SCENARIO ACTIVE2 Active Scenarios โ€” $355K total exposure
$355,200 SLA exposure
CRITICAL

Context Coverage Impact

How each layer of context improves decision quality

โ„น

Illustrative data. Metrics on this page are derived from this demo deployment's seeded Decision Object corpus. They are intended to illustrate the type and directionof improvements AgentForge produces โ€” not to represent any specific production deployment. In a live deployment, all figures are computed in real time from that customer's actual decision history.

Quality: ERP Only
?
54/100
Quality: Full Forge
?
89/100
Quality Lift
?
+35pts
Context Sources
?
4 layers
Decision Quality by Context Layer?
Structured Data Only
54/100
+ Unstructured Docs
68/100
+14
+ Supplier Signals
74/100
+6
+ Pattern Memory
81/100
+7
+ Institutional Memory
89/100
+8

Pattern Memory (+7pts) and Institutional Memory (+8pts) are the two biggest differentiators โ€” neither is available in a conventional planning system.

Outcome Distribution by Context?
Good
Acceptable
Bad
Case Study: DO-2023-SC-0031 โ€” The $31K Lesson
Without Pattern Memory (what happened)
  • โ€ข Agent had no record of SUP-A Q4 capacity constraint (pattern P-002 didn't exist yet)
  • โ€ข Full network expedite ordered โ€” 1,000u from SUP-A at 35% premium
  • โ€ข SUP-A hit capacity wall, delayed 6 days post window
  • โ€ข RetailerX SLA breach + margin destruction: -$31,200
With Pattern Memory (what Forge does now)
  • โ€ข P-002 flags SUP-A Q4 capacity risk before hypothesis is even generated
  • โ€ข P-006 eliminates full-expedite approach at adversarial validation
  • โ€ข P-003 routes to MW-02 โ†’ NE-01 Werner transfer instead
  • โ€ข Result: SLA met, net outcome: +$8,640
This single bad decision generated patterns P-002, P-003, and P-006. The $31K loss has now been repaid ~3ร— in prevented failures.