Who it’s for · Public agencies
Evidence strong enough to survive scrutiny, on the timeline a decision actually runs on.
Evaluation timelines can miss budget and reauthorization windows. Public services also have to continue while evidence is produced. A finding that cannot withstand a determined reading is worth little when the budget is contested. Prova is being built to bring credible, independent evidence to programs while they run, fast enough to inform the decision in front of you.
The bind
A serious evaluation runs on a different clock than the decision.
Some impact evaluations take multiple years, especially when they require primary data, large samples, or long-term outcomes. Budget and reauthorization decisions can move faster. Evaluation can be embedded in a running service through designs such as randomized rollout, lotteries, or credible operational comparisons where the setting permits them. Routine dashboards often count activity; they do not by themselves establish causal impact.
A serious decision benefits from evidence that is independent, explicit about its limits, and available while the decision is still open. Producing that combination can be costly and slow, depending on the question and design.
What Prova does
Credible, independent evidence, built around the work as it runs.
Prova works from the administrative records a program already generates, so measurement runs alongside delivery where the records and permissions allow it. It is being built to establish the strongest claim the evidence honestly supports on a decision-relevant timeline, and to be explicit about what the evidence can and cannot carry, so a finding holds up when it is challenged. Prova Method does not run the programs it assesses, and it does not compete to win the evaluations it reviews, so a finding you can show an auditor, an opposition, or the public comes from an engagement designed to reduce and manage conflicts around its conclusion.
Some things should not be measured, randomized, or claimed, and those lines are clear before the work begins. People described by the data deserve meaningful rights, safeguards, and a voice in how it is governed and used. The applicable legal and governance rights vary by jurisdiction, contract, community, sector, and data type. AI does the labor-intensive reading and synthesis under human supervision. Prova is being built to establish how far that capability lowers the full cost across real engagements. The judgment about what a result means stays with the people who know the policy and the place.
How to start
Start inside the budget cycle.
A short, bounded first step fits inside a budget cycle: weeks, a fixed scope, one question answered well. It can tell you what an existing program can credibly claim, or stress-test an evaluation you are about to commission before the money is committed. When a program is live and worth following as policy moves, measurement can be built in as the work runs, to the same standard.
Start a conversation
Bring the decision and its deadline.
If a program is up for a decision and the evidence will not be ready in time, that is the conversation to have. The first one is a fit check, with no obligation.
Common questions
The questions a careful reader asks first.
Is cheaper evidence weaker evidence?
A lower price does not require a lower standard. Frontier AI models can now perform much of the reading, extraction, and synthesis in bounded tasks. Prova is being built to establish how far that capability lowers the total cost of reaching a given grade in its claim-evidence rubric without shifting work or risk elsewhere. Design labels do not guarantee credibility. A modest claim remains modest, and a thin record remains thin.
AI can be confidently wrong. Why trust it here?
It can, and that is a central design problem. AI does the labor-intensive reading and synthesis; people decide whether a finding holds, what it means here, and where the evidence runs out. Prova is being built so claims retain provenance and uncertainty. Its reliability across real engagements is still being tested.