Scope Control

The Invisible Killer

Designs rarely fail from hard problems; they fail from unbounded ones. Every system connects to everything — auth, notifications, analytics, fraud, chat — and each connection is a rabbit hole. Scope control is deciding in advance where exploration stops.

 THE BREADTH TRAP

 prompt: "design ride matching"
   └─► "well, riders need accounts..."          ← auth rabbit hole
         └─► "...with password reset flows"      ← deeper
               └─► "...and OAuth providers"       ← 20 minutes gone
                     └─► matching never drawn

 THE SCOPED PATH
 "auth exists; I'll treat it as an identity service returning a
  trusted user ID — focusing today on matching."

The In/Out Ledger

Write the boundary down. Unwritten scope re-enters by accident:

In scope (v1)	Out of scope (v1)
Request → match → trip → payment spine	Carpooling, scheduled rides
Rider + driver apps as clients	Driver incentive engine
Card-on-file payment via PSP	Wallets, split fares, regional methods
Basic ratings	In-app chat (SMS fallback assumed)

Two properties of a good ledger: every Out has a one-line reason, and the ledger is visible during the whole design so drift gets caught live.

Deferral Seams: Cheap Insurance for Later

Out-of-scope does not mean unanticipated. A seam costs minutes now:

 PaymentSpine v1:            the seam:
 client → TripService        interface PaymentProvider {
 → PaymentProvider             charge(tripId, amount): Receipt
 (single impl: StripeAdapter)  refund(receiptId)
                             }
                             later wallets/regional PSPs plug in
                             behind the same interface — no redesign

The test of a proper deferral: when the deferred feature arrives, it adds a box, not a rewrite.

Time-Boxed Depth Allocation

Scope control extends inside each kept component — decide depth before diving:

 MATCHING SERVICE (in scope, core):
   deep-dive: geo-index + match algorithm     ← the interesting part
   
 PAYMENTS (in scope, but commodity):
   shallow: call PSP, idempotency keys, done  ← don't reinvent Stripe

 HISTORY (should-tier):
   one line: replicas + cache                 ← move on

Depth follows risk and differentiation, not familiarity.

When Scope Creep Arrives Anyway

Mid-design additions (“can we also support…”) get the three-way response:

1. Park: add to ledger’s Out column with reason.
2. Trade: allowed only if something of equal weight leaves.
3. Seam-check: if trivial to note an attachment point, do so and continue.

Interview Framing

Interviewers inject scope tests deliberately: vague prompts, mid-design feature drops. The scoring behavior is verbal fencing — “that’s out of scope v1; here’s the seam it would attach to” — then immediate return to the core flow. Candidates who chase every branch run out of clock with no architecture; the discipline of returning IS the seniority signal.