The Prova Method · The Architecture
What makes evidence credible, or fragile.
Whether a finding holds up or falls apart is rarely luck. It turns on a specific set of things, and we work each one deliberately, from how a study is designed to how the whole thing stays honest and pays for itself.
The eight layers
What we do to make a finding hold up.
1 · Designing the study
Together we settle what would count as a real effect, and we lock it before any data arrives. That is what keeps a finding honest: no one gets to choose the analysis once they can see which answer it would produce.
2 · Making the data
Most of what a program needs to measure itself already sits in its case notes, attendance, and service logs. We build the measurement into those records, so running the program well does most of the work of measuring it. The evidence comes out cheaper, continuous, and harder to game.
3 · Fitting the method to the place
The standard of proof holds everywhere. The design, the instruments, and the approach get configured to each setting, because a study that works in one place can fail in another for reasons that have nothing to do with rigor, usually political and institutional. Reading those reasons correctly is a large part of the job.
4 · Keeping the judgment human
Reading a context honestly, holding the trust of the people whose data it is, and judging what an effect means here are work only people can do. AI does the heavy lifting around them; the judgment stays human. The whole method is built around that judgment, and the people who hold it.
5 · Compounding with use
Each piece of work leaves something behind: real estimates of how programs behave, the effect sizes and attrition rates you need to design the next study well. The method is built so the tenth study is sharper than the first. We are early, and we will report whether that holds.
6 · Making honesty the rational move
We do the things that would be irrational for anyone who wanted to manipulate the outcome: we pre-register, we report a finding even when it disappoints, and we invite independent checks. The honesty is built into the structure, so telling the truth stays the sensible move even under commercial pressure.
7 · Holding the floor
A serious method has lines it will not cross, whoever is paying and whatever a setting permits. Where it is not legitimate to randomize, to collect, or to claim, we decline, and we say why. Knowing those lines in advance is part of the rigor.
8 · Tracking our own cost of proof
The bet is that the cost of credible proof can fall far enough to expand who can afford to know. We treat that as a claim still to be earned, so the method measures its own cost and reports whether it actually falls. If it does not, we say so.
What holds it up
The three structures everything else depends on.
Eight layers show the full surface. Three of them are load-bearing: if one fails, a whole capability goes with it. Those three are where we spend the most care.
The integrity backbone
Pre-registration, honest reporting, and costly signals work as one chain: each is hollow without the others. Together they are what make a finding trustworthy under pressure. This one is non-negotiable.
The adaptation backbone
The judgment that reads a context, the discipline that turns that reading into the right configuration, and the trust that makes the work possible at all. Together they let one method hold in very different places without losing its identity.
The compounding backbone
Synthesis across engagements, fed back into the method and calibrated by what we have learned. It is what lets the method compound: each engagement makes the next one sharper.
Going further
The architecture runs deeper than this.
There is more beneath each of these layers than a page can hold.
To go further into any of it, that’s a good conversation to have.