Field Nine · Decision Surface · Research environment
The environment Jev searches.
Decision Surface is the environment Jev searches. It records states, actions, trajectories, and boundary crossings, and keeps each crossing as a replayable witness.
States→ Actions→ Trajectories→ Boundary→ Witness
THREAT MODEL
THE SEARCH SPACE IS DEFINED BY WHAT THE ATTACKER CAN CONTROL.
State-space · StatesStates are recorded first.
ActionsPermitted moves branch from each state.
TrajectoriesRecorded paths: actions, observations, outcomes.
BoundaryDiscontinuity in the field, located by search.
WitnessMinimized, replayable record of each crossing.
Autonomous search, limited budget.
Multiple trajectories compete for limited search budget. Branching narrows as the budget is consumed.
ADAPTIVE SEARCH
KNOWN TESTS ENUMERATE FAILURES. SEARCH FINDS HOW THEY ARE REACHED.
Every result carries its trajectory.
Mission, target, environment, and action space are recorded alongside each result. Witnesses become regression harnesses and are retested as systems change.
SIMULATION
THE SEARCH RUNS INSIDE A CONTROLLED ENVIRONMENT.
Mission
The authorized question: what is to be searched, and under what bounds.
Target
The intelligent system under study · model, agent, or full agentic stack.
Environment
The sandbox and its constraints. Nothing runs outside authorization.
Objective
The behavior to reach, stated as a testable condition.
Action space
The permitted moves Jev may select from, observe, and adapt over.
Trajectories
Recorded paths through the space: actions, observations, outcomes.
Findings
Boundary crossings · points where behavior changed consequentially.
Evidence
Witnesses: minimized, replayable records of each crossing.
Evaluations
Regression harnesses, so findings can be retested as systems change.
WITNESS
A WITNESS IS REPLAYED AS A REGRESSION TEST.