One feature, end to end: schema, API, client state and rendering. Practise the loop that tests both halves of the stack in one round.
Fullstack loops rarely run separate frontend and backend tracks. Instead they hand you one feature - commenting with threads, a notification centre, a collaborative document - and ask you to take it from schema to screen in a single conversation. The test is whether you hold the whole contract in your head at once.
That means the failure mode is lopsided depth. Candidates who came up through backend design a beautiful data model and then wave at "the frontend calls this endpoint", ignoring pagination behaviour, optimistic updates and stale-response handling. Candidates who came up through frontend design a slick interaction and then propose an API that cannot be served efficiently.
Startups weight this loop heavily because a fullstack hire is expected to ship a feature without a handoff. Expect questions about what you would cut to ship in a week, which is a scope judgement test as much as a technical one.
60 min
Schema, API contract, client state, rendering and the failure cases between them.
90 min
Build a small working feature across client and server, with real edge cases.
45 min
Usually one medium problem, weighted lighter than in specialised loops.
45 min
Scoping, what you would ship first, and how you decide when to cut.
Less deep on each side, but they penalise gaps harder. You will not be asked to implement a promise polyfill or design a multi-region database, but being unable to explain how your API paginates or how your client handles a stale response is a clear miss.
Your weaker half, and specifically the seam. Most fullstack rejections happen at the contract between client and server - pagination, caching, optimistic updates, error propagation - rather than deep inside either side.
Less so. Big tech usually hires into a specialised track and tests it. Fullstack loops are most common at startups and mid-size product companies, where shipping a feature without a handoff is the actual job.
Take a feature from schema to screen under interview time pressure.