Skip to content
Tech Interview Prep

Top 100 Solutions Architect Practice MCQs

Every question on one page: open one, choose an answer, and see why at once. Signed in, your latest answer to each question counts, and you can answer again after 24 hours.

Curated: · Written: · Reviewed:

Reviewed 100Review pending 0

Answers to reviewed questions count toward this role and the overall leaderboard; your latest answer to each one is the one that counts.

100 questions

MCQ-63How is database indexing best explained to a stakeholder deciding whether to fund performance work?
MCQ-61A three-tier web application must scale horizontally and tolerate the loss of any single instance. Which single change contributes most to that goal?
MCQ-1A new microservice is being designed to handle user traffic. The product owner has only specified the high-level requirement: 'The system must be scalable.' The engineering team needs to define the capacity and scaling strategy before beginning implementation. Which of the following actions represents the most critical next step for the architect?
MCQ-91Which is the best interpretation of a customer describing a requirement as 'real-time' for a management dashboard?
MCQ-6When presenting an architecture to a CTO, a finance director, and two engineers, which structure works best?
MCQ-79A customer with no monitoring asks what to instrument first. Which set gives the most value?
MCQ-31A customer reports that their web app is slow. The team's own tests from the office load it in under a second, the load balancer shows server response times of 90–120 ms, and the database's slow-query log is empty. There is no client-side instrumentation. What should the team do first?
MCQ-62A nightly batch job no longer completes before the business day. What should be established first?
MCQ-47After an incident where the database failed over in 4 minutes but checkout stayed down for 50 minutes, the team must produce evidence about whether its 30-minute recovery objective is met. Which exercise produces that evidence?
MCQ-2Which scope best fits a proof of concept for a project whose main risk is that a legacy mainframe cannot sustain the required query rate?
MCQ-12A read-only reporting job runs on the primary each weekday morning for about 90 seconds. While it runs, p99 API latency rises from 40 ms to 180 ms; the primary's CPU stays near 30%, but its buffer cache hit rate falls from 99% to 71% and disk read IOPS rise about sixfold. Which reason best explains why moving the job to a read replica brings API latency back down?
MCQ-52A customer's analysts run ad hoc SQL against the production primary. Their queries are indexed lookups that finish in under 5 ms and together use under 1% of the primary's CPU. The customer's engineers ship schema migrations every week. Which is the strongest reason to move the analysts' queries onto a separate reporting database?
MCQ-48A payments team's 6 TB Postgres cluster holds seven years of transactions in the same tables the application writes to. Its recovery time objective is four hours, and last quarter's restore drill took nine hours. Which reason best explains why archiving the older years fixes this?
MCQ-92A six-person team runs Postgres on a single EC2 instance with nightly pg_dump to S3. Queries meet their latency targets, Datadog already alerts on the database, and the data is internal with no PII. The engineer who built the setup has left, the last failover was manual and took 40 minutes, and no restore has ever been tested. Which reason should the recommendation to move to managed Postgres rest on?
MCQ-56A six-engineer team is deciding whether to extract the refunds logic from its monolith into a service. They expect the refunds/ledger seam to move at least twice in the next year as partial refunds and chargebacks land. Which consideration should decide the split?
MCQ-71Which is the most reliable indicator that a customer's service boundaries are drawn incorrectly?
MCQ-27A customer wants microservices. Which finding would most support that recommendation?
MCQ-85A team runs one internal service that a single known consumer calls directly over mTLS inside the same VPC. The service already validates caller tokens and enforces a per-caller rate limit. A platform engineer proposes putting the company's API gateway in front of it. What should the team do?
MCQ-15A customer requests a feature that would let support staff view customer passwords. What is the best response?
MCQ-18A six-engineer team runs one product on a single cloud. Its only provider-specific dependencies are a managed queue and an object store, both already behind thin wrappers; moving them to another provider is estimated at three engineer-weeks, and there is no committed-spend contract or meaningful egress bill. Which argument against building a provider-abstraction layer over those two services is strongest?
MCQ-19A customer wants to expose an existing internal API to a mobile application. Which change is most critical?
MCQ-17A customer's requirement is 'real-time reporting'. You get one question before scoping the build. Which one most efficiently determines the architecture?
MCQ-36A search service's acceptance criterion reads: 'p95 latency under 200 ms, measured server-side at the gateway.' Staging (same instance class as production, 20k documents, one request at a time) passes at 12 ms with no errors. The production pilot, 40M documents at 300 requests per second, measures 1.4 s p95, also with no errors. Which addition to the criterion would have made the staging result predictive of production?
MCQ-42A team is adding a read-through Redis cache in front of a Postgres primary that serves 12,000 reads/s at peak. Redis is already chosen, sized at 40 GB for the projected working set, and placed in the same region as the primary; a two-week shadow run at production traffic measured a steady 94% hit rate. Which unanswered question should block the design?
MCQ-46A nightly job pushes 90M records to a vendor's bulk-ingest API in 500-record batches and must finish inside a 4-hour window. The vendor's contract states 500 req/s. A four-hour load test against production holds 40 req/s for the first minute, then settles at 8 req/s for the rest of the test. Which estimate should the job be sized against?
MCQ-22A team is building an order service. The company has a VendorX event-bus contract with a $15,000/month minimum spend through 2027-06-30 and no exclusivity clause; the remaining minimum spend is $360,000. VendorX buses handle 5,000 msg/s per instance, deliver at-least-once, and cost $8,000/month for each additional instance. The order service peaks at 12,000 msg/s and needs exactly-once processing. Today the domain code calls the VendorX SDK directly. Replacing VendorX now would cost the $360,000 remaining minimum spend plus $120,000 migration. Putting the peak on self-managed Kafka behind a domain-owned EventPublisher interface costs $45,000/year and 3 engineer-weeks. What should the team do?
MCQ-67Checkout calls fulfilment synchronously. A 25-minute fulfilment deploy last month returned 502s to checkout and 4,300 orders were abandoned. The two teams already release independently, peak order rate sits well inside fulfilment's capacity, and order events are already written to an audit log. Which justification for putting a queue between them is the one that holds?
MCQ-24A payments team must feed premium and claim entries into a policy administration system that re-prices the whole book in a close-of-day run. The vendor takes writes only as one complete, balanced file per business day and rejects partial or per-entry submissions. Month-end peaks at roughly 400,000 entries, and the one engineer who runs the feed also carries the policy system's pager. Which integration should the team build?
MCQ-5Your team must send order events to a partner that exposes one HTTPS ingest endpoint and no broker or database access, and that goes down for up to four hours at a time. Events have to reach them within 15 minutes of being placed. Their endpoint returns 202 to everything and buffers internally; their ops team sizes that buffer at about 1,200 events per minute, and during their last recovery a 40,000-event catch-up burst over ten minutes made them shed production traffic. Which approach should you build?
MCQ-28A platform team of five must choose a message broker for a workload that requires exactly-once delivery with p99 under 10 ms; neither Kafka nor RabbitMQ meets that requirement well. Two candidates are on the table: Kafka and RabbitMQ. In the first meeting, the Kafka advocates propose four criteria (throughput, partition scalability, replay, ecosystem) and the RabbitMQ advocates propose four (low latency, routing flexibility, ease of ops, existing team skills). The criteria are weighted by a show of hands; each side votes for its own. The result is a 4–4 tie, broken by the director choosing Kafka. The RabbitMQ advocates say the process was rigged. In the second meeting, the lead proposes: 'Before we name any broker, let's agree on the criteria and weights that matter for our workload.' Which outcome is most likely?
MCQ-58A payments company is starting a 12-month, four-phase migration from a Java monolith to event-driven services. Phase 1 has 10 weeks and one two-pizza team. The CTO will cancel the programme unless phase 1 produces evidence that the new stack preserves per-account ordering at the 3,000 TPS peak. Which phase-1 outcome best meets that constraint?
MCQ-60A delivery team replaced a reference architecture's shared message broker with 40 per-tenant queues to satisfy a customer's infrastructure-isolation requirement. The reference broker used tenant ID in message metadata and one consumer group per tenant to guarantee per-tenant ordering. After go-live, tenant A's messages arrive out of order. The team says the isolation requirement forced the change. Which check best determines whether the divergence was a thoughtful application of the reference architecture?
MCQ-100Your payments team runs a 3-node PostgreSQL 14 cluster that handles 4k writes/s at p99 12 ms, with a Python consumer. A new fraud-scoring requirement needs 25k events/s, exactly-once stateful windowing, and p99 under 100 ms. You benchmarked extending the existing path: adding three nodes and batching reached 18k events/s, p99 240 ms, and duplicate deliveries during failover. Apache Flink 1.18 meets the target in a proof of concept, but no engineer on the team has run it in production. Training budget is approved. What should the team do?
MCQ-16Six teams deploy their own services twice a day: no shared release train, and no team can block another's merge. Three incidents last quarter came across one boundary. The producer's nightly export changed a field from minutes to seconds but kept the same integer type, so both builds stayed green and the consumer ran the wrong schedule for nine hours before anyone noticed. Which change to how that boundary is managed would have caught the units change on the producer's side?
MCQ-41A freight operator's brief asks for a 'real-time streaming platform' to replace a nightly batch job that writes truck positions into a database. The brief states 800 trucks, one telemetry ping every 30 seconds per truck, and a goal of 10,000 trucks within two years. The team's draft proposal leads with Kafka, Flink and a multi-region deployment. Which question should you ask before that proposal goes out?
MCQ-87A payments API team measures time from signed contract to a partner's first successful production call: median 24 days. Sandbox and production keys are issued by a human in a weekly batch, median 9 days. The sandbox returns a fixed 200 body for every endpoint, and partners report writing their 429 and decline handling against production. Which investment removes the most of that 24 days?
MCQ-25A retailer's flash sale runs for six hours on a date fixed months ahead. Traffic ramps from baseline to 40x in under 90 seconds, and the autoscaler needs about 8 minutes to add instances. Last year's forecast was 25x and the actual peak was 60x; finance has capped spend at what the 40x forecast costs. Which measure matters most?
MCQ-75A service has run a 3-month pilot with 200 internal users. At that size p95 latency is 180 ms and the error rate is 0.1%. A load test at 5,000 users shows p95 2.4 s, a 4% error rate, and the primary database connection pool saturating at 100 connections. The security review has not started and support has no runbook. Stakeholders want to open the service to 50,000 users next week. Which action is most defensible?
MCQ-64Your team inherits a payments service from a previous supplier and must extend it to a new peak of 800 rps. A load test shows p99 latency of 45 ms up to 600 rps, then a climb to 1.2 s at 800 rps. A flame graph attributes 70% of the time at 800 rps to a global lock held while a config map is rebuilt every 30 s. The code also uses a hand-rolled DI container and a module layout that does not match your reference architecture. Which action should the review recommend?
MCQ-65A billing service has run unchanged for four years. Its author left, the only description of the fee rules is the code itself, deploys are hand-run from a runbook, and the dashboards count requests but say nothing about fee amounts. Finance needs a change to the late-fee calculation live in two weeks, and a wrong fee that reaches customers means a manual credit run. What should the team do in those two weeks?
MCQ-89A customer's RFP requires a system that supports "unlimited" growth. Their peak is 400 write TPS today and has grown 10% month over month for two years. The proposed design is a single Postgres primary with a 200 ms p99 write SLO. Which question should the architect settle before freezing the design?
MCQ-54Phase 2 is quoted 40% above phase 1, and the account is up for renewal. You know the real reason: phase 1 was underpriced. The estimate missed the integration work. The customer's engineering lead wrote that integration and will read your explanation. Which explanation is most damaging to give?
MCQ-53A 40-service SaaS runs 5,000 req/s at peak. Its observability bill is $31k/mo: 100% trace sampling $18k, 300 GB/day logs $9k, and 2M active metric series $4k. The budget is $6k/mo. During incidents the team must answer four questions: (1) is checkout failing before customers report it, (2) which deploy changed the error rate, (3) which tenant is affected, and (4) what is p99 latency per endpoint. The 2M metric series are aggregated and carry no tenant_id or deploy_id. The 100% traces are not indexed by tenant_id or deploy_id. The vendor bundle's logs are unstructured and its traces are not indexed by tenant_id or deploy_id. Synthetic checks cost $200/mo; deploy annotations are free; structured logs carrying tenant_id and deploy_id at 30 GB/day cost $900/mo; tail-sampled traces (all errors plus 1% baseline) cost $1.2k/mo; cutting metric cardinality to 200k series costs $400/mo. Which plan should they buy?
MCQ-10A lift-and-shift business case projects a 25% monthly saving from moving 200 virtual machines to the same provider's IaaS. Which assessment finding most directly invalidates that projection?
MCQ-3A payments team is deciding whether to build an in-house rate limiter or buy a vendor service. Building is estimated at 12 engineer-weeks, and a loaded engineer-week costs $5,000. The vendor charges $3,000 per month. The team's postmortems for a similar service they already run show it consumes about 4 engineer-hours a week in patching, on-call and fixes, at a loaded $125 per hour. Over a three-year horizon, which option is cheaper, and by how much?
MCQ-69A team lift-and-shifted 40 instances, each 8 vCPU / 32 GiB, onto on-demand pricing at $0.40 per instance-hour, running 24/7. The monthly compute bill is $11,680, against $7,000 a month for the hardware it replaced. Measured p99 CPU is 24% of the 8 vCPU, so the team plans to right-size all 40 instances to 4 vCPU / 16 GiB at $0.20 per instance-hour. 18 of the 40 are non-production and are needed only 10 hours a day, 5 days a week, 217 hours a month, so those 18 will be stopped outside that window. What is the new monthly compute bill?
MCQ-49A retail platform team wants to run self-managed Kafka on EC2. The managed offering is $12,000/month and meets the same p99 target. The ops group of six SREs has no Kafka experience and already runs a 1-in-3 on-call rotation; their lead says the objection is staffing, not the technology. Self-managed infrastructure is $4,800/month. Making the ops team Kafka-operable needs two additional SREs at $180,000 fully loaded each per year. The developers' director and the ops director report to the same VP, who owns both budgets and has asked for the annual premium of self-managing over the managed offering. What figure should the staff engineer give?
MCQ-50After a six-month build, a customer asks for a managed service. They want a 15-minute P1 response 24/7 and all change requests included in a fixed monthly fee. Your data from last month: 3 P1 incidents, all between 10am and 3pm on weekdays, averaging 2 hours each. 12 change requests, averaging 6 hours each. Your engineer cost is $150/hour. To offer 24/7 15-minute P1 response, you must staff an on-call rotation: two engineers per night, five nights per week, at $400 per engineer per night; and two engineers per weekend day, two days per week, at $400 per engineer per day. Assume four weeks per month. What is the minimum monthly price you should quote for the requested scope?
MCQ-80A payments client must be live on the card network in 10 weeks, with $400k approved. Its engineering spec asks for 99.99% availability and a 15-minute RPO; the signed network contract requires 99.9% availability and a 24-hour RPO, and fixes the reporting and audit scope. Your team's estimate for the written spec is $1.1M. A single-region, multi-AZ baseline with hourly log shipping meets the contract; the two-region active-active footprint that 99.99% needs adds $440k, and the synchronous cross-region commit path that the 15-minute RPO needs adds $320k. Which response should you send?
MCQ-4A retail customer's RFP asks for 99.99% availability on a new checkout service. The target is not a regulatory requirement, and the business sponsor who owns the budget has final say on scope. Their engineering lead says the current system runs at 99.9% and has had no revenue-impacting outage in 18 months. Which response is most useful before the number goes into the proposal?
MCQ-33A payments team can modernise four of its five legacy components this quarter. The mandate is to cut the risk of unplanned outages, unblock the capability the business has already committed to, and reduce the cost of future change. Which component is the most defensible to leave unmodernised?
MCQ-99A previous supplier built a payments platform that has run in production for four years; it has met its 99.9% availability SLA every quarter, the customer's operations team report no recurring incidents, and the customer has no plans to move off the current stack. The customer's new CTO, three weeks into the job, hands you a requirements document written last month and asks you to review the design, which she says 'gets the fundamentals wrong': it is single-region, reconciles in a nightly batch, and has no autoscaling. What is the most appropriate first step?
MCQ-11A customer's security team rejects the design two weeks before delivery. What is the most effective first step?
MCQ-93A customer's six-engineer team proposes building a bespoke payments-reconciliation engine, arguing that the two commercial products they trialled each lack one report the finance team wants. The customer's scheme rules are the standard ones the vendors already support, and reconciliation is not part of the customer's product. Which recommendation is most defensible?
MCQ-73A payments team must ship a read path in 4 months. Their preferred design is self-managed PostgreSQL 16 with logical replication to a read replica, measured at 40 ms p99 replication lag. The customer's CIO has 26 months left on a managed-cloud commitment and has declined a waiver; the mandate is that all data at rest be stored in that vendor's managed PostgreSQL. The vendor's managed PostgreSQL read replica has 900 ms p99 replication lag and sustains 2,000 QPS; the primary serves reads at 8 ms p99 and has 10,000 QPS of spare read capacity. The read path must return data no more than 500 ms stale and needs 6,000 QPS. What should the architect do?
MCQ-32A payments team runs PostgreSQL 16 with synchronous streaming replication to a standby in a second availability zone. Recovery is a nightly `pg_basebackup` retained for 7 days; WAL archiving is not configured. A migration deployed 11 days ago had an off-by-one in its `WHERE` clause and set `status = 'cancelled'` on 40,000 orders instead of the intended 400. The standby applied the same UPDATE within milliseconds. The error surfaced during month-end reconciliation. Which change would have made the 40,000 original `status` values recoverable?
MCQ-43A payments service writes to Postgres through its ORM, and every write already inserts its audit row in the same transaction. The same tables are also changed by a nightly reconciliation job, an internal admin console, and engineers running ad-hoc UPDATEs during incidents. The auditable columns are a fixed list, writes run at a few hundred rows a minute, and nothing does bulk updates. Compliance must be able to answer 'who changed this row, and when' for any row. Which is the strongest reason to capture the audit trail in the database rather than in the service's code?
MCQ-26An ADR from March 2022 states: 'All PII must remain in eu-west-1.' It records the decision and the date but no source. In January 2025 the team wants to add a read replica in us-east-1. The original author has left. The constraint was imposed by an external party under a time-limited agreement. Which addition to the ADR would have let the team decide whether the constraint still applies without re-investigation?
MCQ-77A team's ADR for a database choice records the decision, the date, the participants, and the benchmark results. Two years later an engineer proposes the option that lost, and the team spends a week re-running the benchmarks before rejecting it again for the same reason. Which addition to the ADR template would have avoided that week?
MCQ-76An order-pricing service keeps each promotion as an enum value plus fixed columns on the `orders` table (`discount_type`, `discount_value`, `min_spend`). Every new promotion has needed an `ALTER TYPE ... ADD VALUE`, a backfill over 40M order rows, a new branch in the pricing code, and a deploy; the last four took 9, 12, 14 and 11 days from request to production. The catalog and order schemas have otherwise not changed in 18 months, and the business expects about 40 promotions next year. Which change most reduces the lead time for the next promotion?
MCQ-35A retailer's finance team counts billing accounts, sales counts prospects plus accounts, and support counts contacts, and the same company can appear in all three counts. Finance's revenue-by-customer report must tie to the ledger, and sales and support must keep their current counts. Which design is most likely to give all three teams numbers they trust?
MCQ-14A vendor hands over a payments service to an in-house team of 3 engineers. The contract requires complete documentation, and the vendor delivers it: 40 runbooks, architecture diagrams, and vault credentials. The team will take over the ingest and API tiers first, then the data platform. The vendor's engineers leave in 10 working days. The team estimates it can rehearse about 6 runbooks with the vendor present in that time. The vendor's engineers recommend starting with the multi-region failover, which they call the most complex procedure, and note that the daily deploy and cache flush are the runbooks they execute most often. Last year's incident log for the service: 62% of Sev-1s were database restores, 21% were queue backlog, 9% were certificate expiry, 8% were other. Which 6 runbooks should the team rehearse?
MCQ-13A bank wants a firm fixed price to migrate a 14-year-old loan-origination system off an unsupported database. The schema and row counts are readable from the production catalogue, but the business rules (rate resets, fee accrual, delinquency transitions) exist only in the code and in the heads of two DBAs who retire in six months. The capital committee approves one fixed number and compares vendors on price. Which response serves the bank best?
MCQ-20Three weeks before a contractually fixed go-live, an audit of the customer's customer-master extract shows 22% of records failing validation. The failures are a mix: truncated and mis-encoded fields, and records where two addresses or two tax IDs disagree with no way to tell which is current. The customer's data steward, the only person who can settle the disagreements, has one two-hour slot next week. What should the delivery lead do?
MCQ-21A payments team gives a vendor's streaming API a two-week proof of concept. Before it starts, they write down the single question it is meant to settle: can the API hold 5,000 events/sec at p99 under 200 ms with our peak payloads, in the configuration we would ship? On day nine it measures 900 ms p99 at that rate. The vendor offers to tune the cluster, the sponsor wants the stakeholder demo to go ahead as planned, and four days remain in the time box. What should the team do?
MCQ-23You are preparing a fixed-price bid for a data platform migration. Three line items carry uncertainty: data migration is 30 person-days ±10% (you have run this exact migration twice); custom report build is 5 person-days per report, and the customer has specified between 12 and 30 reports, to be finalised in week 2; third-party API integration is 15 person-days, but becomes 45 if the vendor's sandbox rate limit is 10 req/s instead of 100 req/s, which a one-day spike can measure. Procurement will sign only against a single number. Which presentation should the bid use?
MCQ-9A fixed-price programme's first user-visible feature is four months out, gated by platform work that cannot be compressed. The customer's sponsor is non-technical, chairs a weekly steering call, and can invoke a break clause each quarter. Discovery is two weeks in. The sponsor's complaint about the previous vendor was that they were kept informed but saw nothing working until month five. What should the team do about the weekly call?
MCQ-8A customer's DR design workshop is in two weeks. The contract commits to a four-hour RTO, the storage platform was replaced eight months ago, and the last restore test predates that change. Last quarter's audit evidence is current and legal signed off the retention window last month. Which is the strongest reason to run a restore drill before the workshop?
MCQ-39A vendor's build starts 1 October 2025 and the plan shows go-live on 24 December 2025, twelve weeks later. The build is estimated at six weeks. Security review takes three weeks and cannot start until the build is code-complete; it typically raises findings that need a two-week remediation window before sign-off. The customer's change advisory board freezes all production changes from 15 December 2025 through 5 January 2026 inclusive. Access to the customer's test environment was provisioned before 1 October. What is the earliest realistic go-live date?
MCQ-90A team plans a 10-week release. The checkout rewrite is well understood and takes 6 weeks. The new payment provider integration needs a 3-week spike; if the spike fails, a 5-week redesign is required and the checkout rewrite must be repeated because it assumed the old provider. The team schedules the checkout rewrite first, then the spike. In week 9 the spike fails. If they had run the spike first and then done the checkout rewrite once, what is the earliest week the release could ship in the failure case?
MCQ-74A team estimates nine weeks of build for a customer's data-platform migration and accepts the customer's fourteen-week go-live date after planning only its own engineering work. Which risk does that estimate most likely understate?
MCQ-94You're the engagement lead on a fixed-date migration for a regulated customer. Your own team has already absorbed two departures and re-baselined twice; the revised plan still fits the date. At the week-three checkpoint, which signal most reliably predicts the date slipping?
MCQ-57A payments team's design reviews circulate the doc 48 hours ahead and list the four decisions with their costs. Last week's 90-minute review still closed none of them. Three of the four decisions were blocked on the partition key, which the schema owner has to sign off; two candidate options differ 3x in write amplification. She was not invited. The schema freeze is in two weeks. Which change is most likely to close the decisions?
MCQ-40A nine-month migration programme keeps a 47-entry customer risk register. Each entry carries a 1–5 likelihood and 1–5 impact score, and the steering committee reviews the top 12 by product in a 30-minute slot it has refused to extend. The twelfth-ranked entry scores 4×3=12 and the twentieth scores 3×3=9. Last quarter two risks materialised that never entered the top 12: the customer's SSO certificate expired and locked out 2,400 migrated users (logged as "SSO integration risk", scored 2×5=10), and a data-residency ruling paused a workstream (logged as "regulatory risk", scored 1×5=5). Neither missed entry had an owner or a next action. After the outage the customer's security lead said she would have flagged the certificate date if anyone had asked her about the SSO cutover. Which change to how the register is maintained would have caught both risks?
MCQ-59You are the tech lead on a fixed-price contract to build a card-payments integration. The statement of work lists 'a daily reconciliation report for the customer's operations team' as a deliverable and requires a signed change order before any work not listed in it begins. Two weeks before acceptance, the customer's ops lead asks for that report to be emailed to their team at 06:00 each day. It is estimated at 2 engineer-days, and the team has 3 engineer-days of unallocated float before the acceptance date. What should you do?
MCQ-70A three-engineer team is scoping the first release of a multi-tenant SaaS that replaces a customer's legacy system. Their rule: defer anything a real user can complete a real task without, unless deferring it costs more than building it now or its absence harms users. Today there are 12 tables, one shared cache namespace, and no customers; by month 12 they project 40 customers. Which deferred item does the rule pull into the first release?
MCQ-34Your team is shipping the first release of a B2B product in six weeks. Every customer will run on one shared deployment and one shared schema, and the first two signed customers are direct competitors. Which item has to be in that release even though a user could complete every task without it?
MCQ-7A company must keep all customer data within one country. The primary database and its read replicas already sit in that country. Which element should the audit check next to catch a residency breach that storage-region reviews miss?
MCQ-37You're the staff engineer at a SaaS vendor. A bank's internal audit found that 3 of the bank's 40 admin accounts on your platform had MFA disabled. The bank's auditor flagged your one-page security posture, which states 'MFA enforced on all administrative access', as the reason the bank's own gap went unnoticed. The page currently lists your controls and the frameworks you align with. What should the page carry?
MCQ-44You are scoping SSO for a 3,000-seat customer whose identity provider is Okta. Their security policy requires that a terminated employee lose access to your product within one hour, and that each user's permissions track their department as recorded in the IdP. Which piece of the integration does that policy make mandatory?
MCQ-45You are the supplier's engineer on a contract where the customer is the data controller. Their engineering manager asks you to add a nightly export of their users' purchase history to a third-party ad platform. The processing is lawful under your contract, but the customer's published privacy notice says purchase history is never shared with third parties, and the engineering manager does not own that notice. What is the most appropriate response?
MCQ-55A regulated customer is evaluating a SaaS product before signing. Its own compliance obligations depend on what data the product handles, who else can reach it, and from where. Which evidence matters most when assessing whether the product meets those obligations?
MCQ-68A customer's security team insists that all personal data stay in Germany. Their platform team says any EU region is fine and prefers Ireland on cost. The provider runs regions in both countries. Which fact decides whether Ireland can be in the design?
MCQ-81A customer requires that support engineers based in another country must never access production data, even during an incident. The data itself already stays in the customer's region. Which control satisfies the requirement?
MCQ-96A supplier designs, builds and delivers a system, and its contract ends at go-live. The customer owns the source and IP, its architects approved the design, and delivery finished on schedule. Three weeks after go-live, the customer's own engineers must change the billing rules with no supplier support available. Which risk is most likely to stop that change shipping?
MCQ-98A payments team sets a 99.95% SLO and a 99.9% SLA. Both are computed from the same request-success ratio over the same 28-day window, and both apply the same maintenance-exclusion rules. What does the 0.05-point gap between them buy the team?
MCQ-51A team ships quarterly. Their release is a three-day manual regression pass, deploys run a shell script over SSH, and rollback is a wiki page of manual steps last exercised eight months ago; there is no error-rate dashboard or per-request tracing. They want weekly releases. Which sequence of investments should they make?
MCQ-66A payments team ships once a month inside a two-hour maintenance window. Two of the last three releases needed rollback; each manual rollback consumed most of the window, and once the service stayed degraded past it. Which deployment automation should they adopt?
MCQ-78A vendor hands a production platform to the customer's own on-call team at the end of an engagement. With one day left, the delivery lead wants the single check most likely to expose that the team cannot yet run the system unaided. Which check?
MCQ-86Your team is handing a payments service to a new on-call team; your access ends when the two-week overlap does. Terraform for every resource, a runbook per paging alert, and CI that publishes test coverage are already in the repo, and the architecture docs are current in the wiki. Which artifact is worth the most of the remaining overlap?
MCQ-88A team runs a single production system that changes about twice a year. The engineer who built it left last month, and the only record of its configuration is a wiki page last edited three years ago. The customer asks why they should adopt infrastructure as code. Which argument is strongest?
MCQ-72You run the checkout platform for a retail customer. At 14:05 you pushed a config change that cut the connection pool from 50 to 5. By 14:20 the customer's checkout error rate is 12% and 40 tickets have landed in their support queue. You know the cause; the rollback will finish around 15:00. What do you send the customer now?
MCQ-95Your team owns the checkout service. A product manager files a requirement: 'Retry every declined payment authorization until it succeeds.' The ticket's stated goal is 'stop losing sales to transient gateway errors.' Last quarter, 0.4% of authorizations were declined: 92% were hard declines (stolen card, closed account) and 8% were gateway timeouts. Retrying the hard declines created 1,900 duplicate holds that support spent three days clearing. What should you do?
MCQ-82A partner's API answers 200 with `{"status":"error"}` for declines, omits `settledAt` on refunds, and enforces 20 req/min per account although the docs say 60. You already generate a typed client from their OpenAPI spec, and 14 call sites use it, each checking `status` itself. Which change removes the most future work?
MCQ-83You have one 15-minute slot with the VP of Operations, who owns the storefront, has no distributed-systems background, and must approve the consistency setting before Friday's release. The analytics team cannot produce numbers before the meeting. Which framing is most likely to get her to a decision?
MCQ-84A consultancy is writing architecture decision records for a client's payments platform. The client's platform team will maintain the system after the supplier hands it over. Which reason best justifies storing the ADRs in the client's repository rather than the supplier's wiki?
MCQ-97A platform team presents a new payments integration to a customer in a single 40-box diagram. In the review the customer's finance director asks which components drive the monthly cost, their security lead asks which hops are still undecided, and two weeks later procurement quotes the message broker as a committed purchase: the team had not yet chosen between Kafka and SQS. Which change to the diagram set would have addressed all three?
MCQ-29Six months after a 40-engineer team spread across three time zones took over a 200-service platform, services are reading each other's database tables directly and three teams have independently re-implemented billing. The architects who drew the original boundaries have all left. Which practice would most reliably have prevented this?
MCQ-30You recommended a managed database service to replace a customer's self-managed Postgres. The sponsor approved it on cost grounds. The platform team: four engineers who are on call for it, hold quarterly failover drills on Postgres, and have never run the new engine in production: objects. How should their objection be weighted?
MCQ-38A customer asks what happens if their supplier's lead architect leaves. Which answer is strongest?