Four altitudes. The one the evaluator is inspecting at decides what they hear, and what they hear is what your case is. Everything past this line is mechanics.
Nobody is rushing you. Atlassian schedules a 45-minute portfolio review. CZI budgets 35 minutes for two cases. You have the time. What you don't get is a second first impression. Reviewers form a working theory of what kind of designer you are early, then spend the rest of the session testing that theory rather than rebuilding it.
Calibrate on the sourcing before you act on it. High confidence that reviewers settle relevance, contribution, and craft level fast; Figma's recruiter puts it at minutes, sometimes seconds. Moderate confidence that the opening of a live walkthrough is what sets the theory they then test, because the published loops document what gets evaluated and not when the read locks. Speculative on any specific number attached to how early. Act on it anyway: your entry point writes their theory, and the entry point has to change with the room.
One case, four openings
Red Cross, opened four ways:
Consequence. "Designing for the worst day of someone's life." Then the constraint stack: untrained volunteers, bad networks, a federal audit trail, six systems that did not talk to each other.
Velocity. "Six months, zero to national deployment. 1,689 cases in the first two weeks."
Systems. "Six disconnected systems became one operational record, with compliance embedded in the workflow rather than bolted on after."
Making. "Let me open the intake queue. Dozens of real states in here, and I'll show you the one I'd change."
Same case, same URL, four different candidates walking out of those four sentences.
The sections run consequence, systems, velocity, making. That order is a ranking by how much your published pages already do the work for you, strongest first. Rehearse it backwards. You don't get to choose which room you walk into, so the preparation belongs where the page carries you least.
Reading the room in the first ninety seconds
You get about ninety seconds of pleasantries to identify the altitude. Three tells. In descending reliability: who is actually in the room, what they ask first, and what they never ask at all.
| Altitude | Who is in the room | Their first question |
|---|---|---|
| Systems | A design VP with long tenure at a large company | "Walk me through your process" |
| Velocity | A founder or product VP at a company under 200 people | "What's the hardest thing you've shipped, and how fast" |
| Consequence | Anyone from clinical, compliance, risk, or safety | "Tell me about a time the system got it wrong" |
| Making | A design engineer, or anyone who asked in advance for a link they could click | "Do you have anything I can play with" |
The design VP mapping almost never misses. The clinical or compliance seat outranks its placement on the panel; those people rarely lead the interview and frequently decide it. Moderate confidence on the first-question mapping. Pattern, not publication. I weight it above the job description because the first question is the least-rehearsed sentence an interviewer says, and because a posting is a high-reliability source for what a company says it wants and a poor one for what the panel actually checks.
Absence is the third tell and the one candidates never use. An evaluator who never raises business impact is not at systems altitude. One who never asks about your team is not evaluating you as a leader, whatever the title on the req says.
Mixed panels are the norm. Pitch the veto. At Director+ the veto usually sits with a peer rather than the hiring manager, because the hiring manager wants you to work out and your future peer is deciding whether you make their quarter harder. Moderate confidence, drawn from watching committees rather than from anything published. If you can't locate the veto inside five minutes, ask who else weighs in and what they care about. The answer names the altitude.
Consequence altitude in healthcare and regulated rooms
The scan. The posting says "experience in regulated environments." Set that aside. What this evaluator is doing is checking whether you have ever designed the state where the system was wrong. Not error handling as a visual treatment. The decision gate: who reviews, with what information, with what authority, and what the system does after they say no. If your portfolio lives on the happy path, everything else you show registers as decoration.
I laid out the anatomy of a real gate in The Crumple Zone Test (Company Battlecards): system state, reviewer time and competence, recognized authority, actual system response, protection from penalty. Use those five as the shape of the walkthrough.
Moderate confidence on this scan pattern. No regulated employer publishes a design rubric the way Atlassian and CZI do. It's inferred from role language and from what these organizations answer for when something goes wrong, which is weaker than a published guide and considerably stronger than instinct.
The entry.
Wrong: "Thermo Fisher's mySupply delivered over twenty million in annual margin at an 83% IRR."
True, published, and exactly wrong for this room. You've opened with the business case to the one person whose entire job is the failure boundary.
Right: "Batch exceptions were surfacing two to four days after they happened. By then rescheduling cost three to five times more. So the first thing I designed wasn't a dashboard. It was the order table, sorted so the exception is the first row on the screen."
Both numbers already sit on your Thermo Fisher page, down in the problem section rather than the hero. Promote them when you speak.
Foreground every state where the system is wrong or uncertain. Red Cross gives you the deepest inventory: duplicate-record warning, FEMA address verification, failed identity checks, escalation, fraud alerts, audit export, supervisor batch approval. Thermo Fisher gives you missing documents, forecast drift, capacity conflict, exception routing, and the human QA release control. Land on the human gate itself, not somewhere adjacent to it.
Compress the dashboards. They are the most attractive artifacts on both pages and the least interesting here. Cut the Thermo Fisher module tour from five modules to two and spend what you recover on what happens between exception detection and human release. The agentic-future sections collapse to one sentence: the release gate stayed human.
Close on consequence avoided, not margin. The number this evaluator retains is what stopped going wrong. Your pages give you the before-state cost rather than an after-state measurement, so say that plainly: exceptions arrived two to four days late, rescheduling ran three to five times more, and the design moved detection to the top of the queue. Naming the gap in your own evidence costs you nothing in this room and buys more than a number you can't source.
Do not force Allē here. Aesthetic medicine looks like regulated proof and isn't. The published page shows patient context, treatment evidence, and provider workflow, with no privacy controls, no clinical approval path, and nothing on error and recovery. Offering it invites a question you cannot answer from published work. Alibaba carries no failure state on the page at all. If either surfaces, say what is and isn't published, then move.
Systems altitude and the enterprise evaluator
The scan. This person has sat through a great many of these and runs one reflex above all others: subtraction. While you talk, they are privately discounting your numbers by the company momentum they estimate was already moving without you. Atlassian states the underlying logic almost mechanically, with individual contributors getting a deep dive into detailed choices and management candidates getting questioned on how they led the team and shaped the result. CZI's interview guide names four signals — Scope and Impact, Design Excellence, Leadership and Ownership, Collaboration — and asks candidates to divide their own work from their team's. High confidence. Two major employers put it in writing.
One question sits under all of it: were you in the room where the trade-off was made, or downstream of it.
The entry.
Wrong: "Alibaba.com does over fifty billion in GMV, and the redesign lifted daily transactions twenty percent."
That hands them the objection before you've earned anything. On a marketplace that size a twenty percent lift raises the attribution question instantly, and you spend ten minutes defending instead of demonstrating.
Right: "Twenty-five percent of the traffic was desktop. Eighty percent of the transaction value was desktop. Nobody in the org had put those two numbers next to each other. Until I did, there was no mandate to fund this work. So let me walk you through how I built the mandate, then the three things we shipped with it."
That is the actual opening of your Alibaba page, spoken. The page leads with the discrepancy, not the money. Say it the way the page says it.
Foreground the diagnostic: 32 interviews, the audit, the gaze-tracking, and the three named failures they produced. Then present the three sprints as three bets you placed, each carrying its own metric. Sprint-level numbers survive subtraction because each one attaches to a decision you can name. Hero numbers attach to a company.
Foreground coherence, which is what this altitude actually means by "systems." Two cases carry it. Equinox+ unified five distinct brands on shared token architecture without flattening any of them, a coherence argument here and a speed argument one altitude down. Allē is your second coherence case: a two-track architecture serving a consumer app and a provider backoffice with opposed incentives, held together by one design language. Present Allē in this room instead of a regulated one and it converts from liability to asset.
Compress the hero metric strip and let them read it. Compress research methodology; Figma's recruiter is explicit that reviewers want the solution path, the iterations, your role, and the learning, not the step-by-step. Compress any "if I were building this today" section to thirty seconds, because an enterprise VP files that as unbuilt.
Build the walkthrough so that "what did you decide there?" never has to be asked. Attach a decision to every screen in the same breath as the screen. "This is the search card. I decided supplier verification outranks price, because the disengagement data said trust failed before price mattered." A slide with no decision on it is a slide to cut.
Velocity altitude at growth stage
The scan. The posting says "thrives in ambiguity," which is an empty container. The founder is doing arithmetic: decisions divided by months. They want proof you ship under constraint and that your presence multiplies a small team rather than requiring a large one. Flipkart's published design guide scores this directly, naming quality and speed of delivery as a dimension and warning against many shallow projects. Moderate confidence as a general pattern. High for that specific rubric, which one employer publishes and the rest do not.
The entry.
Wrong: "Equinox+ unified five premium fitness brands into a single platform serving six hundred thousand members."
Accurate. Reads as scope. Scope reads as slow.
Right: "Three months, zero to MVP, five brands, one platform. Six weeks out I made the launch prioritization call and cut scope, because the date was real and the scope wasn't. Do you want the ship list or the cut list first?"
The three-month timeline and the six-week prioritization decision are both on the page. The cut list is yours to reconstruct out loud, so rehearse the specific item you dropped and the reason. Offering it is the move. Shipping is a subtraction discipline, and almost no candidate volunteers evidence of understanding that.
Foreground clocks and team sizes. Twelve months from zero to live across nine sites on a small delivery team. Six months to national deployment at Red Cross, 1,689 cases in the first two weeks. Three months to MVP at Equinox+. Then the trade-off you named, per case, with the reason attached.
This is the one altitude where the Agentic Labs do unambiguous work. Brand Pulse, Retail Velocity, and Carrier IQ are operable applications, built solo, live and clickable. As proof you can get from idea to working software without a team, they are strong.
Compress research depth. Four personas is three too many; name the one that changed the architecture and move. Compress the design system explorations. Compress anything that reads as deliberation.
Put the clock in the first sentence of every case and "how long did that take?" never gets asked, which beats answering it well.
Six months to national deployment is impressive to a growth-stage VP and mildly alarming to a regulated evaluator, who hears speed as risk. Same fact, opposite reading. Know which room you are in before you reach for it.
Making altitude at AI-native companies
The scan. Where the product is the model, this evaluator is hunting the seam between describing making and doing it. Spotify's published guidance asks candidates to take reviewers on a journey "starring you as the main character," and treats unfinished working artifacts as inspectable signal. Etsy explicitly invites prototypes and live builds alongside finished screens. AI-native staff design postings routinely ask for pixel craft, microinteractions, motion, animated prototypes, and zero-to-one work carried from concept to launch with a visual language built from scratch.
Moderate confidence overall. These employers publish what they want to see. None publish how it's weighted. Anyone who claims to have the rubric is guessing.
The entry.
Wrong: "Carrier IQ is an agentic lab I built that automates insurance quoting across eight carrier portals in about four minutes."
A product pitch to an audience that is going to open the thing anyway.
Right: "Let me just open it." Then, while the intake form loads: "Person, vehicle, address, history, coverage, carrier selection. I'll fill it in while I talk. There's one decision on this screen I'd defend hardest and one I'd change, and I'd rather you see both."
Handing over the artifact inside fifteen seconds is the strongest move available to you at this altitude. It converts you from someone describing making into someone making.
Foreground the intake surface, the most interrogable input design you have published. Then confidence scoring and broker comparison. Then the parallel-workflow model, framed as an interaction decision rather than an engineering one. Use the Watch–Verify–Delegate ladder to explain why the confidence score sits where it sits, not as a standalone framework.
Compress the automation outcome claim; no published adoption or accuracy data stands behind it. Compress the essay to its ladder and its two failure anecdotes: the wrong control on the wrong row, the model that changed silently.
The gap. The Lab pages expose working applications. The material behind them stays unpublished: prompts, evals, traces, version history, the versions you killed. In The Buyer-Angle Map (Portfolio Playbook) I told you to bring one Lab component across every state plus what you killed and why. Those artifacts don't exist publicly. So either publish an iteration history on one Lab, or present the Labs strictly as operable software you shipped alone. The essay proves judgment without proving execution, and at this altitude judgment alone reads as commentary.
One substitution while the gap stands. Your most inspectable craft sits outside the Labs, in the coded interface states inside Thermo Fisher, Red Cross, and Allē: exception rows, queues, ledgers, approval flows, live prototypes, dozens of real product surfaces. The Lab wrappers are short and defer everything to an embedded app. If an AI-native evaluator wants interaction judgment across many states, open Red Cross.
Have all three apps loaded in tabs and rehearse against the live versions that morning. Live software drifts. If a state is broken, name it before they find it, because discovered breakage costs credibility and disclosed breakage costs nothing.
Seams to close before you present next
| Case | The conflict | What to do |
|---|---|---|
| Allē | 18 million and 30M+ members both appear on the same page | Pick one. Never restate it in the room. |
| Carrier IQ | The page says eight portals and four minutes; other surfaces of yours have used five carriers and ninety seconds | Quote the page or fix the page. Do not improvise the count out loud. |
| Thermo Fisher | Margin and IRR sit in the hero; the numbers you actually need at consequence altitude, exception latency and rescheduling cost, sit down in the problem section | Fine for scanning. Know where they are before you speak. |
The rehearsal
Say each entry line out loud. Four times, once per altitude, before any of it gets said in a room. Reading them silently doesn't count.
Every wrong version above is what you'd reach for under pressure, because it's the version your own page leads with. Narrating a portfolio top to bottom is the default failure mode, and it's a failure mode precisely because the pages are good. They were built to be scanned in silence. Nobody in that room is scanning in silence.
Your pages lead with metrics. Your mouth mostly shouldn't.
- Read the recruiter's own scan: Figma's Korin Harris says senior reviewers may have only minutes, or even seconds, to parse type of work, relevance, and approach, and that excessive process detail makes their eyes glaze over, which is worth reading before you rebuild any case narration.
- When a Head role is not executive: Amplitude's Head of Product Design posting states outright that this is not a role that manages up and delegates down, asking instead for systems-level direction and pixel-level craft across a 15-person team, which is the clearest published example of title diverging from altitude.
- Explanations can make reliance worse: a CHI experiment with 199 participants found cognitive-forcing interventions reduced overreliance on AI more effectively than standard explainability, though participants rated the interventions that helped them most least favorably, which is a sharper story for a consequence-altitude room than "we added a confidence score."
- What a strong regulated mandate looks like in writing: Maven Clinic's VP of Design posting publishes the CPO reporting line, executive access, a roughly 15-person multidisciplinary team, and accountability for clinical outcomes alongside growth metrics, making it a useful benchmark for judging vaguer healthcare postings.