Collection is easy. Rejection and context are the product.
Every Driving Virginia product uses the same discipline: define the decision, collect primary evidence, reject noise, separate fact from interpretation, and stay quiet when nothing material changed.
Define
We name the business, property, geography, decision, timing, and evidence threshold before collecting records.
Collect
Official APIs, filings, agendas, program pages, public datasets, and primary records are retrieved instead of trusting snippets.
Normalize
Sources become a common evidence model with location, dates, status, source health, limitations, and identifiers.
Connect
Related signals are grouped only when the available evidence supports a connection.
Challenge
A separate review tries to disqualify the conclusion, find contradictory evidence, and expose weak assumptions.
Deliver
The brief shows verified facts, labeled interpretation, confidence, unknowns, and one useful next step.
Source changes trigger alerts. AI wording does not.
A model can describe the same record differently from one run to the next. That is not a change in the world. Material-change detection uses official revision numbers and normalized page fingerprints.
A deadline, amount, eligibility rule, program status, or official source version changed.
The official source is unchanged, even if the generated summary would use different wording.
Human review remains in the pilot
The research engine is automated. Customer-facing judgment is reviewed while we measure usefulness, noise, traceability, and cost. Applications, lender contact, payments, and customer-profile creation do not happen automatically.