing-lag range, value magnitude). Rows whose
+   only available filing is a much-later one (common for the oldest years, where the original
+   10-K predates XBRL) are marked `filed_reliable = False` rather than shipped with a
+   misleading date.
+
+## Columns
+
+| column | meaning |
+|---|---|
+| `ticker`, `cik` | company identity |
+| `concept` | Revenue / NetIncome / OperatingCashFlow / EPSDiluted / DilutedShares / Assets / StockholdersEquity |
+| `xbrl_tag` | the exact SEC tag the value came from (full provenance) |
+| `fiscal_year`, `period_end` | the period the value covers |
+| `first_filed` | date it first became public (the point-in-time stamp) |
+| `lag_days` | first_filed − period_end (the lookahead gap) |
+| `filed_reliable` | True if the original 10-K is in XBRL (trust the date) |
+| `original_value` | value as first reported |
+| `latest_value` | value as most recently reported |
+| `restated` | True if later revised >0.5% |
+| `qa_status` | `clean` or the specific flag raised |
+
+## This sample's coverage
+
+- 40 large-cap US companies, 7 concepts (revenue, net income, operating cash flow, diluted EPS, diluted shares, assets, equity), up to 12 fiscal years each
+- 3,212 point-in-time rows; revenue history depth averages 11.7 years
+- 3,152/3,212 rows carry a reliable filing date (mean lag 43.4 days, max 61); 60 oldest-year/edge rows flagged for resolution
+- 187 restatements detected (same-tag revisions >0.5%, including 10-K/A amendments)
+
+## Known limitations (we mark them, we don't hide them)
+
+- **Oldest-year filing dates**: 60 rows where only a later XBRL filing exists; flagged, not faked.
+  (Resolvable by cross-referencing the EDGAR submissions index — on the roadmap.)
+- **Annual only** for now; quarterly (10-Q) point-in-time is the next build.
+- **Banks/insurers**: "revenue" is an approximate concept for financials; treat JPM-type names with care.
+- **40-company sample**: the full US universe is a pipeline run away — gated on demand, not on capability.
+
+The entire pitch is the line above each of these: a clean dataset *tells you what it doesn't know.*

README.md

@@ -0,0 +1,62 @@
+---
+license: cc0-1.0
+language:
+  - en
+pretty_name: Point-in-Time US Equity Fundamentals (Sample)
+tags:
+  - finance
+  - stocks
+  - fundamentals
+  - backtesting
+  - sec-edgar
+size_categories:
+  - 1K<n<10K
+task_categories:
+  - tabular-regression
+configs:
+  - config_name: default
+    data_files: pit_fundamentals_history.csv
+---
+
+# Point-in-Time US Equity Fundamentals (Sample)
+
+A free, verifiable sample of **point-in-time** US equity fundamentals from SEC EDGAR — 40 large-cap companies, 3,212 annual rows. Every value is stamped with the date it *first became public*, so a backtest only ever sees what was knowable at the time. **No lookahead bias.**
+
+This is the open sample of [Tradevo Data](https://tradevodata.com); the full API covers 5,214 companies.
+
+## Why point-in-time matters
+
+Most fundamentals datasets store one row per fiscal period keyed by the *period-end* date, and silently overwrite numbers when companies restate. Joining on period-end hands your backtest data months before it was public; using restated values lets it "see" corrections that didn't exist yet. Both quietly inflate returns. This dataset fixes both, by keeping the original first-reported value and its filing date.
+
+## Columns
+
+| column | meaning |
+|---|---|
+| `ticker` | US stock symbol |
+| `cik` | SEC Central Index Key |
+| `concept` | one of: Revenue, NetIncome, Assets, StockholdersEquity, OperatingCashFlow, EPSDiluted, DilutedShares |
+| `xbrl_tag` | the exact XBRL tag the value came from |
+| `fiscal_year` | issuer fiscal year |
+| `period_end` | fiscal period end date |
+| `first_filed` | **the date this value first became public** (the point-in-time key) |
+| `lag_days` | days between `period_end` and `first_filed` |
+| `filed_reliable` | whether the filing date is high-confidence |
+| `original_value` | value **as first reported** — use this for point-in-time backtests |
+| `latest_value` | most recent revised value (may post-date `first_filed`) |
+| `restated` | true if later amended by more than 0.5% |
+| `qa_status` | per-row quality verdict (`clean` or `FLAG:…`) |
+
+## Usage
+
+```python
+from datasets import load_dataset
+ds = load_dataset("Tradevodata/point-in-time-us-equity-fundamentals-sample")
+```
+
+For point-in-time correctness, filter to `first_filed <= your_as_of_date` and use `original_value`.
+
+## Provenance & license
+
+Source: U.S. SEC EDGAR (public domain). Data is CC0-1.0. See `METHODOLOGY.md` for exactly how it was built. Not investment advice; verify against primary filings before relying on it.
+
+Full dataset + JSON API with `as_of` queries: **[tradevodata.com](https://tradevodata.com)** · methodology + code: **[github.com/christianpichichero-max/pit-fundamentals](https://github.com/christianpichichero-max/pit-fundamentals)**