SHADOW MODE DIAGNOSTIC
Deploy Lakhowal in Shadow Mode — non-blocking, zero production risk. Watch permits in real time. Measure enforcement overhead. Build confidence before activation.
THE DEPLOYMENT MODE
Lakhowal operates in parallel to your AI execution stack. Every action proposed by your AI is classified and evaluated by the Gamma Stack — but the result does not block or gate the action. Your AI fires as normal.
Every evaluation, every predicate result, every Γ computation is logged and made queryable. You can see what permits would have been issued in real time.
Your AI models, your application logic, your infrastructure — nothing changes. Lakhowal intercepts and observes at the network or SDK boundary.
Shadow Mode is designed for quick deployment. API integration, predicate configuration, and live monitoring can be operational in one to two weeks.
SHADOW MODE INSIGHTS
What is the latency cost of the Gamma Stack? p50, p95, p99, p99.9. Most customers see ≤10ms additional latency in real workloads.
What percentage of actions would be permitted vs. denied? Which action classes trip which predicates? Where are the governance boundaries breaking down?
Which predicates are most selective? Which are redundant? Optimize predicate configuration before moving to enforcement mode.
Did a security incident correlate with denied permits? Did a compliance violation relate to a predicate that should have been stricter? Find the signal.
Is your AI behavior changing over time? Are new action types emerging? Is Integrity-Flux (I_φ) trending upward, indicating reasoning instability?
DEPLOYMENT PROCESS
We understand your AI stack, action types, and governance requirements.
We design the predicate set for your action classes. You review and validate.
We provide the SDK or API endpoint. You add two lines of code (or zero, if using HTTP interception).
Configure action class mappings, threshold rules, and alert rules.
Shadow Mode goes live. Permits are computed and logged in parallel. Zero impact to your production traffic.
Run diagnostics for 1–4 weeks. Measure overhead. Identify predicate tuning opportunities.
Decide whether to move to Inline Gateway (fail-closed enforcement) or remain in Shadow Mode.