Module shadow_planning 

Side-effect-free, defense-aware shadow planning.

This layer shows how the current plan would change under an observed defensive posture, without changing anything. It filters one already authorized plan into a dependency-safe subsequence and computes an explainable delta against the current plan. It never issues a request, mutates the planner, runtime, knowledge, or experience state, or reorders the real plan.

Defense evidence is not a second planner: it never adds an action and never raises an action’s utility. For each existing candidate it only decides to allow, deprioritize, or suppress. The per-observation recommendation is the existing recommend policy — this module reuses it and adds only the resource-scoped aggregation and the monotonic class mapping.

Structs

DefenseAwareShadowPlan

The current plan, the defense-aware shadow plan, and the delta between them.

PlanAdjustment

One action kept but deprioritized in the shadow plan.

ResourceDefenseObservation

One observed response leg for a resource, with its supporting evidence.

ResourceDefenseSignal

A resource-scoped, corroborated defense signal.

ShadowPlanDelta

The explainable difference the defense signal makes to the current plan.

SuppressedAction

One action removed from the shadow plan.

Enums

DefenseInteractionClass

Closed classification of how an action interacts with a target.

InteractionDecision

What defense recommends doing with one existing candidate.

Functions

decide

The monotonic, exhaustive mapping from a recommendation and an interaction class to a decision. This is the single place the mapping lives.

defense_aware_shadow_plan

Computes the current plan, a defense-aware shadow plan, and their delta.

explanation_code

A stable, replay-safe explanation code for one decision.

render_explanation

Renders a stable explanation code as human-readable prose.