Labs
Rules are not enough if an agent keeps violating them. Boundary Learning Score measures whether an agent adapts to enforced boundaries: fewer repeated blocked attempts, fewer stale retries, smaller requests, better use of required_next_action, and less operator noise.
The score does not prove an agent is safe. It measures behavior under a defined boundary challenge.
Why this matters
A blocked request is not only a failure. It is feedback. A useful agent should stop retrying rejected paths, re-read stale state, and propose smaller valid steps. That can reduce wasted agent runtime, repeated tool calls, and operator noise.
01
Agents should stop retrying paths the boundary already rejected.
02
Fewer blocked attempts and stale retries mean less wasted agent work.
03
Humans should review real decisions, not repeated noise.
04
The score makes boundary feedback visible across comparable runs.
Boundary Learning Score
Raw boundary events become opportunity-normalized scoring evidence.
Raw counts are not enough. A harder run can contain more chances to make mistakes. Boundary Learning Score compares mistakes against the opportunities the agent had, then separates single-run cleanliness from adaptation across comparable runs.
The base evidence.
How many chances did the agent have to make each kind of mistake?
Raw signals are grouped into explainable score components.
A single-run view of how cleanly the agent behaved.
Did behavior improve across comparable runs?
How reliable is this comparison?
What the score measures
01
Does the agent keep trying the same rejected action?
02
Does the agent act on old state, or does it re-read before retrying?
03
Does the agent reduce blast radius and propose smaller, valid steps?
04
Does the agent follow feedback such as required_next_action before trying again?
05
Does the agent reduce noise so humans can focus on real decisions?
Guardrails
Planned lab
MazeRunner is one planned lab for this score: movement becomes impact, and every move must pass through a boundary decision.
Boundary challenge
Any small target environment can become a boundary-learning challenge. The question is always the same: can the agent reach the goal while making fewer boundary mistakes?
