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 ValidatorValidates spatial boundaries and node adjacency at every routing step, preventing entities from crossing impassable boundaries or exceeding capacity constraints.
Supported bySP-002 Geometric Boundary Enforcer - 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 TrackerPreserves 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.
Supported bySP-007 Trajectory Predictor - 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 EscalatorEnforces temporal deadlines and decay timelines as hard constraints on the routing solution, escalating conflicts before they result in violations.
Supported byTE-005 Duration Realist
The Evidence
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.
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.