Degraded
The Embedder failed and EmbedFailurePolicy.FALL_BACK_TO_COMPUTE stepped aside, so the call ran uncached.
This is the one failure mode that otherwise leaves no trace. A rejected candidate produces a MissReason and moves a counter; a fall-back produces neither, because it is not a miss — no lookup ever happened. A team running with EmbedFailurePolicy.FALL_BACK_TO_COMPUTE and a flapping embedder would watch its hit rate collapse with nothing naming the cause, and would reasonably suspect the part that is instrumented, the guards. Counted in CacheStats.degradedLookups.
A RetryingEmbedder that exhausts its attempts arrives here too: retrying is transparent to the cache, so the final throwable is what cause carries. There is no separate "retries exhausted" event, because the cache cannot see that a retry happened at all.
Constructors
Properties
Which entry point stepped aside.
The policy that decided to step aside, always EmbedFailurePolicy.FALL_BACK_TO_COMPUTE today.
The prompts the failed call covered — one element for the single-prompt entry points, the whole batch for SemanticCache.getOrPutAll. Never empty.