How the data is made.
Every number in 100bps starts life in a government filing and ends on your screen with its provenance intact. This page walks the whole path. It exists because our first principle — the record & from the record — is only worth something if you can inspect how the record is handled.
1. The raw record
The foundation is the U.S. Department of Labor's Form 5500 series, published through EFAST2 — the annual filing every covered retirement plan must make — together with SEC and state investment-adviser registration filings (Form ADV). We work from hundreds of thousands of filings across multiple filing years, including amendments and superseding filings, not just the latest snapshot. The filing years and "data through" dates are printed in the product, on every surface.
2. Untangling who is who
Filings name service providers however the filer typed them: we've seen 825 spellings of Fidelity and 22 of CAPTRUST. Provider names are resolved to canonical entities across ~2.9M records using pattern matching, identifier clustering, and automated classification — currently 91% of 106,994 provider names resolved. The remaining 9% stay unresolved and are shown as filed, not guessed.
3. Decoding the money
Schedule C service-provider compensation is reported in coded categories that mix advisors, recordkeepers, and asset managers. We split those roles apart using the filed service codes — so an "advisor" in 100bps means the filing says advisor, and fee ratios are computed from filed compensation over filed assets. The arithmetic is traceable: every computed field's inputs are visible in the filing record it came from.
4. Reading the fund menus
For plans that file the detail, the fund lineup lives inside Schedule H attachments — as documents, not data. We extract those holdings with automated document extraction, then validate every extracted lineup against the plan's own filed balance sheet before it earns a place in the product. Lineups that fail validation are excluded, not approximated. Where no fund menu is filed, the product says so — marked, not hidden.
5. Keeping time
Filings amend, supersede, and contradict each other across years. We keep the full chain — which filing is canonical, what changed between years, when the advisor line went quiet — because the signal lives in what moved. Temporal context is stored as first-class data, not recomputed on the fly.
6. What never happens
- No estimates or modeled guesses presented as data.
- No contact lists bought from data brokers.
- No customer information ever sent to an AI provider — automated extraction reads public filing attachments only.
- No number on a marketing page that isn't regenerated from the same database the product runs on.
We show our work — you make the call. That's the product.