6.2 Shared Remote Cache

Reason about cross-client cache identity across declared action dimensions, local state, HTTP and REAPI caches, hidden inputs, and bounded path mapping.

Deploy a maintained shared cache and prove remote reuse across two isolated Bazel clients.

Separate cache identities, permissions, namespaces, and writer trust so one producer cannot silently poison every consumer.

Attribute missing or wrong shared-cache results, bound the affected cohort, and recover without destroying evidence.

Decide whether one class of cache reuse saves latency and cost after lookup, transfer, and materialization are counted.

Design AC/CAS retention and cold-cache recovery so metadata never promises unavailable outputs and repopulation stays within fleet capacity.

Diagnose high-cardinality directory outputs by structure and per-file work, not aggregate bytes alone.