Agentic Maneuverability Score: Difference between revisions

The '''Agentic Maneuverability (AM) Score''' is a protocol for cognitive resource allocation within [[AetherOS]]. Its primary function is to determine the "virtuousness" of deploying a specific AI agent for a specific task, given the real-time state of the system's available resources.

The score is the output of a calculation that treats the agent, the task, and the system not as abstract software entities, but as components in a single cyber-physical system. It grounds abstract computational requests in the concrete realities of supply and demand, drawing direct inspiration from the principles of [[Energy–maneuverability theory|Energy-Maneuverability Theory]].

 

 

The AM protocol is designed to prevent the misuse of agents by ensuring the system only commits to actions that are '''virtuous'''. A virtuous action is one where the capabilities of the agent are well-matched to the demands of the task, and the capacity of the system is sufficient to support the effort without compromising its own integrity.

The AM Score protocol is a direct implementation of the '''[[OODA Loop]]''', a strategic framework for rational decision-making in dynamic environments.

* '''Observe''': The system gathers real-time data on three components:

** The '''[[System Maneuverability Score]]''' (Y): The available computational and energy capacity of the host system.

** The '''Agent''' (Z): The intrinsic characteristics of the selected AI model, drawn from the [[Agentic Maneuverability Almanac]].

** The '''Task''' (X): The projected demands of the proposed mission, drawn from a registry of task profiles.

* '''Orient''': The system applies the Universal Agentic Maneuverability Equation to synthesize the observed data into a single, coherent metric: the '''Task Load'''. This orients the system by providing a unified understanding of the mission's cost.

* '''Decide''': The final '''Virtuousness Score''' (System Maneuverability / Task Load) provides the basis for a clear, threshold-based decision.

The core of the protocol is the '''Universal AM Equation''', which calculates a projected '''Task Load'''—a dimensionless quantity representing the holistic stress a given mission will place on the system.

* '''P''' is the size of the agent's underlying model in '''billions of parameters'''. This represents the agent's "cognitive mass" or inertia.

* '''W_p''' is the '''Parameter Weight''', a tunable constant of the ecosystem that scales the influence of model size.

* '''T_o''' is the '''Expected Output Tokens''', representing the anticipated volume of cognitive work required by the task.

* '''W_t''' is the '''Token Weight''', a constant scaling the cost of generating each unit of output.

* '''F_c''' is the '''Cognitive Load Factor''', a multiplier derived from a qualitative assessment of the task's intrinsic difficulty (e.g., "low", "medium", "high").

* '''V_c''' is the agent's '''Cognitive Velocity''', measured in tokens per second (tps) on a baseline hardware profile. This represents the agent's raw processing speed or "thrust."

The final '''Virtuousness Score''' is then calculated by comparing the system's capacity to the projected load:

A score '''greater than 1.0''' indicates a virtuous action, meaning the system has ample capacity to handle the load efficiently. A score '''less than 1.0''' indicates the action will stress the system, consuming more resources than are sustainably available and potentially leading to degraded performance or instability.