Compliance rules your product can actually cite.
Minimum wage, final paycheck timing, breach notice, paid sick leave, cottage food, short-term rentals — pulled from official government sources, every rule carrying its statute citation and a link to the source it came from. For the teams already maintaining these tables by hand.
What it is
The table you've been maintaining by hand.
If your product touches payroll, HR, hiring, food, lodging, or customer data, somewhere in your codebase there's a spreadsheet of state rules that somebody has to keep current. RuleFeed is that table, kept current for you — structured, versioned, and cited back to the government source, so your support team can answer "where did this number come from?" without a research project.
- Jurisdiction
- West Virginia
- Rule
- Final paycheck timing — voluntary vs. involuntary separation
- Value
- Wages due for work performed before separation are payable on or before the next regular payday, whether the employee is discharged or quits.
- Citation
- W. Va. Code § 21-5-4
- Source
- code.wvlegislature.gov — the statute text itself
- Independent review
- ✓ passed — two models, no disagreement
That's a real published row, receipt and all.
The review
Two models. Both have to agree.
A model that invents a citation will confirm it just as confidently.
That's the quiet problem with AI-built reference data: the same system that wrote the row usually checks it, and it cannot see its own blind spots. Agreeing with yourself isn't verification.
So RuleFeed doesn't do that. One AI employee gathers each rule and cites it. A second AI employee — deliberately a different model family — reviews that row adversarially, tasked with refuting it rather than approving it: is the citation real and specific, is the source an official government page, does the rule actually say what the row claims. A rule reaches the feed only if both agree.
Right now 10 rules are held out of the feed because the second reviewer wouldn't pass them. You'll never see those — that's the point.
Disagreement is logged, not buried
Every verdict lands in a separate ledger the builder can't edit, and rides along on the published row so you can see what was checked and when.
Held rules stay held
A held rule is dropped from the feed automatically, and stays out until it's re-sourced and re-reviewed. Unreviewed rules are held too — silence never ships.
Bad sources get caught
Real catches so far: a rule sourced to a law-firm blog, a city rule citing a handbook superseded four years earlier, and citations pointing at a bill that was only ever introduced.
The receipt travels with the data
Statute citation, source URL, the date it was last verified, and the review verdict — on every row, so an auditor's question has an answer.
What's in it
Fifteen topics, ninety-one jurisdictions.
Deepest where small businesses get tripped up most, and growing.
Need a topic that isn't here? Say so when you sign up. The list is driven by what the people on it actually need — and adding a topic is a matter of days, not quarters.
Straight talk
What's real today, and what isn't.
You're being asked for your email, so here's the honest state of things rather than the roadmap dressed up as a product.
Working today
- 330 published rules across 91 jurisdictions, each with a statute citation and source link
- The two-model independent review, running as a gate before anything publishes
- A versioned export with a changelog, so you can diff one release against the last
- Every row dated, so you always know how fresh it is
Not built yet
- Live change alerts — we watch for changes on a schedule, not the moment a legislature moves
- A hosted query API — today the feed publishes as structured files, not an endpoint you call
- Sub-state coverage beyond the cities already listed
- Human legal review — everything here is machine-built and machine-reviewed
This is triage-grade data, not legal advice. RuleFeed is built to give your team a cited head start instead of a blank page — not to hand down a legal conclusion. It isn't a substitute for a lawyer, and the compliance call in your product stays yours. We'd rather say that plainly now than in a footnote later.
Get early access
Tell us what you're building and which rules you're maintaining by hand. Early access is free, and the people on this list decide what gets built next — starting with whether change alerts are worth building at all.
This opens an email to us — a real person reads it and answers. We'll only write to you about RuleFeed, and you can tell us to stop at any time. No payment, no card, nothing to cancel. Rather write it yourself? info@templeaisolutions.com
Who's behind it
Built by a studio that runs on this.
Temple AI Solutions builds custom AI employees for businesses — and runs its own companies the same way. RuleFeed came out of that work: we needed compliance data we could actually stand behind, found that checking AI work with the same AI doesn't count, and built the two-model gate to fix it. See the rest of our work →