Four phases.One closed loop.

Each phase is designed to catch the failure mode of the one before it. The loop terminates only when measurement shows the intervention moved the underlying behavior.

1. PHASE

   01

   ▸ executing

   Observe

   We begin where the behavior actually happens — in product telemetry, transaction logs, support tickets, and conversations we are explicitly invited into. No deck-first research. The first artifact is a map of where the signal lives.

   • Subroutine
     Behavior surface mapping
   • Subroutine
     Stakeholder interviews
   • Subroutine
     Evidence audit
2. PHASE

   02

   ▸ executing

   Instrument

   We redesign the sensors. Event schemas, survey instruments, observational protocols — each engineered against the decision they need to inform. The instrument is treated as a first-class artifact, versioned and reviewed.

   • Subroutine
     Schema design
   • Subroutine
     Sampling architecture
   • Subroutine
     Bias profile documentation
3. PHASE

   03

   ▸ executing

   Model

   We translate observation into structure: decision pathways, segmentations, behavioral drivers. AI is deployed against specific named judgments, not the task as a whole. Every model ships with its known failure modes.

   • Subroutine
     Hypothesis framing
   • Subroutine
     Quantitative validation
   • Subroutine
     Bounded AI synthesis
4. PHASE

   04

   ▸ executing

   Deploy

   Findings are wired into the systems that act on them. We measure whether the action moved the underlying behavior, then feed that result back into the next loop. Research that does not change behavior was diagnostic, not done.

   • Subroutine
     Decision-aligned artifacts
   • Subroutine
     Intervention wiring
   • Subroutine
     Closed-loop measurement