Session 5 Deliverable | Author: Chase Petersen | Includes 100-Triple Domain Graph & Risk Audit
YARDMASTER's Knowledge Graph bridges physical assets (trucks, plants) with epistemic constraints (time, compliance). The system tracks 9 core entity classes through live API ingestion from Jonel (batching) and TruckTrax (telemetry).
| Entity Class | Key Attributes | Source of Truth |
|---|---|---|
| Asset: Truck | Capacity, Washout_State, GPS_Loc | TruckTrax |
| Asset: Plant | Status (Online/Offline), Silo_Inventory | Jonel API |
| Actor: Driver | HOS_Remaining, Union_Break_Status | Dispatch Logs |
| Material: Mix_Design | Cure_Time, Aggregate_Type, Chemical_Additives | Jonel Schedule |
| Event: Pour | Time_Since_Batch, Target_Volume, Job_Site | Jonel + TruckTrax |
Below is a 100-triple snapshot of the Knowledge Graph representing a standard fleet state at 10:00 AM across two plants, 20 trucks, and active job sites. This is the exact semantic structure the Generative AI queries before suggesting a reroute.
A resilient AI must acknowledge its epistemic boundaries. YARDMASTER synthesizes APIs, but it cannot see physical reality. If these gaps are not accounted for, the AI will recommend catastrophically flawed reroutes.
The Assumption: The Knowledge Graph sees a 10-yard truck.
The Reality: Due to hardened concrete buildup (fins/drum), the truck may physically only be able to safely haul 8.5 yards without spilling. If the AI max-loads a degraded truck for a critical reroute, it creates a DOT violation and a mess in the yard.
Mitigation: V1 assumes standard capacity, but requires the Batchman to visually verify the hopper before dropping the load. V2 will require manual fleet maintenance inputs overriding the 10-yard default.
The Assumption: GPS API shows the truck is "On Site" and pouring.
The Reality: The truck is physically at the geo-fence, but the job site entrance is blocked by a stuck pump truck or a mud pit. The 90-minute concrete clock is ticking, but the truck isn't actually pouring.
Mitigation: YARDMASTER tracks "Time On Site Without Pour Rate." If a truck breaches 20 minutes idle inside a geo-fence, it throws a localized anomaly alert to the dispatcher, preventing the AI from dispatching more trucks into a bottleneck.
The Assumption: The mix design in the drum exactly matches the Jonel API record.
The Reality: A site super rejects the slump (too stiff), and the driver manually adds 10 gallons of water from the truck's saddle tank, fundamentally altering the cure time and structural integrity of the mix.
Mitigation: YARDMASTER cannot physically track saddle tank usage. Therefore, any load that sits on site > 45 minutes defaults to [MIX_STATUS: UNVERIFIED] in the graph, forbidding the AI from calculating its structural validity if it is rejected and brought back to the yard.