KiRecommendedSpatial

Logistics & Supply Chain

The Problem

The route passes through a segment the vehicle physically cannot traverse. Bridge heights, road widths, and weight limits are not encoded in the token sequence. The model plans in description space where every path is syntactically valid, and physical constraints only surface at execution time.

How Ejentum Solves It

One API call forces your model to validate every intermediate state between start and end positions, enforcing physical constraints that text-based planning ignores.

The Failures

  • 01

    The Pattern

    Routes pass through geometrically impassable segments or violate infrastructure clearance limits

    Why It Happens

    The model plans in description space where routes are sequences of location names. Physical constraints like bridge heights, road widths, and weight limits are not encoded in the token sequence.

    The Resolution

    SP-001Topology Validator

    Validates spatial boundaries and node adjacency at every routing step, preventing entities from crossing impassable boundaries or exceeding capacity constraints.

  • 02

    The Pattern

    Full-speed direction changes planned for loaded vehicles that physically cannot execute them

    Why It Happens

    Momentum and inertia are continuous physics that the model treats as discrete decisions. "Turn left" is a valid token regardless of the vehicle's mass, speed, or load distribution.

    The Resolution

    SP-018Kinetic Momentum Tracker

    Preserves inertia and momentum in every spatial calculation. A 20-ton truck at highway speed cannot execute a 90-degree turn, and the agent is not allowed to assume otherwise.

  • 03

    The Pattern

    Delivery windows, perishable decay timelines, and driver-hours regulations ignored in schedule optimization

    Why It Happens

    Temporal constraints interact with spatial routes nonlinearly. A route that is optimal by distance may violate delivery windows. The model optimizes one dimension without holding the others constant.

    The Resolution

    TE-011Deadline Escalator

    Enforces temporal deadlines and decay timelines as hard constraints on the routing solution, escalating conflicts before they result in violations.

The Evidence

+7.4pp on spatial tasks

EjBench, 30 spatial tasks

Route planning has single correct answers constrained by physics. A single scaffold that enforces intermediate state validation outperforms multi-ability injection. Haki regressed -2.8pp on spatial tasks in EjBench: four perspectives confused constraint tracking.

EXT-SP-200.1430.857 Haki

Task required tracking object locations through perspective shifts in a narrative. Baseline attributed knowledge to a character who could not have observed the event. Haki anchored each character belief state to what they actually witnessed. Correctness flipped from 1/3 to 3/3.

Run your next route optimization through the API. See how the scaffold catches the boundary violation your planner missed.