Three of four checkpoints fail. The one that passes barely does. Twenty minutes on the H1 buys you more than everything else on this list combined.
Juno. I pulled the live teaser this morning before writing a word, which is the rule for these. Nothing has shipped since the July 24 audit. Last-modified still reads July 19. File size identical to the byte. Every string I flagged in July is sitting where I left it. Delta of zero, and four recommendations stay open: the H1, the per-sprint annotations, the component rack, the CTA wording.
The eight-second read on what's live: the buyer gets a mood, then a right rail that is, credit where it's owed, doing real work. Nothing in that order tells them you generated this project rather than got handed it. The rail rescues you if the reader is patient. Plan for the ones who aren't.
Eight seconds is my own working model out of search-committee work, not a published finding. I went looking for a credible one and there isn't one. Soft on the duration. Hard on the ordering.
| Checkpoint | Verdict | Fix |
|---|---|---|
| 1. Title | Pass, downgraded | Optional, do last |
| 2. First screen | Fail | 20 minutes |
| 3. Craft signal | Fail | A weekend |
| 4. Call to action | Fail on accuracy | 1 hour |
The only question this buyer is asking
Your buyer for this case is a VP Design or a CPO at an enterprise platform company. Stripe. Salesforce. Atlassian. Brex. People running large design orgs where the unit of work is an end-to-end system and a single screen is the smallest thing on the table.
Every finalist who reaches their shortlist can produce the history line. Led three cross-functional sprints across homepage, search, and product detail pages. All true, and inert. It reports what your hands did and stays silent on whether any of it happens without you in the building.
I've briefed candidates through enough Director-plus searches to say this without softening it. The panel scores whether the work exists in your absence.
Which is why this audit is short. You already answer that. The sentence is written, in your voice, sitting in your intro paragraph:
"Nobody had named that gap yet. I named it, built the research case, secured the mandate, and redesigned it."
The whole case, compressed, pitched at exactly the altitude the buyer is scanning for. It is currently fourth in the hero. Behind the eyebrow. Behind a headline that withholds its subject for atmosphere. Behind four sentences of desktop-traffic diagnostic. Three eye movements before anyone meets the only line establishing authorship.
The content is there. The order is wrong, and fixing it is mostly cut and paste.
Checkpoint 1 — Title Reopened
Current: Alibaba.com — Redesigning B2B Trust at $50B+ GMV — Juno Chen
Verdict: Pass, downgraded. I closed this in July. Graded against the mandate test it's a B-plus, because redesigning is an execution verb and execution verbs invite the wrong inference.
Optional replacement:
Alibaba.com — Owning B2B Trust at $50B+ GMV — Juno Chen
Owning says you held the mandate. Redesigning says somebody handed you one. Stakes are genuinely low here: this string renders in a browser tab and a search result, nowhere near the buyer's field of vision. Do it last. Or don't.
Checkpoint 2 — First screen Open since July 24
Current H1: The signals before anyone touched a screen.
Verdict: Fail. This is why the audit exists.
Give the rail its due. Scale clears the fold: $50B+ in annual marketplace volume, 200K+ suppliers, 25M+ desktop sessions. Your title clears it too, Head of Design & Research, North America. So do the outcome numbers. A patient reader finishes fully briefed.
The headline is the thing that patient reader has to survive first. It withholds the subject, withholds the scale, makes no ownership claim, and occupies the one position where the buyer decides whether the next fifty seconds belong to you.
Replacement:
<h1>I named the trust gap nobody owned inside a $50B+ GMV marketplace.</h1>
<p class="hero-sub">Research case built across 32 interviews. Mandate secured.
Three cross-functional sprints. +20% daily transactions.</p>Two things are doing work in there. The verb is named, which settles generate-versus-execute inside the opening clause. And the 32 interviews currently live only in your page metadata, where no human being will ever encounter them. Promoted into the subhead, that figure converts "I had a strong opinion" into "I built the evidence and won the argument."
Yes, the rail repeats $50B+ GMV two inches to the right. Leave it. Repetition is how a scanned page gets retained.
Keep the 25%/80% desktop diagnostic exactly where it is in the body. Strong second beat. Bad opener.
Checkpoint 3 — Craft signal Open since July 24
Current: No product imagery in the hero, just a low-contrast background numeral. Below the fold: nine before/after browser mockups across the three sprints, plus a complete component inventory covering navigation, search, cards, product detail pages, onboarding, global elements and design tokens.
Verdict: Fail, and inverted from the question I was handed. Also, before anyone proposes a hero screenshot: I checked, and I'm ruling it out. Thermo Fisher, Red Cross and Allē all run the identical hero construction. Editorial headline left, structured brief right, no product image. Break that pattern on CS-01 alone and it reads as inconsistency rather than emphasis. Hero stays clean.
The craft evidence is there. It's pitched one level below where it should sit. Nine mockups labeled only by state (Before · 2022, After · Logged Out) prove that pixels moved. They say nothing about who decided the pixels should move. And a full component rack inside a teaser quietly reclassifies you from design leader to systems designer, which is the specific demotion this audit exists to stop.
Fix one. Annotate every redesigned mockup. Format: Decision: [what changed] — [what it bought].
Homepage — Decision: B2B sign-in surfaced above the fold — attacked a 37-point sign-in gap.
PDP — Decision: Trade Assurance moved above the fold — put the guarantee where the hesitation was.
Search — Decision: [your call] — [what it bought].The 37-point figure is yours, out of the sprint diagnostic. If you can substantiate the gap actually closing, write closed. If you can't, leave attacked and let the transaction number carry the outcome. Write the search line yourself; I'm not inventing a decision you made.
Fix two. Collapse the component rack.
Component system: 40+ elements across navigation, search, cards, PDP, and onboarding
Four to six thumbnails underneath. Full inventory moves behind the case. The 40+ is carried over from my July count off the visible rack, not a number you gave me, so verify before it ships.
Annotations are the only item on this list that pays two readers simultaneously. The VP Design reads a placement decision and registers craft judgment. The CPO reads the same fifteen words and registers decision ownership. That split aperture is the argument in "Same Anxiety, Different Room", and a phrase-length annotation is one of the few artifacts that serves both without splitting the difference.
Checkpoint 4 — Call to action Open since July 24
Current: Request the full case study, in the hero and again in the closing block, linking straight through to cs-01-alibaba with no request step anywhere in between. No form. No email capture. What sits on the other side is your client-side overlay.
Verdict: Fail on accuracy. You're charging a toll you never collect.
Replacement, both instances:
View the full case study →
Someone clicks expecting a request flow, lands directly in the case, and has just learned that a word on your page wasn't literal. This page runs entirely on numerical precision. Teaching that lesson thirty seconds in costs more than the click was worth.
If you want the gate, and there's a real argument for knowing who reads this, build the form and write Request access to the full case study. Either choice is defensible. Copy describing a mechanism you haven't built is not.
One number to cut
The outcome row carries four metrics: +7% DAU · +20% transactions · +2.2pt NPS · −47% security concerns. All three sibling cases carry three. Cut +7% DAU. It's the softest number in the set, and it dilutes the three that are genuinely hard to produce.
Priority stack
- Twenty minutes. Swap the H1, add the subhead. Nothing else on this page moves what the buyer believes about who started the work, and the copy is already written. You're relocating text, not drafting it.
- One hour. CTA correction, DAU cut. Two string edits.
- A weekend. Annotations, rack collapse. Worth the weekend. But both improve a page the buyer has already committed to reading, and right now your headline is making that commitment decision for them. Badly.
- Don't bury the lede: Adobe.com's Director of Design says the most interesting thing in a case study is the outcome and that candidates should open there, which is the same ordering argument this audit makes about your hero, stated by someone on the buying side of the table in Adobe Design's account of what hiring managers actually scan for.
- Two rooms, two rubrics: Atlassian separates management evaluation (how you led the team and shaped the result) from craft evaluation (typography, text density, image framing) into different interview blocks, which is worth reading before you decide how much annotation weight the teaser should carry — see the Atlassian Design Interview Handbook.
- The gate that isn't: the CTA problem is downstream of a bigger one, since the full Alibaba case route ships its HTML to unauthenticated browsers and hides it with a client-side overlay — fix the mechanism before you write copy that promises one.
- Equinox+ metadata is still generic: Alibaba, Thermo Fisher, Red Cross, and Allē all carry scale or outcome in their page titles, but the Equinox+ teaser still ships as "Equinox+ · Teaser · Juno Chen," leaving the five-brand, three-month MVP story invisible to search and tab.