Skip to content
Tech Interview Prep
Technical interview guide

Requirements Gathering & User Stories

Turning a vague need into a written, testable requirement: user stories, acceptance criteria, and edge cases.

Read
30 min
Practice MCQs
25
Interview Q&A
25
Edition
v4
Editorial status
Reviewed

Scope: GOV.UK Service Manual, Scrum Guide 2020, ISO/IEC 25010:2023, WCAG 2.2, RFC 2119, OpenAPI, and NIST guidance current 2026-08-31.

Curated: · Written: · Reviewed:

25 questions

MCQ-4A developer who has just joined from a team that wrote stories as a role and a feature only asks why every story here needs a 'so that' clause. Which reason is correct?
MCQ-25An insurer's quoting service and billing service both implement the term "policy effective date". The glossary entry reads: "Effective Date. The date on which coverage begins." After a release, 1,800 of 40,000 policies (4.5%) show a proration mismatch between the two services. On every one of the 1,800, the signed-proposal timestamp and the first-payment timestamp fall on different calendar days; on the other 38,200 they share a calendar day. Both services pass their own unit tests. Which action addresses the cause of the mismatch?
MCQ-8A team is finalizing acceptance criteria for `POST /payments`. The draft lists: a $10 payment with a valid card returns 201 and charges $10; a payment of $10,000 returns 201 and charges $10,000; a payment above $10,000 returns 422 and charges nothing; a payment with an expired card returns 402 and charges nothing. The service contract also sets a $0.50 minimum charge. A support ticket shows a $0.01 payment returned 201 and the card was charged $0.01. Which acceptance criterion does the support ticket show is missing?
MCQ-5A team ships a CSV bulk-import feature. The story's acceptance criteria read: "Add a streaming parser; create an ImportJob class; handle errors gracefully." All 42 unit tests pass and the story is marked done. Two weeks later a customer uploads a 2,000,000-row file with one malformed row at line 1,900,000; the import aborts after 40 minutes and commits nothing. The customer's contract requires that one malformed row must not discard the rest of the batch, and that the rejected row be reported with its line number. Which acceptance criterion would have forced the team to build and verify that behaviour before calling the story done?
MCQ-6A checkout API's release gate is an automated CI job: it runs a 10-minute load test at 300 rps, the load generator's sustained ceiling, against staging, which runs the same instance type as production. The metrics pipeline exports p50, p95 and p99 latency per endpoint; it exports neither mean nor max. The current criterion, "checkout must be fast under peak load", cannot pass or fail on its own. Which rewrite can this job evaluate?
MCQ-7A team's search requirement reads: 'p99 latency under 200 ms.' Staging and production run the same instance type and replay the same query mix. Staging passes: 25 rps over a 200k-document index gives p99 = 110 ms. Production runs 400 rps over a 9M-document index and reports p99 = 1.6 s. Which rewrite of the requirement would have made the staging result predictive of production?
MCQ-10A working group publishes a protocol spec with two requirements: 'Implementations MUST NOT reuse a nonce within a session' and 'Implementations SHOULD use TLS 1.3 or later.' A vendor ships a client that reuses nonces when its entropy pool is exhausted, and negotiates TLS 1.2 because its load balancer cannot do 1.3. The vendor asks to be listed as conformant. Which assessment is correct?
MCQ-20A three-step onboarding form was built from a clickable Figma prototype. In the prototype, each Back button is wired to the previous frame, so tapping Back on step 3 shows step 2 with its fields empty. Engineering reproduced that behavior exactly. After release, users who go back to fix a field lose everything they typed on the earlier steps. What did the prototype fail to specify?
MCQ-19A team's deletion job removes a user's rows from the `events` table. The next morning the same user ID is back in `events`. The pipeline rebuilds `events` nightly by replaying the Kafka topic `events-raw`, which retains messages for 30 days. The privacy requirement says only that user data must be "handled in line with applicable law." Which addition to the requirement would have prevented the reappearance?
MCQ-12A team's design review for a multi-tenant billing service records one design decision under each of ISO/IEC 25010's product quality characteristics: security: per-tenant key rotation; reliability: idempotent retries; maintainability: contract tests per tenant. Every characteristic gets an entry and the review is signed off. Which statement about the review is accurate?
MCQ-14A trading system ships a build where orders over $1M skip the mandatory second approval. The postmortem finds requirement TRD-9 ('orders over $1M require two approvers') traces to the regulation that mandates it, to the module that implements it, and to 3 tests: all passing. The tests assert against a mock that always returns 'approved', so they pass whether or not the threshold is enforced. The matrix reports TRD-9 as fully covered. What does the matrix fail to capture?
MCQ-22A booking service's acceptance suite passes: all 62 agreed scenarios green, 200 concurrent users, a 45-minute soak, and no seat sold twice when 50 clients request the same seat within one second. The suite runs on a single-node staging deployment. Two weeks after release on a four-node cluster, two clients 3 ms apart on different nodes both receive seat 14C. What does the passing run establish about that defect?
MCQ-9A checkout flow is up for a procurement review that asks for a WCAG 2.2 AA statement covering the whole flow. The evidence pack holds four things: a tracking sheet of the 47 applicable Level A and Level AA success criteria; a clean axe-core CI run across all 5 templates, whose enabled rules map to 18 of those rows; a keyboard-only pass of the cart page; and a 5-participant screen-reader session on the payment page that found 9 failures, including a dialog that opens without taking focus and an error summary that appears only visually. On this evidence, what should the QA lead do with the AA statement?
MCQ-13A payments API's contract states that POST /charges accepts an Idempotency-Key header, that the server stores each key with the first response it produced for 10 minutes, and that it returns that stored response for any repeat of the same key inside the window; after 10 minutes the key is evicted. A client's request times out and it retries the identical request with the same key at t = 0, 20 s, 3 min, 8 min, and 12 min. Every attempt that reaches the server and is not deduplicated creates one charge. How many charges does the ledger record for this order?
MCQ-11You own the 12-node analytics cluster that runs two nightly jobs. Search has filed a mandatory requirement that the index rebuild finish by 07:00 to hold its 07:00 SLA. Finance has filed a mandatory requirement that the revenue batch finish by 08:00 to produce the 09:00 report. Both name the same 02:00–08:00 window, and the batch reads the index the rebuild writes. Last month's runs: the rebuild's p95 is 2h50m on 8 nodes; the batch's p95 is 2h40m on 10 nodes. Neither team reports to you. What do you do?
MCQ-21A team is 10 days from a contractual go-live when a late change requires a new `region` column on a 240M-row `events` table in PostgreSQL 14. `region` is derived per row from the existing `country` column, and the settlement query needs it NOT NULL for every row at go-live. The team's migration tool rewrites a table at about 9,000 rows/sec, and the only maintenance window before go-live is 4 hours. What should the team do?
MCQ-1An ops team asks for a bulk re-route button, saying it will save two hours a day. Analytics show the re-route screen is opened 14 times a day, with a median session of 40 seconds. Sprint planning is tomorrow. What should the team do first?
MCQ-3A payments team is deciding whether to build saved payment methods. Evidence on hand: 340 Q3 support tickets mention re-entering card details, filed by 18 of the team's 1,200 active accounts; those 18 accounts carry 34% of ARR. A 500-response in-app survey, self-selected from a banner, put "save payment methods" in the top three for 22% of respondents. Two enterprise prospects asked for it on sales calls. Which claim about the need is the evidence strong enough to support?
MCQ-2A PM at a 400-account B2B analytics tool gets a request from the admin of its largest account (18% of ARR): 'Add a CSV export button to the dashboard.' Last quarter's support tickets: 11 mention exports, 9 from that same account, and 7 of those 9 say they already pull the data through the API but want it 'without writing code.' Telemetry shows the API export endpoint is called 40 times a day by 6 accounts. Which next step does the evidence support?
MCQ-24A team shipped saved report filters to all 12,000 accounts three weeks ago. Telemetry shows 1,800 accounts created at least one filter, 1,620 of those used a filter again in weeks 2 and 3, and 220 tickets were tagged "filter lost after logout" (previously zero). For the repeat users, median time from login to first report fell from 11 minutes to 2 minutes. The stated need was that analysts rebuild the same six filters every Monday. Which conclusion does this evidence support?
MCQ-17A team holds a 90-minute refinement session every week, but it walks the entire 120-item backlog in order, so the top items get a few minutes each. Sprint Planning still overruns by 40 minutes because the top items arrive as 13-pointers with no acceptance criteria. The team has no capacity for another meeting. Which change fixes the overrun?
MCQ-23A payments team's refinement sessions include the product manager, the tech lead, and two backend engineers. Over the last quarter, 26 of the 40 stories that passed refinement needed rework after release. The rework log groups the causes and the average fix time: | cause | stories | avg fix | | --- | --- | --- | | acceptance criteria omitted the data-retention period, failing privacy review | 13 | 1 day | | API contract changed after the iOS client had shipped against it | 6 | 4 days | | no rollback step in the on-call runbook | 4 | 2 days | | screen-reader flow broken | 3 | 3 days | Which single change to the refinement attendee list removes the largest share of this rework, measured in engineer-days?
MCQ-16A team closes 18 points per two-week iteration across three pairs. For the next iteration one pair is on support, leaving two pairs and about 12 points of capacity. 'Saved payment methods' is a 21-point story with three acceptance criteria: save a card, pay with a saved card, delete a saved card. A customer demo is booked for the end of the iteration, and the customer will only sign off on something they can use. Which split gives the team something usable at that demo?
MCQ-15A payments team has eight weeks to ship reconciliation. Finance's rounding rules will change roughly a third of the reconciled figures, and finance can only judge those rules against real numbers. The backlog is sliced by layer: ledger schema (3 weeks), ledger writes (2), reconciliation API (2), finance admin UI (1). Which plan gets finance judging real numbers earliest?
MCQ-18During design review, the team logs five assumptions. Each has an owner, a validation method, a decision date, an estimated probability of being wrong, and a cost if wrong (rework, delay, or lost revenue). The team can validate only one this sprint; validation cost is roughly equal for all five. Which assumption should the team validate first? | Assumption | P(wrong) | Cost if wrong | | --- | --- | --- | | Payment provider supports idempotency keys | 0.60 | $150k | | Vendor API p99 < 200 ms | 0.10 | $800k | | Users accept 2 s cold start | 0.90 | $60k | | Mobile app can reuse web session token | 0.40 | $200k | | Search index can rebuild in < 5 min | 0.30 | $250k |