Skip to content
Tech Interview Prep
Technical interview guide

Developer Experience Metrics

Measuring whether a platform is actually working: lead time, deployment frequency, and developer satisfaction.

Read
40 min
Practice MCQs
25
Interview Q&A
25
Edition
v3
Editorial status
Reviewed
Relevant for
Platform Engineer

Scope: DORA software delivery metrics and measurement guidance, SPACE, DevEx, AWS, Google Cloud, and Microsoft platform engineering guidance current 2026-08-31.

Curated: · Written: · Reviewed:

25 questions

MCQ-3A platform group's "commits per engineer" dashboard was gamed within a quarter: commits split, reviews rubber-stamped, estimates padded. A staff engineer proposes the SPACE framework, which requires measuring several independent dimensions at once because any single productivity signal can be optimized directly. Which set of dimensions does SPACE prescribe?
MCQ-6A team's deploys are frequent, but when a bad release reaches production the service stays degraded for hours. Leadership wants a DORA metric that captures how long the team is down after a bad release. Which metric should they add to the dashboard?
MCQ-14A load test sends 1,000 requests to a checkout service: 950 complete in 200 ms, 49 in 400 ms and one in 6,000 ms. The dashboard shows p50 = 200 ms and p99 = 400 ms (nearest-rank). Which statement is supported by this run?
MCQ-8A platform team wants a single deployment event stream that feeds both deployment frequency and change failure rate for a named service. Their CI runs on every merge to main, about 40 times a day; the CD pipeline promotes to staging automatically and to production after approval, about 3 times a day. Which definition should they record as a deployment?
MCQ-19Your team governs a 'checkout conversion' metric. Its spec reads: numerator = orders completed, excluding internal accounts and fraud-flagged traffic; denominator = checkout sessions started, excluding internal accounts and fraud-flagged traffic; window = calendar week, Monday 00:00 to Sunday 23:59 UTC. For the week of 3–9 March the session table holds 62,480 rows: 1,340 from internal accounts, 890 flagged as fraud, 2,000 from a staging environment, the rest external. 3,150 orders completed in the window, of which 120 came from internal accounts and 60 from fraud-flagged sessions. Two dashboards report different values for the week. Which value should the governed metric report?
MCQ-22A product team replaces the definition of 'weekly active users': the old definition counted any session in the trailing 30 days and last week reported 1,240,000; the new one counts at least one core action in the trailing 7 days and this week reports 470,000. The ETL job was swapped on Monday, so the chart now shows a 62% week-over-week drop. The metric also feeds a monthly board report whose last six months were published under the old definition. What should the team do?
MCQ-5A platform team is building a DORA dashboard for 400 engineers and 60 services. Services change owning teams every few quarters, some services are split across two repositories, and one repository hosts three services. Leadership wants to see whether a given service's delivery performance improved after last year's CI migration. What should each metric be scoped to?
MCQ-7A team's change failure rate is low, but a quarterly review finds that a large share of its deployments were follow-up releases fixing something a previous deployment broke, and because those fixes shipped successfully they never counted as failures. Which metric should the team add to its dashboard to make this visible?
MCQ-23Your payments team's lead time for changes, first commit to running in production, has gone from 1.9 days to 2.8 days over two quarters. Last sprint's 40 merged changes averaged 5 hours of coding before the pull request opened, 50 hours waiting for a first review, 5 hours from first review to merge, and 8 hours from merge to production. Three engineers cover every review on top of their own sprint work, and roughly 12 pull requests sit open and unreviewed at any moment. Which change is most likely to bring lead time back toward 1.9 days?
MCQ-2A platform team of 8 engineers tracks commits per engineer per week on its dashboard. Last quarter: 12 commits/engineer/week, 18 merged PRs/week, median PR diff 210 lines, 34 story points completed/week. This quarter: 21 commits/engineer/week, 17 merged PRs/week, median PR diff 85 lines, 34 story points/week, and CI pipeline runs up 70%. Which conclusion does the data support?
MCQ-1At a 250-engineer company, the last DevEx dashboard ranked teams by deploys per engineer. The payments team (40 engineers) came last, but its engineers spend 35% of their time waiting on a 22-minute build. Leadership will decide next month whether to fund a build cache for that team. What should the DevEx team do first to make the next dashboard drive that decision?
MCQ-4A platform team's delivery metrics look healthy, but engineers report that code review stalls for days, the service catalog is too large to navigate from memory, and they rarely get an uninterrupted block to build. Which set of dimensions should the team instrument to capture these three developer-side factors?
MCQ-13A platform team tracks onboarding with three medians: days to first deploy (TTFD), days to first independent production change (no buddy), and days to first solo on-call shift. Last quarter's cohort: 2, 34, 61. The prior quarter's cohort: 9, 28, 55. The director sees TTFD fell from 9 to 2 and calls onboarding a win. Which reading of the three medians is correct?
MCQ-11A platform team reports 92% adoption: 368 of 400 backend engineers created at least one service from the golden-path template this quarter. Deploy telemetry for the same quarter shows 12,400 production deploys, of which 2,900 went through the platform's deploy pipeline. 291 of the 368 adopters created their service during a two-week migration hackathon and their repos have had no commits in the six weeks since. Which reading of the adoption number is correct?
MCQ-12An internal developer platform serves 118 services across 40 teams, with #platform-help as the only support channel. Tickets filed fell from 1,180 in Q2 to 690 in Q3, tagged as: 'env provisioning' 520 to 40, 'pipeline failures' 300 to 295, 'access and permissions' 210 to 205, 'deploy config' 150 to 150. In the same window, paved-road deploy share went from 84% to 87% and the number of services deploying at least once a week went from 112 to 115. The only platform change in Q3 was a self-service environment-provisioning flow shipped on day 2, and median time from new-service creation to first production deploy fell from 3.2 days to 1.1 days. Which reading of the ticket drop do the numbers support?
MCQ-15A dashboard ranks services by 1-hour error rate. Service A: 3 errors in 250 requests (1.2%). Service B: 1,200 errors in 250,000 requests (0.48%). A sits at the top of the list. Which of these is the defensible reading?
MCQ-16A platform org publishes a monthly league table of change failure rate (CFR) across 30 teams. Team Atlas: 40 deploys/week, CFR 15%, and every failed deploy auto-rolls back in a median 6 minutes. Team Borealis: 1 deploy/month, CFR 0% across the last 12 months, and its rollback is a manual restore that takes about 4 hours. A director reads the table and orders Atlas to freeze deploys until its CFR is under 2%. What does the table actually support?
MCQ-10A platform team replaced a manual review gate with auto-approval for requests that match the hardened template. Median provisioning time fell from 3 days to 25 minutes, and weekly volume rose from 60 to 210 environments. The team's counter-metric: the share of newly provisioned environments that fail the nightly policy scan within 24 hours of creation: moved from 3% to 39%. The team lead calls the result acceptable because 39% is still under half. Which reading of the counter-metric is correct?
MCQ-9A payments team watches p99 latency for the /charge endpoint climb from 240 ms to 2.1 s over a week. Telemetry shows the tail is confined to requests carrying a `retry=true` header (6% of traffic); server-side handler time for those requests is 55 ms, and the remaining ~2 s sits in the client's time-to-first-byte span. No deploy shipped in that window. The team interviews the two client teams that set `retry=true`: both retry once after a 2 s timeout, and both added that retry last month to work around intermittent 503s from a shared connection pool. What does the interview change about the diagnosis?
MCQ-17A platform team rolls out a new build cache to 40 of its 120 teams in week 1; the other 80 teams get it in week 4. In the four weeks before the rollout, the 40 treated teams averaged 500 build-minutes per team per week and the 80 control teams averaged 480. In the four weeks after the treated teams got the cache, and before the control teams did, the treated teams averaged 350 and the control teams averaged 420. The lead wants the best estimate of the cache's effect on build minutes per team per week. Which figure belongs in the report?
MCQ-18A platform team compares 30 teams that adopted its new CI/CD platform with 30 that did not. Adopters show a median lead time of 3.2 days against 6.4 days for non-adopters. Before adoption, 24 of the 30 adopters were already in the top quartile for deploy frequency, against 6 of the 30 non-adopters. Within each pre-adoption deploy-frequency quartile the lead-time gap is 0.3 days. What does this comparison support?
MCQ-21Your platform team surveys all 5,000 engineers on whether the deploy-tooling migration is on track. Of the 1,000 still on the legacy tooling, 600 respond and 80% of them say it is on track. Of the 4,000 on the new tooling, 800 respond and 40% say it is on track. The draft slide reads "57% of engineers say the migration is on track." What should the team report?
MCQ-25A 400-engineer org runs a quarterly developer-experience survey. Q1 drew 244 responses and 9 recurring themes; the platform team shipped fixes for 3 of them and announced the changelog on the internal blog, which analytics show 40 people opened. Q2 drew 232 responses. For Q3 the team added 12 questions and a monthly pulse survey; responses fell to 88 and the top write-in is "nothing ever changes." Which change is most likely to stop that write-in in Q4?
MCQ-20A weekly engineering-metrics table stores one row per engineer per week with a `reviews_completed` count. The dashboard publishes each team's mean `reviews_completed` for the week. The Mobile team has three engineers, and last week's published mean was 12.0. The Mobile manager knows that engineer A completed 4 reviews and engineer B completed 20. The privacy review requires that no published figure let a manager recover an individual's value. Which change to the dashboard meets that requirement?
MCQ-24Once a year the platform team retires at most one series from its 30 largest, enough to keep the observability bill flat. Five candidates are on this year's list. | series | active series | purpose, as written in the runbook | what the last 90 days show | | --- | --- | --- | --- | | `api_5xx_rate` | 128,000 | backs the API availability SLO | 11 pages, every one closed by a rollback or a config fix | | `checkout_p99_latency` | 61,000 | backs the checkout latency SLO | 9 pages, all closed by a config or capacity change; 2 dashboard viewers | | `payment_recon_mismatch_count` | 6,000 | reconciles the ledger at the monthly close | read by the finance close script on the 1st of the month; no alert rule and no panel attached | | `kv_cache_hit_ratio` | 44,000 | added to answer whether a 30-minute TTL holds the cache hit ratio above 0.90 before the TTL change went to all regions | flat at 0.94 to 0.945 since the TTL change shipped 11 months ago; its four rules at 0.90 have not fired; the engineer who created it left 9 months ago and the March review that read it was cancelled | | `codec_memory_overhead` | 19,000 | checks the new codec against the 512 MB per-connection plan | reads 22% over plan; the migration is paused at 40% and its lead reads this series daily | Which one should be retired this year?