Describe pair-programming challenge under time pressure

# How interviewers evaluate this - Collaboration: Do you align quickly, communicate clearly, and avoid blame? - Decision-making under pressure: Do you timebox, choose pragmatic tradeoffs, and document decisions? - Learning agility: How do you handle unfamiliar tools (e.g., curl/PostgreSQL differences)? - Ownership and outcomes: Do you quantify impact and land the project despite detours? # Structure your answer (STAR-L) - Situation: Brief context, scope, and time pressure. - Task: Your responsibilities and the goal. - Action: Communication, work split, and decisions (timeboxes, fallbacks, tradeoffs). - Result: Concrete outcome and metrics (time lost/recovered, delivery status). - Learning: What you’d change next time (process, tooling, guardrails). # Example answer (adapt with your details) Situation: My partner and I had 4 hours to build a small REST endpoint to ingest events, store them in PostgreSQL, and expose a read endpoint. We agreed to pair for design and split for implementation. Task: I owned the DB schema/migrations and idempotent upsert logic; my partner owned the HTTP handler and a script to exercise the API. Actions: - Kickoff alignment (15 minutes): We wrote a one-page “Definition of Done” with 3 acceptance examples (happy path, duplicate event, invalid payload), success criteria (all examples pass), and explicit out-of-scope items. - Work split: We used a driver/navigator rotation for the first hour to agree on the skeleton, then split tracks with a Slack channel for decisions and a 20-minute timebox per unknown. - Handling unknowns and delays: 1) curl quoting (lost ~25 minutes): Posting nested JSON via curl kept returning 400 errors due to shell quoting differences. We switched to sending from a file (curl -H "Content-Type: application/json" --data @payload.json http://localhost:3000/events) and added a Make target (make send-event). This removed quoting risk and sped iteration. 2) PostgreSQL mismatch (lost ~30 minutes): My local container lacked the uuid-ossp extension required for uuid_generate_v4(), and CREATE INDEX CONCURRENTLY failed because our migration tool wrapped it in a transaction. Under time pressure, we made two decisions: generate UUIDs in application code (library call) and create a normal index (non-concurrent) since the dataset was tiny. We documented these tradeoffs in the README. 3) Requirement ambiguity (~10 minutes): The prompt was vague about idempotency scope. We timeboxed, picked a pragmatic rule (idempotency by event_id within 24 hours), and wrote a test to encode that decision. - Communication: We narrated thinking while pairing, posted a running decision log, and used short syncs every 30–45 minutes to re-plan. Result: Despite ~65 minutes of delays, we delivered the MVP in 3 hours 45 minutes with 12 tests, a clear README, and a scriptable way to exercise the API. The endpoint handled duplicates correctly, and our decisions were explicit and reversible. Learning and what I’d do differently: - Pre-flight guardrails: Pin tool versions in docker-compose, include init SQL for required extensions, and provide a test harness (curl/httpie script) upfront to avoid manual quoting issues. - Timebox/decision log: Keep 15–20 minute timeboxes on unknowns and maintain a visible decision log so we can revisit tradeoffs after the deadline. - Example-driven alignment: Start with 3–5 acceptance examples to eliminate ambiguity and keep both partners focused on outcomes. - Pairing protocol: Use explicit driver/navigator roles early, then split with clear interfaces and integration checkpoints. # Checklist you can use - Before coding: Align on acceptance examples, timebox length, and a decision log. - Environment: Pin DB versions; install needed extensions; standardize on one HTTP client (curl/httpie) with file-based payloads. - During execution: Timebox unknowns; prefer reversible, low-risk decisions; document tradeoffs. - After: Quantify impact and outcomes; reflect on process improvements you’ll adopt next time. # Pitfalls to avoid - Vague stories without measurable impact. - Blaming your partner or the tools instead of owning the process. - Skipping what you’d change next time (interviewers look for self-correction). # Mini-template (fill in) - Situation: [timebox, scope] - Ambiguity/unfamiliar tools: [requirement gap, curl issue, PostgreSQL version/extension] - Actions: [alignment, split, timeboxes, fallback decisions] - Result: [delivery status, tests, time lost/recovered] - Learning: [process/tooling guardrails you’ll implement next time]