Requirements checklist

Run the conversation from the race — everything else follows:

□ the race                two users, one seat: first tap wins, at tap time
□ holds + TTL             picking = exclusive claim, 5 min; expiry frees seats
□ payment position        external call AFTER the hold; success/fail/timeout
□ failure semantics       fail -> release hold; timeout -> TTL handles it
□ retry-safe confirm      duplicate success callbacks: same booking, no error
□ seat map anchor         per SHOW, not per screen (layout vs availability)
□ scope                   pricing/discounts/food out; single process, in-memory

The core model

Show           owns Map seat -> AVAILABLE | HELD | BOOKED  (the ONE truth)
               screen keeps only the immutable layout
Hold           id + seats + owner + expiresAt; all-or-nothing unit
               lazy expiry: checked at the top of every mutating call
placeHold      one synchronized check-then-claim — the race is decided
               here and nowhere else; ANY seat taken -> claim nothing
confirm(holdId)  idempotent (booking keyed by hold id); expired -> clean
                 fail; only live unconfirmed holds book
Lifecycle      AVAILABLE -> HELD -> BOOKED; expiry/release -> AVAILABLE
Timeout        NO code path — the TTL already bounds every hold

Principles, at their decisions

• Single source of truth: seat state has one owner (the show) and one writer path; the race has exactly one battleground.
• Cohesion: the hold object carries claim, owner, and deadline together — release, expire, confirm are one operation each.
• Idempotency: confirm keyed by hold id — same input, same booking, no second effect.
• Separation of concerns: immutable layout (and later, pricing) apart from hot availability.

Complexity facts

placeHold      O(k) for k wanted seats + O(holds) lazy-expiry sweep
confirm/release  O(k); all under the show's lock — contention is per show
memory         O(seats + live holds) per show; expired holds reclaimed
               lazily (correctness) + optional sweep (hygiene)

What earns points, per report dimension

• Requirements & interface — you opened with the F7 race, pinned the TTL, asked about duplicate callbacks, and caught the per-show seat-map anchor.
• Core design & invariants — one writer for seat state, all-or-nothing holds, booking only from a confirmed live hold, all stated unprompted.
• Extension probe — the round will move the requirements; land the change on a seam (layout side, a new guarded transition, the one claim path) and say what survives.
• Complexity honesty — lock scope named (per show), lazy expiry's cost on the mutating path acknowledged.
• Communication — the timeout row ("no code path — the TTL handles it") and the stale-search caveat said out loud.

Ready? Sit the live mock → — the interviewer will run a twist this chapter deliberately hasn't shown you.