pub struct VerifiedCacheConfig {
pub ttl_secs: u64,
pub max_entries: usize,
pub redis_url: Option<String>,
}Expand description
Verified-response cache (§8.4). See Escalation::verified_cache.
Fields§
§ttl_secs: u64How long a proven answer stays replayable.
A proof has a shelf life: the model, the gate, and the world all move. Short by default because replaying a stale answer confidently is worse than paying to re-derive it.
max_entries: usizeHard cap on entries; oldest evicted first. In-process store only — a Redis-backed cache is bounded by its TTL and the server’s own eviction policy.
redis_url: Option<String>Redis URL for a shared cache across replicas. Absent (default) = in-process only.
Without it the cache is per-replica: behind N instances the same answer is cached N times over, the hit rate drops roughly by N, and — the part that matters — a retraction only reaches the replica that received the feedback, leaving the other N-1 serving an answer that has been disproven.
Requires the redis-cache build feature. Configuring a URL without it is rejected at
startup rather than silently ignored: a cache that quietly stays per-replica looks
identical to one that is working.