Capstone: The Mock Loop

The research-challenge round is the only part of the loop where the panel watches you think, not just hear you narrate the past. Treat it as a live audition for the job, not a quiz.

Cursor runs a paid, project-heavy onsite that mirrors the engineering loop. You will get a prompt about the developer and the product, then you design a study and defend the method choices to a panel of researchers. The bar is senior-IC: a plan that ships insight fast, holds up under pushback and connects to a roadmap decision someone actually has to make.

A worked promptuse this as your rehearsal target

Work the same prompt every time you practice so you can feel yourself getting faster: Cursor wants to know whether developers trust the Agent to make multi-file changes unsupervised - design the research. It is a real Cursor tension, it mixes attitude with behavior and it forces a method choice between what people say and what they do.

1. 1Frame the decision. Trust here is not one thing. Split it into the decision it drives: should the Agent default to auto-applying multi-file diffs or keep a review gate? Name that the moment you start.
2. 2Sharpen the question. Turn the vague "do they trust it" into answerable sub-questions: where does trust break (which file types, which blast radiusHow much breaks if a change goes wrong; the scope of potential damage.), what recovers it (diff preview, undo, tests passing) and whether stated trust matches actual accept/revert behavior.
3. 3Pick the method for the question. This is generative-into-evaluative. You need behavioral signal on what developers already do, plus moderated sessions to explain why the revert happened.
4. 4Define participants and sampling. Segment by stakes, not by demographics: solo side-project users versus engineers shipping to a shared production repo. Trust thresholds differ enormously across that line.
5. 5State analysis up front. Decide before fieldwork how you will code the qual and which behavioral metric (revert rate on multi-file Agent edits) anchors the quant. Pre-committing stops post-hoc storytelling.
6. 6Map findings to decisions. For every plausible result, write the product move it triggers. A plan that cannot change a decision is a literature review.

The 5–8 slide planwhat the panel expects on the page

Whether it is a whiteboard or a take-home deck, the skeleton is the same. Keep it to one idea per slide and lead with the question, not the method.

Two versions, alwaysthe speed read is part of the test

Cursor cares about running good-enough research fast. Show that you can dial rigor up and down on purpose by presenting the same question at two scopes.

Pre-write the defensethe panel will push; plan for it

Every method choice is a tradeoff and the researchers on the panel will probe each one. The classic gut-check is "do this in half the time" - they want to see you cut scope without losing the spine of the study.

• "Why not just a survey?" Surveys capture stated trust, not the moment someone reverts a diff. You need behavior plus the explanation behind it and a survey gives you neither cleanly.
• "Why moderated, not unmoderated?" Trust failures are emotional and contextual; you need to probe the hesitation in real time. Unmoderated would miss the pause before someone clicks reject.
• "Half the time - what goes?" Drop the survey, keep telemetry plus five sessions. You lose the ability to size the problem, but you keep the directional answer and the mechanism.
• "What if your sample is biased toward power users?" Acknowledge it, then show the mitigation: deliberately recruit from the at-risk segment (shared-repo engineers) and flag the readout's generalizability limit.

The portfolio deep-dive is where most senior candidates quietly lose the loop. They give a tour of how clever the study was and never say what changed because of it.

Cursor's deep-dive runs about an hour on one or two case studies, with probing on tradeoffs, pushback and what you would do faster. Pick a single mixed-methods case you can run cold in under ten minutes, then leave the rest of the hour for the panel to dig. The case has to show qual and quant working together, because that is the practice they are hiring for.

The PAPAIL spineProblem · Approach · Process · Artifacts · Impact · Learnings

1. 1Problem. One sentence on the decision at stake and why it was hard. Not the topic - the tension. "The team disagreed on whether to invest in X and we had no data."
2. 2Approach. The method mix and why this mix for this question. Name the one tradeoff you made consciously.
3. 3Process. Compress hard. The panel does not need your screener; they need the two or three moments where your judgment mattered.
4. 4Artifacts. Show exactly one - the synthesis that drove the decision, not your raw notes. It should be readable on screen in five seconds.
5. 5Impact. Lead with this, even though it comes late in the acronym. What decision changed, what shipped or got killed, what the measurable effect was.
6. 6Learnings. One honest thing you would do differently, framed as a faster or sharper version, not a confession.

Pre-load the three probesthey will come; have crisp answers ready

One artifact, one vivid momentthe two things people actually remember

An hour of talk decays in memory to two images. Choose them on purpose: a single synthesis artifact that shows your thinking and one verbatim user moment that makes the finding undeniable.

The craft deep-dive moves fast and probes your reflexes. They are checking whether your method instincts are sound when you don't have time to hedge.

This round tests mixed-methods range: qual rigor, survey and sampling design, experiment and stats literacy and how you reconcile signals that disagree. The senior tell is that you name each method's blind spot as you reach for it. Confidence plus self-awareness reads better than confidence alone.

When the obvious method is wrongknowing what NOT to run is the senior signal

Sizing a study, fasthave a defensible rule of thumb ready

For qual, you size to saturation, not to a magic number. For quant, you size to power. Be able to say both cleanly and to flex the number to the decision's stakes.

• Qual: roughly 5 sessions per distinct segment to surface major themes, extending until new sessions stop producing new codes (saturation). A higher-stakes decision buys more sessions, not a different rule.
• Survey / quant: size to detect the smallest effect that would change the decision (the minimum detectable effect) at a fixed confidence and power, then report the precision you actually achieved, not just the point estimate.
• The honest move: when the recruit is hard or the clock is short, run fewer and flag the confidence level rather than pretending small-n is conclusive.

When qual and quant disagreethe reconciliation question they love

Conflicting signals are a feature, not a failure. The disagreement usually means each method is measuring a slightly different thing and resolving it is where the real insight lives.

1. 1Check they're measuring the same thing. Telemetry might show high feature usage while interviews report frustration - both can be true if usage is forced, not loved.
2. 2Trust behavior for what, qual for why.* Logs tell you what happened at scale; sessions tell you the mechanism. Let each answer the question it's good at.
3. 3Look for the segment hiding the average. A flat aggregate metric often masks one delighted segment and one struggling one that the qual surfaced.
4. 4Resolve, then re-collect if needed. If the conflict is decision-critical and unresolved, name the targeted follow-up that would break the tie rather than averaging the two into mush.

Spot-the-bias on demandthey will hand you a flawed question and a flawed test

Triangulate one Cursor question three waysthe design-a-mixed-method drill

A common live task: take one question and design three complementary methods, each covering another's blind spot. Use the same trust prompt from section one.

The values round reads for authenticity above all. Cursor is checking whether you actually use the tool and care about the developer or whether you're a strong generalist UXR reciting transferable answers.

Expect truth-seeking, ownership, pace and "why this team," sometimes folded into the onsite or a founder conversation for senior roles. Keep every story under two minutes and end each one on the outcome. The fastest way to fail this round is a polished answer that could have been told by someone who has never opened Cursor.

Four behavioral prompts, four themesrehearse one tight story per theme

The 2-minute why-Cursor narrativewhy this product, why this user, why now

1. 1Why this product. Be specific and current: you use Cursor, here's the workflow it changed for you, here's what's still rough. Generic "AI is exciting" gets discounted instantly.
2. 2Why this user. Professional developers are a sophisticated, opinionated, technical user - say why researching an expert user energizes you and where it's harder than researching a consumer.
3. 3Why this team. A small, flat, talent-dense research function where you'd be roadmap-driving from day one. Name that you want the scope and the speed, not a big-co research org.

One Cursor friction you'd researchthis is the authenticity proof

Have a real friction you've personally hit, stated as a researchable hypothesis with a method. This single answer does more to prove you use Cursor than anything else in the loop.

Three senior questions backasked questions reveal level

• "What research has actually changed a roadmap decision here in the last quarter and what made it land?"
• "Where does the team currently trust intuition over research and is that the right call?"
• "As the research function grows, what's the first ritual or piece of infrastructure you'd want built - and would I own that?"

A mock loop you don't score is just rehearsal theater. The point is to find your weakest dimension while it's still cheap to fix.

Treat the rubric as a gap finder, not a score to admire. Mark the weakest proof, turn it into one practice rep and keep the artifact you would show an interviewer.

Run all four mock rounds back to back, then grade each one against the same five dimensions a Cursor panel weighs. Score honestly - generous self-scoring is how people walk into the real loop blind. Anything under 4 is a gap and the plan below tells you which module to revisit for it.

The five-dimension rubricscore every round 1–5 on each

Common failure modes and their fixdiagnose your low scores fast

The one-week gap planconcrete, daily, re-testable

1. 1Daily Cursor use. Use it as your real editor every day this week and log frictions as you hit them. By Friday you have a stack of authentic, specific material for the values round.
2. 2One re-recorded case study. Re-record your portfolio walkthrough opening with impact, cut to under ten minutes and re-watch with a timer until no sentence is padding.
3. 3One quant refresher. Drill sizing-to-MDE, the qual/quant reconciliation move and spot-the-bias on a survey and an experiment until they're reflexive.
4. 4Revisit your weakest module. Map your lowest rubric dimension to the track module that covers it and rework that one before re-running the mock.