The problem this addresses
A state's website shows you the law as it stands today. It rarely shows you what the same provision said in the year you are being asked about, where it lived before it was renumbered, or which amending act moved it. Anyone reconstructing that reads the current text and hopes.
The gap is not that the answers are hard to find. It is that a lookup which returns nothing looks exactly like a lookup that returns an answer, and a provision that was renumbered into a different subject looks like a provision that never changed.
Current work
Unclaimed property
Uncashed payroll and the business-to-business exemption in Delaware, Illinois, Minnesota, Virginia and California, with every amending act in the window read rather than summarized. Written for the practices that defend examinations, where the year under audit decides the answer.
Large load and data center electricity tariffs
Terms extracted from utility commission proceedings so they can be compared across states: eligibility thresholds, contract terms, minimum take, exit fees, collateral, capacity reduction. One row per instrument per stage, because a proposed term and an ordered term are different facts and the difference is what a rate case turns on.
Water withdrawal and allocation
Permitting and allocation records in Arizona, Minnesota and Virginia. Earlier stage than the other two, and described that way rather than dressed up.
How the work is checked
Every quoted span is re-read off the source before publication. A citation that names a page which does not contain the quote fails the build, so a mis-cited value cannot ship. This is mechanical, not a promise.
What is missing is declared, not omitted. A document that could not be obtained stays visible as a gap with a stated reason. “We did not look” is not a reason.
Absences are recorded as findings. Where a provision does not exist, the record says so and names what substitutes for it, because an empty cell and a deliberate silence are different facts.
The reason for the machinery is simple: a comparison table is worth something only if a reader can check any cell in it against the document it came from. A ninety-five percent accurate compliance dataset is worth approximately nothing, because the reader cannot tell which five percent they are holding.
What this is not
- Not legal advice. The work reports what instruments say; it does not advise on how a regulator would apply them to particular facts.
- Not a database of everything. Each piece states its scope on its own front page: which states, which provisions, which years. The scope is narrower than the subject.
- Not automated extraction dressed as research. Values are read by a person and then checked by machine, in that order.