Google

What they grade on

• Depth on distributed systems primitives (consensus, replication, sharding)
• Precision in reasoning about consistency and latency
• Ability to go deep on storage engines when probed
• Handling of ambiguity — Googlers ask the vague question deliberately
• Willingness to challenge your own assumptions when new info arrives

Deal breakers

• Confusing eventual and strong consistency
• Refusing to go deeper when asked 'why?'
• Ignoring the interviewer's constraint

Super helpful signals

• Citing Spanner, Bigtable, or Colossus with the specific property that applies
• Discussing 2PC + Paxos when relevant, and their trade-offs
• Reasoning about SSTables, LSM trees, or Bloom filters when discussing writes

Delivery advice

• Be precise. 'Eventually consistent within 100ms' beats 'eventually consistent'.
• When the interviewer says 'go deeper', they mean it — pick the most interesting sub-problem and go 3 levels down.
• If you cite a Google system, cite the specific property (Spanner's TrueTime, Bigtable's tablet servers, etc.).
• Volunteer alternatives you rejected and why — Google grades the discarded options too.
• Reserve time for the deep-dive; Google interviewers often hold back their favorite question until minute 25.

Red flags to avoid

• Vague on consistency semantics
• Skipping the storage layer
• Not asking clarifying questions