Skip to content
Tech Interview Prep

Top 100 Technical Product Manager 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-60A "weekly activated users" KPI runs COUNT(DISTINCT user_id) over an IN-list of 12 event types that includes cta_view, activation_complete and imported_data. Week 1 reads 80,000. The eligibility contract defines an activated user as one who fired activation_complete or imported_data. The audit: activation_complete 41,000 distinct users; imported_data 12,000; 6,000 users fired both; cta_view 62,000; the union of all 12 types 80,000. The roadmap and six hires were sized on the 80,000. What is the KPI's correct Week 1 value?
MCQ-61A dashboard titled 'average discount per order' runs SELECT AVG(discount_pct) FROM orders and reports 20.0%. The orders table holds 100 rows: 60 have a promo code and their discount_pct averages 20%, and 40 have discount_pct NULL because no promo code was used. The product spec defines NULL in this column as 'no discount applied'. The PM wants the average discount across all 100 orders. Which change should the review approve?
MCQ-58A revenue dashboard runs `SELECT SUM(o.amount) FROM orders o JOIN order_items i ON i.order_id = o.id`. Three orders exist: two at $100 (one with 2 items, one with 1 item) and one at $300 (3 items). The dashboard reports $1,200; finance's ledger says $500. What is the correct fix?
MCQ-55A signup funnel reports a 71% activation rate. The step is computed as: ```sql SELECT count(*) FILTER (WHERE e.event = 'activated') AS activated, count(*) AS joined_rows FROM signups s INNER JOIN activation_events e ON e.user_id = s.user_id; ``` `activation_events` holds at most one row per user. 10,000 users signed up in March: 4,400 emitted an 'activated' event, 1,800 emitted other events but never activated, and 3,800 emitted no events at all. Which conclusion is correct?
MCQ-59A daily job writes one last-touch channel per conversion. Monday's run wrote conv 8817 → paid_search; Tuesday's rerun over the identical snapshot wrote conv 8817 → email, moving $90k of spend between the two channels. The job runs: ```sql SELECT t.conv_id, t.channel FROM touchpoints t JOIN (SELECT conv_id, MAX(ts) AS max_ts FROM touchpoints GROUP BY conv_id) m ON t.conv_id = m.conv_id AND t.ts = m.max_ts; ``` and keeps the first row per conv_id. conv 8817 has two touchpoints at ts = 10:00:00, channels paid_search and email. Which option is the correct diagnosis and fix?
MCQ-56A checkout funnel is defined as ordered steps `view_pricing` → `start_checkout` → `pay`. The dashboard reports 2.1% start_checkout→pay, while finance's ledger shows 8.4% of pricing-page sessions ending in a paid invoice. The `pay` event fires only on the client-side success page; 12% of invoices are settled through a server-side webhook that never renders it. Growth killed a checkout experiment on the 2.1% figure. Which change addresses the cause?
MCQ-14Helio sells a team-invite product. In Q2 it recorded 50,000 signups. The anti-fraud filter flagged 10,000 of them as automated or internal before billing. 25,000 of the remaining accounts clicked Invite teammate. 8,000 sent an invite that a teammate accepted within 28 days of signup, and another 1,200 sent an accepted invite only after day 28. The signed KPI contract (v2024-06) defines activation as first accepted invite within 28 days of signup over the denominator signups excluding accounts flagged internal or automated. The growth deck reports Activation 32% against a 30% benchmark and asks the board to triple the referral budget. Which figure should the analyst report to the board, and on what basis?
MCQ-13A docs product's PRD names total signups as its North Star. Last quarter a referral contest added 220k signups: total signups went 500k → 720k. Of those contest signups, 91% never created a doc. Weekly collaborators went 41,000 → 41,200; docs with at least two editors went 18,000 → 18,100; infrastructure cost rose $28k/month. The board cites the 44% signup lift and the PRD asks to keep total signups as the North Star and add a signup dashboard. What should the review require before approving?
MCQ-57On Jan 8 a paid template campaign ran for seven days. It brought 3,000 signups, 2,400 of them on Jan 13-14. The growth dashboard reports calendar-week retention (active Jan 8-14 and again Jan 15-21) at 40%, up from 25% the prior week. The product team's tenure-based day-7 retention for the same Jan 8-14 signup cohort is 18%, down from 25% for the prior cohort. Day-7 counts a user only if active exactly on the seventh day after signup. The board is about to fund a repeat campaign on the strength of the 40%. Which read is correct?
MCQ-62A consumer app reports DAU on UTC calendar dates. US Pacific users (PST, UTC−8) are 45% of DAU with activity peaking 18:00–23:00 local; India (IST, UTC+5:30) is 20%. The UTC series shows a 9% dip every Sunday. To recover it, growth sends a re-engagement push at Sunday 20:00 UTC. The next week DAU is back to baseline, but India uninstalls rise 0.6% and support logs complaints about overnight notifications. What is the correct read and fix?
MCQ-66A consumer app counts DAU by device id: 100,000 on the day before a login-based identity merge ships. Telemetry shows 9,000 of those device-days belong to people active on a second device the same day. The day after the merge, DAU reads 91,000 while sessions, signups and revenue are flat. What should the team do?
MCQ-63A growth team spent $180,000 on a paid channel last quarter. The acquisition dashboard shows 12,000 signups and a CAC of $15, and the team plans to double the channel's budget. A fraud audit then finds that 4,500 of those signups were automated accounts that never verified email; of the remaining signups, 3,000 activated. Which action should the team take before deciding on the budget?
MCQ-67A subscription product ships a pricing-page change on Monday 09:00. Its activation metric is built from events that can arrive up to 48 hours late. At Monday 17:00 (T+8h), the dashboard shows activation at 41%, down 6 points from the 47% baseline, so the team rolls back. On Wednesday 09:00 (T+48h), the same Monday cohort recomputes to 47%. Which conclusion should the team act on?
MCQ-65An A/B test was configured for a 50/50 split and expected 200,000 users. It logged 103,000 in control and 97,000 in treatment. The primary metric shows treatment up 2.1% with p = 0.03. What should the team do?
MCQ-64An experiment on the checkout page assigns each page view to treatment or control with a client-side coin flip. Over 12 days it logs 480,000 page views from 96,000 users. The metric is 7-day purchase rate per user: treatment 6.8%, control 6.5%, +4.6% relative, p=0.01. A diagnostic shows 41% of users in the treatment group also had at least one control page view. Which change addresses the cause?
MCQ-15A team ran a 14-day, 50/50 A/B test of a one-page checkout across 210,000 sessions. Before launch the PM registered the rule: ship only if conversion lifts at least 2% relative to control, and abort if any guardrail (refund rate, checkout p95 latency, accessibility defects, support contacts) moves past 1.4x its control value for the full window. Control: 3.6% conversion, 1.1% refund rate, 240 ms p95, 12 accessibility defects. Treatment: 3.9% conversion (p = 0.03), 1.9% refunds, 410 ms p95, 140 accessibility defects. Which decision follows from the registered rule?
MCQ-71A checkout redesign moves conversion from 2.0% to 2.2%. The funding deck that paid for the work assumed the redesign would add 10 percentage points of conversion, and its headline reads "+10% conversion". Before the win is claimed in the quarterly review, what does the data actually support?
MCQ-72A grocery-delivery app ran free delivery through September in 14 metros. Portland and Austin stayed on paid delivery because the promo eligibility flag was misconfigured there, and both sit close to the median promo metro in size and order frequency. Orders per active user came in as follows: | Metro group | August | September | | --- | --- | --- | | 14 promo metros | 3.0 | 3.6 | | Portland, Austin (no promo) | 3.3 | 3.5 | The growth team wants to keep free delivery permanently and is quoting the promo metros' 3.0 to 3.6 rise as the promo's effect. What is the best estimate of the promo's effect on orders per active user?
MCQ-73A checkout redesign ran for two weeks. Overall conversion rose from 3.2% to 4.0%. In the US it fell from 6.0% to 5.5%; in the EU it fell from 2.0% to 1.8%. US share of sessions rose from 30% to 60%. The product review asks whether to attribute the overall lift to the redesign. What should the analyst conclude?
MCQ-39A checkout service reads the flag `checkout_v2` from its config service. The flag has a default of `false` and two rules, evaluated top-down with the first match winning: rule 1 returns `true` for `userId` in the segment `beta_testers`, which is 5% of all users; rule 2 is a 40% rollout that returns `true`. A user who matches neither rule gets the default. What share of all users is served the new checkout?
MCQ-4Two items compete for the same 6 engineer-weeks this quarter. The team's RICE rubric sets confidence from the reach evidence: a query over the full target population = 100%, a sample drawn from it = 80%, an internal opinion = 50%. 'Bulk export': reach 1,000 accounts/quarter from a query of every account with more than 500 rows; impact 2; effort 2 engineer-weeks. 'Custom roles': reach 2,000 accounts/quarter from 12 interviews with users drawn from the target segment; impact 3; effort 4 engineer-weeks. Which item ships first?
MCQ-9A payments team is sequencing three work items for the next quarter. Each has a measured Cost of Delay per week and an estimated duration: a checkout retry fix (CoD $60k/week, 6 weeks), a fraud-rule update (CoD $30k/week, 2 weeks), and a settlement-report rewrite (CoD $24k/week, 4 weeks). The team will run them back-to-back in the order chosen. Using WSJF, which order should the team commit to?
MCQ-5A team is ranking two RICE items for a 12-month horizon. Assume 160 hours per person-month and 52 weeks per year. Item A: reach 1,600 users/year, impact 2, confidence 80%, build 2 person-months, and 10 support hours/week after launch. Item B: reach 1,200 users/year, impact 1, confidence 100%, build 1 person-month, and 2 support hours/week after launch. The team currently ranks A first using coding-only effort. Which statement gives the correct lifecycle RICE scores and the resulting priority?
MCQ-94A checkout team scores a redesign with RICE: reach 2,000,000 users per quarter, impact 3, confidence 0.8, effort 2 engineer-months. 2.4M, the roadmap's highest; the next item scores 1.1M. The accessibility audit finds the new Place order control is a div with no role, so the 16,000 quarterly screen-reader users (0.8% of reach) cannot complete checkout at all. Screen-reader use is not a declared account attribute. Fixing the control adds 1 engineer-month. The legacy checkout is decommissioned in the same release because it fails the same audit. Counsel says the block excludes a protected class. What should the team do?
MCQ-6Your team has 10 engineer-weeks next quarter after on-call and holidays. Three requests are on the table: a contract-mandated audit fix (4 weeks, must ship by week 8 or the $1.2M renewal is blocked), a latency migration (5 weeks, p99 900ms→250ms, no external deadline), and a feature for the largest account (4 weeks, no deadline, but their champion says they will evaluate a competitor if it slips two quarters). The VP asks for a plan. Which plan should you commit to?
MCQ-7A B2B SaaS team has a fixed six-week release window and a backlog of 14 items that stakeholders have each tagged Must. The release's defined viable outcome is that the enterprise prospect can sign and existing customers keep processing lawfully. The backlog includes a GDPR consent receipt (legal), SSO for that prospect (sales), an invoice logo (sales), a CEO colour tweak, and ten others. Velocity over the last three releases is about five items per window. How should the PM handle the Must list?
MCQ-20A payments team of eight engineers runs two-week sprints: 80 engineer-days of nominal capacity, minus 8 for the on-call rota and 8 for interrupt-driven support, leaving 64 engineer-days of planned work, and the roadmap books all 64 in each of sprints 4 and 5. Discovery happens in whatever time is left; across the last three sprints the team logged six hours of it, and the pricing change that shipped on an untested assumption was rolled back nine days after launch (that assumption had been rated the least risky of the quarter's bets). Three assumptions behind the next roadmap bets need spikes the tech lead estimates at 8, 6 and 6 engineer-days, and repricing must be decided before the contract renewal at the start of sprint 6. Over the last three sprints, on-call and support consumed 15, 16 and 17 engineer-days. Which plan does the capacity arithmetic support?
MCQ-48A payments team of six engineers plans in 13-week quarters. Twelve features a quarter touch the pricing module, and estimates on those features have climbed from 3 days to 9 days. Two of last quarter's four rollbacks were in pricing. The team is deciding how much of next quarter's capacity to set aside for pricing's debt. Which plan is right?
MCQ-100A 10-engineer team plans a 13-week quarter at 40 h/week, so the roadmap commits 5,200 engineer-hours to features. Last quarter's timesheets show 1,300 hours went to KTLO (on-call pages, security patches, customer escalations) and the team logged 780 hours of overtime (6 h/engineer/week) to hit the feature dates. Two engineers have said they will leave if the overtime repeats. Next quarter the team wants a feature commitment it can hit at 40 h/week with no overtime. Which commitment does last quarter's data support?
MCQ-24Two engineers, four working days left in the sprint. The Sprint Goal is the checkout tax fix: it needs four more days of both engineers, and the bug is costing about $3k/day in refunds. On day six the VP of Sales forwards a board member's request: a filtered CSV export for a $400k renewal that closes in three working days. The export is estimated at four engineer-days. What should the PM do?
MCQ-27A six-engineer team has eight weeks before a pilot with 40 operations admins, all from the same mid-market customer. Capacity is 48 engineer-weeks. The pilot succeeds only if an admin signs up, invites two teammates and runs one saved report unaided. One-segment end-to-end scope estimates at 11 engineer-weeks; the layer breakdown is data model 16, API 15 and React screens 17, and nothing integrates until the last track lands. Two smaller segments, 4 engineer-weeks each, sit on the roadmap for after the pilot. Last quarter the same team closed 12 layer stories in seven weeks and its first end-to-end demo failed on tenancy for 9 of the 20 pilot accounts. Which plan should the team commit to?
MCQ-74A two-engineer platform team has 8 weeks left in the quarter: 16 engineer-weeks after KTLO and on-call. Two items are committed. Migrating off a deprecated auth SDK takes 6 engineer-weeks; the vendor's EOL is in 4 weeks, it offers no extended support, and after the EOL the pinned fork carries an unpatched CVE with auth p99 going from 180 ms to 900 ms during incidents. The $240k prospect's SSO feature takes 14 engineer-weeks and its contract date is the end of the quarter, 8 weeks out; the prospect is still in security review and has already moved its date once. What should the team do?
MCQ-10A B2B SaaS team runs a Now/Next/Later roadmap. Later holds 14 items; three carry a date or quarter: 'Q3', 'Oct 15', 'June 30'. The June 30 date is the marketplace integration, which sales put on a renewal slide in March to close a $480k account. Engineering has not started discovery on it, and the date has already moved twice. Two accounts churned in Q1; both exit notes cite the roadmap slide. What should the team do?
MCQ-25A B2B SaaS company keeps one canonical roadmap: 14 items, each labeled Committed, Directional, or Later. Three artifacts reach customers and the board. The board Gantt is generated nightly from canonical and matched it within 24 hours all quarter. The 22-slide sales deck template is generated weekly from canonical. A 40-slide shared library, hand-maintained by PMM, holds customer-proof slides that AEs paste into decks. The support KB page was last edited seven months ago and carries no dates. Last quarter, 9 of 12 enterprise deals cited a Marketplace date that appears in the library's Marketplace slide (H1 2026) but not in canonical, where Marketplace is Later/directional; one deal settled at a $250k credit. Which change removes the cause of the 9 citations?
MCQ-11A B2B SaaS vendor's sales team attached the Q3 roadmap PDF as Exhibit B to an MSA. The PDF lists 11 items under 'Later'; one is already marked Won't. The customer signed and now cites Exhibit B as a delivery commitment, demanding all 11 by quarter-end. Legal estimates 6 weeks and $180k in credits to unwind. Which response is the defensible product decision?
MCQ-12Q3 planning minuted "18% ARR growth" as a commitment. The team's evidence-based range was 12–15%. September closed at 13.4%, and two directors called the quarter a broken promise. The board pack goes out Friday. What should the planning lead put in it?
MCQ-26A growth team's Q3 recap opens with "47 stories shipped, 100% of committed scope delivered, no P1s." The board's goal for the quarter was activation, the share of new signups completing the first key action within 7 days, which moved from 31% to 31%. What should the engineering lead put in the recap?
MCQ-16A B2B editor team ships a template gallery. In the launch month editor adoption rises 18% and the launch dashboard is green. Finished documents are flat at 0% growth, and paid-seat revenue falls 6% that quarter and another 4% the next. The team's metric tree claims adoption → finished docs → retained paid seats. What should the team conclude about adoption as a leading indicator?
MCQ-17A tax-filing product ships a "stay in product" badge that recommends related articles after each field is completed. The following month: sessions per user rise 30%, return completion falls from 71% to 64%, median time-to-complete rises from 22 to 31 minutes, and filing-fee refunds reach $62k. CES after submit drops from 4.4 to 3.9. The team's HEART dashboard reports only sessions per user, so it shows the launch as a success. Which change to the goals-signals-metrics mapping is correct?
MCQ-1A team's backlog story reads: "As a seller, I want a two-way Salesforce sync so that I can see accounts." They shipped it in 11 weeks. Two months on, 64% of sellers still copy fields by hand and support logs 180 tickets/week on duplicate accounts, while the sync's own dashboard reports 99.4% record sync success. The PM has one sprint before planning. What should the PM do?
MCQ-2A wallet product has 2,100,000 enrolled accounts. The story that shipped SMS OTP said only that "user accounts must be secure": it named no asset, no attacker and no residual-risk owner. Six weeks after launch, forensics put the 340 account takeovers on two vectors: 268 where a SIM swap moved the code to the attacker's handset, and 72 where a real-time phishing proxy relayed the code from the victim's own phone. Median loss per takeover was $1,180. Policy now requires every story, before it is marked Ready, to state the residual risk its chosen control leaves and the person who owns it. The rewrite is testing app-generated TOTP with no SMS fallback, and the attack mix is expected to hold. Which statement belongs in the acceptance criterion for that control?
MCQ-3A checkout team's launch gate reads "checkout is fast and retries are safe." QA signed it off on office Wi-Fi. Two weeks after launch, RUM shows p95 Pay→order-id of 4.6s on 4G against the 800ms budget, 9% of payment attempts double-submitted, and $47k in chargebacks. The team is rewriting the gate. Which replacement criterion fits this situation?
MCQ-28Two tickets are candidates for the next sprint. Ticket A scores 14/14 on the team's Definition of Ready checklist (criteria, estimate, design link, test plan) but its empty-state and error-state fields both read "TBD". Ticket B scores 6/14, missing the estimate and test plan, but in a ten-minute walkthrough two engineers and one designer each described its empty, error, and success states from memory and agreed. Last quarter the team shipped a 14/14 ticket whose empty-state field read "TBD"; the release rendered a blank page for accounts with no saved card, 22% of new accounts never added one, and the fix took 5 days. Which ticket is Ready, and on what basis?
MCQ-19An integration team pre-registered a kill line before GA: at week 8, sunset the integration if fewer than 25% of the 800 accounts that installed it have completed a successful sync, or if support load exceeds 15 hours/week. At week 8, 200 accounts have attempted a sync and 168 completed successfully; support load is 12 hours/week. The launch dashboard is green: it tracks installs against the 150-account launch goal, and 800 accounts installed. What does the pre-registered criterion require?
MCQ-21A PRD's stakeholder section lists a 47-person distribution list and a RACI with 3 people marked Accountable, 12 Consulted, and 31 Informed. The escalation path reads "raise it in #product". The last two scope cuts took 3 weeks and 19 days to land, and the launch slipped 11 days past its event window. What should the reviewer require before approving?
MCQ-22A platform team owns a shared auth service used by all 40 enterprise accounts. The Q3 GA of a new token-refresh flow is on the critical path, with 2 of the team's 6 engineers on it. A VP of Sales asks for a per-tenant fork of the refresh path for one account ($400k ARR, renewal in Q4) whose actual blocker, a missing SCIM attribute mapping, is already in the shared flow's Q3 scope. The last per-tenant fork in this service took 6 weeks to unwind and doubled auth incidents for two quarters. What should the team do?
MCQ-23A payments platform team and a mobile team left a joint design meeting with different memories. Payments built a retry queue with fixed 30-second intervals; mobile shipped exponential backoff with jitter. The staff engineer who proposed the chosen approach has left, and the only shared trace is a Slack thread with two competing summaries. The incident review wants a durable fix. Which action best addresses the root cause?
MCQ-29A payments team is on day 8 of a 10-day sprint for a checkout release. The PRD excluded keyboard-only navigation for the new address form. A late change asks to include it. The company's Definition of Done requires keyboard access on all customer-facing forms, and the launch date is fixed by a marketing campaign. The team has one QA engineer and no spare sprint capacity. What should the review decide?
MCQ-30A payments team's board shows the story 'Add CSV export to reports' as Done: all four acceptance criteria pass, the demo ran green against a stubbed payment service, and the code merged Thursday. Friday's release produced a 5xx spike: 8,000 users hit a partial write, the logs carry no trace id, and rollback meant redeploying yesterday's build, which took 47 minutes. The team wants the one change that would have caught this before the release. Which change does the situation call for?
MCQ-46A payments team ships a checkout change every two weeks. Last quarter three accessibility defects reached customers: keyboard focus was trapped inside the card-entry iframe, so a keyboard-only user could not reach Pay; after a validation error focus stayed on the submit button, so screen-reader users never heard the error; and a country dropdown built from divs announced nothing. All three passed the team's axe scan with zero violations. Each was filed as a follow-up ticket, median report-to-fix was 19 days, and one customer left a public review. The definition of done currently requires unit tests, code review and a clean axe run. Which change to the definition of done would have kept these three defects from reaching customers?
MCQ-47A multi-tenant billing API authenticates the session, then loads an invoice by primary key: `SELECT * FROM invoices WHERE id = :id`. The row carries `tenant_id`. A pen test showed an authenticated user of tenant 7 requesting `/invoices/90210`, tenant 12's invoice, and getting 200 with the PDF. The fix must ship this sprint and must make a cross-tenant request indistinguishable from a missing invoice. Which change does that?
MCQ-70A vendor stores patient messages in Postgres. Client-to-service traffic uses TLS 1.3; message bodies are AES-256 encrypted with a KMS key whose policy allows the application role to decrypt; support engineers can run an admin query that returned 1,240 message bodies during a 30-minute incident. A hospital RFP asks: 'Are messages encrypted so that no vendor employee can read them?' Sales must answer in writing. Which answer is accurate and defensible in the audit?
MCQ-81An EU enterprise customer will sign only if the DPA clause 'customer personal data is stored and processed exclusively within the EU' is true at signature. Engineering maps the flows: primary Postgres and nightly backups in eu-central-1; a product-analytics feed to us-east-1 that sends only daily aggregate feature counts with no per-user rows; a US-hosted support desk where agents paste ticket bodies containing customer names and emails; and CDN access logs in us-east-1 holding client IPs with 30-day retention. Which remediation set makes the clause true?
MCQ-85A B2B SaaS sells a "Dedicated" tier at 3× the standard price. Isolation today: shared Postgres with RLS (a tenant_id predicate on every table, covered by tests), one KMS key per tenant, one backup bucket per tenant. Support runs a third-party BI tool through a service account that holds BYPASSRLS so it can also run schema migrations. A pilot audit logged one BI query returning rows from 4 tenants. The contract clause buyers will sign reads "no other tenant's data is reachable by your staff." Which change makes that clause true?
MCQ-95A partner integration needed MFA disabled for its service accounts. The exception was approved in a Slack thread, applied with an `mfa-exempt` label, and never given an expiry. Fourteen months later 8,000 production accounts carry the label and one of them was used in a credential-stuffing incident. Which change to the exception process would have caught this before the incident?
MCQ-43SeatWave sells concert seats. Seat inventory is written to one Postgres primary in eu-west-1; the seat map and the per-event like counter are read from a replica in eu-central-1. On 12 March a 14-minute link failure cut the replica off. The primary stayed up and kept accepting bookings, but the booking flow checks availability against the replica before writing, so 220 seats were sold twice and refunds plus goodwill came to $48k. The like counter drifted by up to 90 seconds and no user complained. The venue contract forbids a double-sold seat. Which policy should SeatWave write down for the next partition?
MCQ-45A PRD for a 'faster repeat checkout' feature names one purpose: cut repeat-checkout time for signed-in users. The proposed analytics event schema writes five fields: the card token, the cardholder name, the billing postcode, a device fingerprint, and a keystroke stream from the address form. The PRD states three uses for the table: measuring time-to-checkout from the event's start and submit timestamps; counting how often a saved card is reused, which needs the card token; and breaking checkout time down by region, which needs the billing postcode. Retention is 400 days. The metric is computed over a rolling 90-day window. Which review decision does purpose limitation require?
MCQ-36A payments platform is retiring API v1. v2 has been live for two quarters. The deprecation dashboard shows v1 at 0.8% of requests, so the team schedules deletion for Friday. A pre-delete audit finds one white-label partner still calls v1 from a cached client; its traffic is 0.8% of requests but 12% of GMV, and it has no v2 integration. A kill-switch can restore v1 routing within 15 minutes. What should the team do before Friday?
MCQ-52A signup service writes to a Postgres `users` table and publishes `user.created` events (Avro) consumed by 14 services. To satisfy a new KYC rule, one deploy makes `middle_name` required end to end: the signup form, the event schema, and the `users` column. Within an hour, signup completion in the 9 locales where middle names are rare falls from 71% to 53%, and the fraud scorer rejects every new event with `Missing required field: middle_name`. The ledger writer, which reads the `users` table directly, is unaffected. Which change restores signup completion and keeps the KYC field?
MCQ-40A payments ledger appends to a primary in us-east-1; replicas in us-east-1, eu-west-1 and ap-southeast-1 stream from it asynchronously with apply lag p50 300 ms, p95 1.8 s, and 6.2 s during the 01:00 UTC bulk import. After POST /transfers returns 200, the app renders the confirmation screen and the receipt from a replica's copy of the sender's balance. On 2024-11-04, 3,400 transfers had a second tap and $1.1M was debited twice, because the receipt showed the balance from before the debit. Audit requires the receipt to state the post-debit balance, and the transfer API's p99 write budget is 120 ms (85 ms today; a round trip from us-east-1 to eu-west-1 is 140 ms and to ap-southeast-1 is 210 ms). Which change fixes the duplicate debits?
MCQ-49A parcel API returns `weight` in pounds. A new EU carrier will start reading the same endpoint next week and expects kilograms. Its integration is fixed: it can read a new field but cannot switch to a new API version or perform unit conversion before onboarding. Existing US carriers read `weight` as pounds and handle 1,100 parcels/day. The team plans to change `weight` to kilograms in place and update the docs. What should the team do?
MCQ-50A team replaced a $48k/year vendor with an in-house build: 8 engineer-weeks at a fully loaded $4,000 per engineer-week, a $30k one-time migration and compliance review, and 0.5 FTE of ongoing maintenance at $220k fully loaded per year. The vendor's fee covered 24/7 support and included a 30-day exit with full data export. The launch deck compared the $32k of build labour against the vendor's $48k year-1 fee and called it a win. Which statement about the three-year comparison is correct?
MCQ-51A user-facing "similar cases" feature runs on a managed vector database holding 40M embeddings. The vendor's bulk export API is documented at 10,000 records/hour, the contract renews in six weeks at a 3× uplift, and the product's disaster-recovery requirement is a 4-hour restore. The team plans to sign the renewal and treat the export API as the exit path. Which assessment of that plan is correct?
MCQ-31Your orders API returns HTTP 200 with body {"status":"OUT_OF_STOCK"} when inventory is unavailable. The OpenAPI 3.1 document passes Spectral lint with zero errors, and CI blocks merges on that gate. Three partner clients treat any 2xx as order-created; last quarter 1,900 out-of-stock orders were fulfilled anyway, costing $210k in warehouse variance. Which change stops the recurrence?
MCQ-37A payments team adds an optional `loyalty_tier` string to the `GET /orders/{id}` response so the web checkout can render a tier badge. The OpenAPI diff marks it nullable and not required, and the spec linter flags the change as additive. The canary runs at 5% of read traffic. Eight minutes in, the Java checkout service, a generated Jackson client with `FAIL_ON_UNKNOWN_PROPERTIES` at its default, throws `UnrecognizedPropertyException` on every order response; its 5xx rate goes from 0.03% to 100%. That service reads 30% of all orders, its requests carry `Accept: application/json;charset=UTF-8`, and its generated client is checked in and shipped unchanged by the release train, which is 11 days out. What should the team do?
MCQ-32A payments API exposes POST /v1/charges. The mobile SDK retries any request that has not returned within 2 s. Under peak load the endpoint's p99 rose to 3.2 s, and a 50,000-charge test run produced 4,000 duplicate captures, each a second charge on the same card for the same order. Which change makes the retries safe?
MCQ-33An export API accepts work with POST /v1/exports: the handler commits a job row and replies 202 Accepted with the job id, and a worker pool then runs the export, which takes 40 s to 6 min per file. The official client library aborts a POST at 2 s and immediately re-sends the identical body. Last week users tapped export 50,000 times, the server logged 53,000 POSTs and committed 53,000 job rows, and server-side POST latency was p99 4.1 s with 6 percent of first attempts running past the client's 2 s limit. Every POST body carries a submission_id that the client generates per tap and re-sends unchanged on the retry. Users also re-export the same file minutes later with different parameters, and those jobs are wanted. No new client build can ship this quarter. Which change should ship?
MCQ-34An invoices list endpoint runs `SELECT ... ORDER BY created_at DESC LIMIT 50 OFFSET :n`. A nightly batch inserts ~2,000 invoices/min, stamping `created_at` from each invoice's business date, so hundreds of rows share one `created_at` value and id order does not match `created_at` order. Support logs 40 'missing invoice' reports a week, all from customers who paged past page 50 during the batch. The team replaces OFFSET with a cursor: `WHERE created_at < :last_created_at ORDER BY created_at DESC LIMIT 50`. The reports continue. Which change removes the missing-invoice reports?
MCQ-35A checkout API returns HTTP 500 for every failure. A nightly batch submits bad SKUs at 400 requests/min, and all of them come back 500; the third-party client SDK is pinned until its next release in six months, decides retries from the status code alone (retries any 5xx up to five times, never retries a 4xx), and ignores the JSON body. That path now peaks at 2,400 requests/min, p99 latency on the endpoint rises from 120 ms to 900 ms during the batch, and support MTTR is 4.2 h because the tickets carry no error type. A warehouse outage also returns 500. Which change reduces the retry amplification while keeping genuine dependency failures retried by the same SDK?
MCQ-38Your API's free tier allows 60 requests/min per key, but that number is only in a support macro, and every 429 has an empty body. Last quarter, 38% of new developers who hit a 429 in week 1 never made a second request, and support spends 22 hours/week explaining the limit. Paid tiers are 600 and 6,000 rpm. Which change should ship first?
MCQ-83A feed API issues opaque cursors with a 30-minute TTL. When a cursor is older than the TTL the endpoint returns `401 {"error":"cursor expired"}`. A shared client interceptor treats every 401 as an auth failure: it clears the stored token and routes to the login screen. Users who leave the tab open and come back later get logged out; support logs roughly 40 "logged out for no reason" tickets a day, and 12% of feed requests carry an expired cursor. The feed is mutable, new items arrive continuously, and a third-party client also consumes the API. Which change fixes the reported symptom?
MCQ-96You are retiring `/v1/export`. Thirty days of telemetry show 41,000 requests from 260 API keys: three keys produce 91% of the requests, the other 257 produce 9%, and 40 of those 257 call exactly once a month for billing reconciliation. A 48-hour canary disabled the endpoint for 5% of requests and reported zero errors, with the canary cohort's request rate matching the control. The full disable is scheduled for Friday. What should you do?
MCQ-41A retailer's CDN caches `GET /product/:id/price` with `Cache-Control: max-age=3600`. A flash sale ends at 14:00, but the CDN keeps serving the 20%-off price until the entry expires at 14:47; 1,900 orders are placed at the stale price and the retailer absorbs 310 chargebacks. Price requests are 40% of the 12,000 rps peak, and the origin saturates at 3,000 rps. Which change contains the next incident without saturating the origin?
MCQ-42A video service enqueues transcoding on upload, and the web UI shows an unbounded spinner until the job finishes. A load test with re-uploads disabled drains the queue to zero in 90 seconds at p99, with the worker pool at 70% utilization. In production, 18% of uploads are re-uploaded after roughly two minutes of spinner, and each re-upload re-enqueues the full source file; peak queue depth reached 4,200 jobs and drain reached 11 minutes. Uploads are up to 2 GB from mobile clients. Which change addresses the cause of the 11-minute drain?
MCQ-77You're reviewing the PRD for offline editing in a collaborative document editor. Edits buffer in a shared write-back cache node and flush to the primary store every 10 minutes. The nodes are stateless containers with no local disk, the flush interval is fixed by the storage team's write-amplification budget, and the cache tier has no headroom for a second copy. The incident review records one node death that lost 6 minutes of unflushed edits across 900 documents open at the time, with no client-side recovery; support tickets cite it and trust NPS fell 11 points. The PRD's acceptance criteria are a green launch dashboard and an NPS re-check 30 days after launch. Which statement should the review accept?
MCQ-76A notes app writes to a single primary in us-east-1 and serves reads from regional replicas. Replica lag is p50 40 ms and p99 3.2 s. A user edits a note on their phone, then opens the same note on their laptop 1.5 s later; the laptop's read hits a replica that has not applied the write, so the user sees the old body, adds a paragraph, and saves over the phone's edit. Support logs 200 "lost my edit" tickets a week. The primary already runs at 70% CPU, and moving EU reads to it would add 120 ms RTT. Which change removes the reported symptom?
MCQ-78A worker pops welcome-email jobs from SQS (30 s visibility timeout), POSTs to the email provider, then acks. p99 send latency is 800 ms. Since last week's deploys, 400 of the 40,000 daily welcomes (1.0%) arrive twice, and every duplicate pair in the logs is a POST that returned 200 OK, then the worker being killed mid-deploy before its ack, then SQS redelivering the job 30 s later. The provider returns a receipt id and honours a caller-supplied idempotency key for 24 hours. Deploys roll the workers about ten times a day. Which change cuts duplicate welcomes to zero without losing any?
MCQ-79A live mock-interview product keeps each room's transcript buffer and in-flight turn state in the app node's heap and routes WebSocket upgrades through a load balancer with 60 s cookie-based sticky sessions. At peak the cluster runs 8 nodes holding 40,000 rooms, spread evenly. Ops needs a rolling deploy, one node at a time, that completes inside a 10-minute window at peak and drops zero transcript lines; today a replaced node loses the rooms it held, and a client that reconnects onto another node starts with an empty buffer. Which change meets that requirement?
MCQ-80A gateway sheds load with 429 once inbound exceeds 10,000 QPS; real peak traffic is 3,000 QPS. The mobile SDK retries any 429 or 5xx immediately, with no cap and no jitter. During the last incident inbound reached 15,000 QPS and the shed window ran 40 minutes instead of 12. The team will make one change. Which change keeps inbound below the 10,000 QPS shed point when the same 3,000 QPS peak returns?
MCQ-44A payments API carries a 99.9% availability SLO over a 30-day window, which is 43.2 minutes of error budget. On day 12 the burn dashboard shows 34 minutes consumed. A feature branch is ready to ship; its 24-hour canary shows a 0.05% error rate, and the team expects that rate to hold for the remaining 18 days. The release manager wants a go/no-go. Which call does the budget arithmetic support?
MCQ-84An export service shows a bare spinner while a job runs; p95 is 10 minutes. Of all export jobs, 22% are duplicates. 13 points of that are the same user re-submitting the same export within 90 seconds: the spinner gives no sign the first job is alive. The other 9 points are a partner integration that re-submits the same export every 5 minutes until it polls a completed result; no user is involved. The team ships progress stages, an ETA range, a cancel button, and a completion notification shown in the web app, but no server-side deduplication. What duplicate-start rate should they expect after ship?
MCQ-86A shared pool of 24 workers serves 300 tenants. Acme's nightly reindex is a parallel scan that claims every free worker and totals 600 worker-minutes of work; it starts at 01:00 and must finish by 02:30. While it runs, the other 299 tenants' p95 goes from 45 ms to 610 ms against a 200 ms p95 SLO at 99.9% monthly attainment, and the contract refunds 10% of the monthly fee per 0.1% below target, capped at 100%. Acme pays $1,800/month, the other 299 pay $400 each, a worker costs $250/month, and the infra budget has $2,500/month of headroom. The other 299 tenants need all 24 workers to hold 45 ms; at 12 workers their p95 is 260 ms. Which change fixes the breach within budget?
MCQ-87A multi-tenant API gateway serves all tenants from a shared 400 rps pool. Current total draw is 240 rps, of which Tenant A accounts for 20 rps. Platform policy caps any quota grant at a pool utilization of 80% (320 rps); the pool sheds load above 400 rps, and measured p95 is 40 ms below saturation and 180 ms at saturation. Tenant A asks for its cap to be raised to 200 rps for a 7-day launch. The pool is provisioned at 400 rps and the next capacity expansion is next quarter. What should the platform team do?
MCQ-88The CSV importer shipped 30 days ago. Weekly tickets tagged `importer`: | Week | Tickets | | --- | --- | | 1 | 40 | | 2 | 25 | | 3 | 12 | | 4 | 6 | Median handling time is 15 minutes. The PRD's kill criterion reads: "kill the feature if support load in the week of the 30-day review exceeds 2 h/week." The review is Monday, at the end of week 4. What should the PM do?
MCQ-89A nightly reconciliation job that computes next-day invoices has failed every run since 02:00. If it has not succeeded by 06:00, roughly 12,000 customers will be charged wrong amounts. Nothing customer-facing is broken right now: the API, dashboard and payment flow all work. It is 03:05, and the on-call has a fix re-running that should finish by 03:40. The incident commander asks what to communicate.
MCQ-91A platform team must migrate a shared CI fleet of 30 runners to a new image before the old image loses support at 16:00 on Friday. The migration runs 13:00–16:00 and drains runners in configurable batches; the batch size sets how many runners are drained at once and does not change the migration's duration. The platform team set the batch to 10, leaving 20 runners serving jobs. Baseline load from other teams holds 15 runners busy throughout. A checkout squad's release pipeline needs 8 concurrent runners to keep its deploy queue under the 5-minute SLO during a fixed 2-hour retail promotion window, 14:00–16:00. The platform team's migration dashboard is green. What is the correct call?
MCQ-92A messaging platform's 'Pro' plan costs $99/month and includes 20,000 outbound SMS segments; segments beyond the allowance are billed at $0.005 each. Every segment costs the platform $0.008 in carrier fees. The median Pro account sends 25,000 segments a month, and at that usage the plan's contribution margin is negative. Which change to the plan makes the median account contribution-positive?
MCQ-97A payments team is migrating its ledger from a legacy Postgres database to a new service. For three weeks both systems have received every write, while reads still come from Postgres. The nightly reconciliation job compares the two and reports that 0.4% of accounts hold a different balance: the new service is missing refunds that arrive in a nightly batch file it does not yet ingest. The team wants to flip reads to the new service this sprint and call the migration done. What should they do?
MCQ-68A self-serve analytics product serves a per-account dashboard whose headline metric is lifetime revenue, computed by scanning a 12 TB `events` table. The table is partitioned by day into 365 partitions of about 33 GB, and the dashboard query filters on `account_id` only. The dashboard gets 500 loads a day across 2,000 accounts, and the warehouse bills $6 per TB scanned, so it costs about $36,000 a day. Lifetime totals must stay on the dashboard, and every account must be able to load it at any time. Which change fixes the cost?
MCQ-75An orders service is sharded by tenant_id across 40 shards. The admin order page reuses the order-edit transaction: the sharding layer opens a transaction on every shard the query touches and holds it until the admin navigates away. A prototype search box that accepts a customer email without tenant_id opened transactions on all 40 shards for 8 minutes; checkout p95 went from 180 ms to 12 s. Product wants an all-tenant checkbox on that search box, used about 30 times/day. The sprint has room for one change. Which plan is defensible?
MCQ-82A retailer's search service runs BM25 candidate generation over the full index, then a cross-encoder reranker over every candidate. Measured p95: 50 ms for candidate generation and 1.0 ms per candidate for the reranker, with 1000 candidates per query, so the pipeline lands at 1050 ms p95. The launch gate is p95 ≤ 300 ms, and the reranker is the point of the project: it lifts NDCG@10 from 0.31 to 0.44. Which change ships that lift inside the gate?
MCQ-8Your Q2 roadmap review has to settle on-prem. A regulated-EU bank staffed a 9-person project for an on-prem build 4 months ago; there is still no binary, the account has paused USD 1.2M of ARR, and the AE has twice told them "next half". On-prem asks run 9 of 340 enterprise accounts over 18 months, all of them regulated EU, while cloud ARR is up 12% QoQ and the next two quarters are committed to multi-region cloud. The committee will not build on-prem this horizon. Which action should it take?
MCQ-18A checkout team must be on 3-D Secure 2.x by 1 November. Certification takes four weeks, so engineering freezes the release branch on 1 October. One decision is still open: certify the existing redirect layout or a new inline sheet. Each takes 2 engineer-weeks to build and there is capacity for only one of them. A researcher proposes a study running 1 September to 27 October (1 researcher, 3 engineers for sessions, about $110k) to understand friction in the payment step. The standing release bar for any layout change is 6 of 8 first-time buyers completing payment unaided. What should the team do?
MCQ-53A payments team plans GA in 10 days for a payout feature. The PRD's success metric is 'merchant receives first payout within 24 hours of signup', target 99%. The launch dashboard tracks only API 500 rate and p99 latency. In a 5% canary, 400 merchants signed up; the API 500 rate is 0.1% and p99 is 180 ms, so the dashboard is green. Support finds 14 of those 400 merchants stuck at 'verify bank' with no payout, but the payout API returned 200 because a webhook retry succeeded. Which review decision best fits the launch requirement?
MCQ-54Checkout release R42 adds a nullable `orders.total_cents` column and puts the new pricing path behind flag `pricing_v2`, default off. The flag-on path applies a new discount rule and writes only `total_cents`; the flag-off path reads only `total`, the float column in production today. A backfill job copies `total` → `total_cents` for 12M existing rows in about 35 minutes and never writes `total`. The team's rollback promise is that flag off restores prior user-visible behaviour within 2 minutes, at any point in the release. Their plan: deploy the migration, deploy the code, flip the flag, run the backfill overnight. What does that plan actually deliver?
MCQ-98GA is announced for a keynote six days out and the date cannot move. The launch checklist has three Must gates: security review (green), rollback drill (green), and billing metering (amber. Annual-plan usage is under-counted, so roughly 12% of accounts would be billed $0 for their first month; the fix lands in three weeks). Monthly-plan metering is correct. What should the team do?
MCQ-99An instant-retry payment flow ships to 20% of checkout traffic on day 0. Its launch doc pre-registers the criteria: success is a +0.5 percentage-point conversion lift at the day-14 read; guardrails are duplicate-charge complaints under 1.0 per 1,000 orders and payment p95 under 900 ms; any guardrail breach that holds for two consecutive daily reads triggers an immediate halt and rollback. The day-9 read: conversion +0.3 pp; duplicate-charge complaints 0.4 per 1,000 on control versus 2.6 per 1,000 on treatment; p95 720 ms on control versus 1,180 ms on treatment; both guardrails have been red on every daily read since day 5. What should the launch team do at the day-9 read?
MCQ-69A team is reviewing a PRD for the March board pack. sub_snapshot holds one row per day, written at 23:00 UTC by a nightly job; a failed load leaves no row. sub_events is the append-only billing webhook log of cancellations and starts, timestamped in UTC. The pack must show active subscribers as of 31 March 23:00 UTC, and the subscriber close has to tie to the billing export - the pack's 3% tolerance covers MRR rounding only. The PRD states: March close = the sub_snapshot row with the greatest load_date. The facts: the 30 March 23:00 snapshot holds 40,000 active subscribers; the 31 March 23:00 load failed (the billing API returned 503), so no 31 March row exists; sub_events for 31 March records 1,120 cancellations and 240 starts; 1 April has a 09:40 UTC dunning batch of 1,400 cancellations plus 500 starts; the 1 April and 2 April snapshots both read 38,220; the pack is assembled on Friday 3 April. Which statement should the review accept?
MCQ-90A billing service carries a "technical debt" roadmap item: a 14-week rewrite of the invoice module. The module produces a 2.1% invoice defect rate, the refund median is 9 days, and ledger on-call pages are up 3x. The last rewrite of this module slipped from 14 to 26 weeks with no user-visible change, and two product bets were cancelled to staff it. The sponsor wants the debt item to stay on the roadmap. What should the team do?
MCQ-93Your team shipped a redesigned cancellation flow: the Cancel control is now a small grey text link under a full-width "Keep my 40% discount" button, and clicking it opens a three-step retention survey. Two weeks of data: trial-to-paid conversion +8% relative, monthly churn flat, "cannot find cancel" support tickets 3× baseline, chargebacks 2.4× baseline, and legal flags exposure under the FTC negative-option rule. The PRD review has to decide what happens to the flow. Which decision is right?