LOADING
Core explanatory content is server-rendered. This state is only for a route segment that is still resolving.
OPTIMIZATION
Use model-target optimization such as DSPy or GEPA behind canonical behavior identity, independent evaluation, staleness rules, explicit promotion, and canonical fallback.
DEFINITION
An optimized projection is a model-target execution artifact, not a new policy source. Canonical behavior remains the governed authority that defines what the system is intended and allowed to do.
HOW IT WORKS
Compose the effective behavior program from exact active scope authority.
Produce the canonical model-facing projection while hard guards remain runtime-enforced.
Give DSPy, GEPA, or another optimizer a bounded objective for one exact model target.
Bind the optimized artifact to the exact profile, revision set, composition digest, artifact digest, and target identity.
Evaluate the optimized artifact independently against the canonical behavior it is intended to execute.
Promote only through release authority; invalidate on source or target drift and fall back to canonical execution when optimization is no longer eligible.
PRODUCT PROOF
These fields describe the product evidence Vira preserves for this part of the lifecycle.
CONTROL BOUNDARIES
The optimizer receives a governed behavior identity; it does not author new hard policy or broaden scope.
An optimized artifact is eligible only for the model target and composition identity it was evaluated against.
Lesson, binding, ancestor authority, registry, privacy, deletion, or target changes can invalidate optimized execution.
When optimized authority cannot be proven, valid canonical execution remains the controlled fallback rather than silently reusing stale optimization.
LIMITATIONS & FAILURE STATES
DEVELOPER HOOK
Model-specific execution can change while the same canonical behavior identity remains available for evaluation, fallback, and explanation.