Methodology

Sources

• Structured filing data: the U.S. Department of Labor's Form 5500 datasets (Form 5500, Form 5500-SF, Schedule H and Schedule C), loaded per plan year. When a plan amends a filing, the latest version replaces the original.
• Audit reports: for larger plans, the independent auditor's report attached to the filing PDF on EFAST2. Its notes describe the match, vesting and other plan features.

We serve retirement plans only (no health and welfare plans). Figures are labelled with the plan year of the filing they come from, not the date we published them. Filings arrive 7 to 10 months after a plan year ends, so the newest data is usually one to two years old.

Employers and plans

Each plan is identified by its sponsor's EIN and its plan number. We group plans under the employer name people search for (for example, a sponsor filing as “Example Holdings, Inc.” under “Example”), with manual overrides where names are ambiguous. An employer page shows every plan we could link to it, each as its own section.

What gets a page

An employer gets a full, indexed page only when at least one current plan has three or more data points beyond its identity: employer contribution per participant, administrative expense ratio, a recordkeeper from Schedule C, or published match, vesting or enrollment details. Other employers still appear in search and in the old 401(k) finder.

Match and plan features: the filing's own words

Employer pages are quote-first. For the employer match, and for vesting, automatic enrollment, Roth contributions and loans, we show the sentences from the notes in the independent auditor's report exactly as they appear in the filing, with the plan year, the PDF page and a link that opens the filing at that page. We don't paraphrase, shorten or correct the quoted text.

1. We fetch the filing PDF from EFAST2 and locate the auditor's opinion and the notes to the financial statements.
2. A language model reads only those pages, points to the sentences that describe each feature and gives the page number. The quoted passage is then extended to whole sentences from the page text by a fixed rule, not by the model.
3. Quote check. A deterministic check confirms the passage appears word for word on the cited page. Only verified passages that describe employer contributions are published. A passage that fails the check isn't shown at all.
4. The model also proposes parsed values: the formula (“100% of the first 4% of pay”), the maximum match, the vesting schedule and so on. In a frozen out-of-sample test these were right about 88% of the time, which is not good enough to publish unchecked. So parsed values are shown only after a person has reviewed them (see below), marked “Reviewed”.
5. If a feature isn't shown, we either couldn't find a verified passage or it is awaiting review. “Not shown” never means “no”.

The quoted text describes the plan as of the filing's plan year. Plans change, so check your current Summary Plan Description before relying on it. Search engines and AI assistants that summarize our pages see the same quote; the page's description and FAQ answers repeat the filing's words, not a machine-read number.

Maximum match (reviewed rows only) is the most the employer adds as a percent of pay when you defer enough: 100% of the first 3% plus 50% of the next 2% is 4%. Discretionary matches show the rate the employer declared for that year when the notes state it. Matches capped as a share of the IRS deferral limit or in dollars are shown as written and are not converted to a percent of pay. Rankings and averages such as the best 401(k) matches use only reviewed formulas, and appear once enough have been reviewed.

Human review

Before anything is shown, automated checks decide which rows can be published at all. Quality gate. Rules check that each quoted sentence appears on the cited page, that tiers are sensible, that the formula is consistent with the employer contributions reported on Schedule H, and that the recordkeeper agrees with Schedule C. Automated adjudication. Rows the gate flags are read again by a second, independent model pass that compares each value with the quoted sentences; rows it can't confirm stay unpublished.

A person then works through a review queue, highest-traffic employers first, reading each parsed value against the filing. A value they confirm is shown with a Reviewed mark; a value they correct is replaced with the corrected one and marked reviewed; a value they reject is taken off the site. Until then the page shows only the filing's words and says that our summary appears after review. Corrections you send us are checked against the filing the same way.

Metrics

What we don't do

• We don't rate or recommend plans, and nothing here is financial advice.
• We publish no information about individual participants. Form 5500 contains none, and we keep it that way.
• Rankings and data are never paid placements. Affiliate links, where they appear, are labelled (see our affiliate disclosure).

Corrections

If something is wrong, tell us. A person checks the filing and fixes or removes the field.

General information from public sources, not financial, tax or legal advice.