Notes on historical options data.
We only publish what we have actually measured. These are the places data goes wrong — survivorship, split adjustment, snapshot timing, and price — with the numbers behind each claim.
Survivorship bias in options backtests: the ~35% you can't see
Five years of the breadth universe, counted: 1,573 names on 2020-03-16, 1,521 on 2026-10-30. We probed the old list against recent sessions and 546 of the original names no longer quote. That is the fraction of your "full market" that a survivor-only dataset quietly drops. Here's what it does to a backtest, and how to test whether you're affected.
Read it →How option strikes get adjusted for a 10:1 split (NVDA, 2024, step by step)
On 2024-06-10 the 10.00 call was re-listed as a new contract at 1.00 and a fresh 10.00 series appeared alongside it. Most feeds won't tell you the old series changed. We walk the exact rows, the mapping rule to carry a position across the ex-date, and why "adjusted" is a lie if the feed has no marker for it.
Read it →End-of-day options: what "the 15:45 close" means — and what it doesn't
Every EOD option file claims a "close" snapshot. Ours is the last consolidated top of book at or before 16:00 ET, stamped 16:00:00. That is a mark, not a fill. This is what that distinction costs you when you backtest execution, and why we stamp the time instead of pretending.
Read it →Where to get historical US options data in 2026 (and what it will cost)
DoltHub, a public Parquet mirror, Kaggle, OptionsDX, and the paid vendors — mapped against what each actually has (start date, fields, delisted coverage, size). Then the honest math on what a 2008-2025 full chain is worth, and when it is worth paying for at all.
Read it →