Every headline figure in this application carries a documented basis. The table below is generated directly from the same claims registry the interface renders from, so it cannot drift from what is shown on screen. Each entry states the value, the scope it applies to, whether it is measured, modeled, simulated, or a design target, and how it was produced.
| Label | Value | Scope | Basis | How produced |
|---|---|---|---|---|
| Deterministic resolution rate | Not measured | no measured rate on real cases | Simulated data | No deterministic-resolution rate has been measured on real cases; the 12-month trend tile shows simulated data only. |
| Deterministic resolution rate | 70% | architecture floor | Design target | Minimum deterministic-gate coverage the control fabric is designed to hold; not a measured result. |
| Deterministic resolution rate | 85% | ambition target | Design target | Stretch objective for deterministic coverage after pilot-phase rule tuning; not a measured result. |
| Tier 3 LLM cost per case | $0.0040 | modeled, per published unit pricing | Modeled estimate | Computed from published per-token vendor pricing against representative Tier 3 prompt and completion sizes. |
| Broad-LLM baseline cost per case | $0.0450 | modeled industry baseline | Modeled estimate | Reference cost of mediating every exception through a frontier-class LLM, modeled from published unit pricing. |
| Modeled LLM-cost reduction (not measured) | 91% | modeled LLM-cost reduction (not measured); baseline: every exception sent to a broad LLM at $0.045 per case vs Tier 3 at $0.0040 | Modeled estimate | Difference between the modeled tiered-routing cost and the modeled broad-LLM baseline; not a billed figure. |
| False-positive rate | 3.8% | synthetic dataset · against generated ground truth | Simulated data | Measured against labels produced by the synthetic case generator. Indicates pipeline behavior only; not a prediction of real-world false-positive performance. |
| Golden-set result | 15 of 16 (94%) | latest stored golden-set run, n=16 | Golden-set measured | 15 of the 16 labeled golden-set cases matched their expected disposition in the latest stored run; a 200–300 case pilot baseline has not yet been established. |