Fetched live August 1, 2026, 1:58 PM PDT. No credentials, no session. The request came back HTTP 200 and 154,227 bytes: the whole case, every section and metric, handed over before the password check runs. The gate is a script sitting in the page body. It reads local storage and bounces an unauthorized browser to /login.html after the content has already arrived. Same behavior as July 24. Last-Modified reads 8:38 this morning. Treat that header as low reliability; static hosts rewrite it on deploy. It tells you the file moved, not that anything inside it changed. One sentence changed. The rest is eight days old.
This case beats most speed cases and it will still fail a craft review, because it never descends to the floor where craft gets judged. The highest-value item this week is not a copy pass. It is producing one annotated component iteration, because the page currently has none.
Delta against July 24:
- Closed: one. The data-analysis sentence now reads "I spent a week in the data before the architecture became obvious."
- New: nothing. No new sections, no new assets, no revised copy in the five decision blocks.
- Open: six, one of them half-done.
The check nobody says out loud
A panel reviewing a compressed timeline runs a test it never announces: did the speed eat the craft? They go looking for the evidence that would be missing if it were true, and when they fail to find it they do not file "no evidence." They file "confirmed."
Prose will not move that. The only thing that retires it is work shown at the altitude where cutting a corner leaves a mark.
System level is architecture and rule. Which layer is shared across brands, which one varies, what object a member actually navigates from. Your five decision sections live here, and here you are above market. "Shared where behavior needed consistency, variable where brand meaning mattered" is a defensible systems position. Most portfolios never get this far.
Underneath sits program level: scope, duration, headcount, outcome. Your hero and Operating Scope handle it well. It is also the altitude that is easiest to fake.
Component level is one element and its history: three versions of the class card, why the second one died, what the third inherited from it.
July 24 flagged the timeline pressure going quiet as the reader descends. That still holds. But I ran the wrong test. I checked whether the constraint stayed visible inside the craft sections. I never checked whether those sections carried component evidence at all. They do not. They never have. So this is baseline for that standard, not a regression against it.
The cost of that lands at portfolio level, not page level. Equinox+ and Red Cross are the two cases carrying non-expert trust calibration in your portfolio: ordinary people, no domain expertise, deciding whether to act on what a system puts in front of them. Red Cross holds the vulnerable-population claim. Equinox+ is supposed to prove the same mechanism at consumer scale. Read as scope reporting, it proves nothing of the kind.
Six dimensions, scored
| # | Dimension | Verdict | Fix |
|---|---|---|---|
| 1 | Constraint as design driver | Insufficient | One constraint sentence per decision block |
| 2 | Data-to-architecture derivation | Sufficient, qualified | Add sample size, window, source |
| 3 | Multi-brand tradeoff | Rule sufficient, tradeoff no evidence | One contested brand request, resolved |
| 4 | What got cut | Insufficient | Give one cut its reason |
| 5 | IC + manager proof | IC sufficient, leadership no evidence | Two sentences on a call you made |
| 6 | 4.8★ traceability | Asserted, not evidenced | Qualify the rating, quote two reviews |
1. Constraint as design driver — insufficient. One passage on the page does this properly, and it is genuinely excellent:
"With 6 weeks left and engineering capacity for 3 brand worlds at launch, I prioritized SoulCycle, Equinox, and Pure Yoga — the three brands with the highest instructor-followership density in the retention data. Precision Run and HeadStrong shipped in the first post-MVP update."
Constraint, criterion, decision, consequence. It appears once and then never again. Inside Decisions 01 through 05 the visible copy never touches the deadline, the six-week mark, or engineering capacity. Five architecture stories that could have been produced on any schedule at all.
Fix. One sentence per decision, this shape: "[Constraint] meant I could not [the better option]. I chose [what shipped] because [criterion], and accepted [specific cost]." Fill it with facts already on your page: "Six weeks out, with engineering capacity for three brand worlds, I could not ship all five. I chose SoulCycle, Equinox, and Pure Yoga on instructor-followership density in the retention data, and accepted that Precision Run and HeadStrong members would open a launch app with no home in it." You have four more of these in your head. None of them are on the site.
2. Data-to-architecture derivation — sufficient, one qualifier. Best chain on the page:
"Members who followed at least one instructor within their first 30 days retained at nearly double the rate of those who didn't. The clearest signal in the data — and it wasn't in the brief."
Then the move: if followership predicted retention, the primary object had to be the instructor, not the content category and not the brand. A reviewer can trace that. Leave it alone.
The qualifier is sourcing. No sample size, no cohort window, no retention definition. Anyone analytically literate registers that absence and quietly discounts the number without telling you. Nine words closes it: "(n = [figure], [window] retention, [cohort dates], via Equinox data science)." If the figures are unrecoverable, say so: "directionally — I did not have a clean control cohort in the timeframe." Stating the limit beats leaving the hole open.
3. Multi-brand tradeoff — rule sufficient, tradeoff no evidence. You give the rule, the token architecture, five finished brand worlds, five Creative Directors each controlling their own expression without breaking anybody else's system. What you do not give: a contested request, a rejected brand treatment, an abstraction that failed and had to be rebuilt. Not thin. Zero.
Two readings are available to a panel. Either the tradeoff was trivial, or you sanded it off before publishing. Neither one helps you. A rule with no cost attached is documentation; the compromise it forced is the part that reads as judgment.
Fix. One paragraph inside Decision 04, under a subhead reading "Where the rule broke." Which brand asked for what. Why the shared layer could not carry it. What you gave them instead, and what they gave up. Not the most dramatic instance on the project. One real one.
4. What got cut — insufficient. Four MVP exclusions are listed: two brand worlds, personalized recommendations, tracker integrations, social and challenge mechanics. Only the brand sequencing carries a reason. A list of cuts reports scope. Attach a reason and it becomes a decision.
Fix. Use the recommendations cut. Personalization is the obvious lever in a fitness product and it is what a reviewer will press on. "Personalized recommendations lost to the QR club credential. Recommendations needed engagement data we wouldn't have until after launch; the credential was the only thing that made digital and physical feel like one membership on day one. Recommendations shipped in [period]." Adjust to what actually happened. Put the loser and the winner in the same sentence.
5. IC craft sufficient, leadership behavior no evidence. Four designers and a researcher, two hired mid-project, five brand PMs, two engineering leads, five Creative Directors, data science. That is team composition. It is an input, not proof you led anything. One row does real work and should survive verbatim:
"Owned: IA, cross-brand navigation model, design system, token architecture. Influenced: roadmap sequencing, MVP scope, brand-team sign-off process."
Everything else is absent: no review where you sent work back, no standard you set that somebody else then applied, no disagreement you closed.
Fix. Two sentences in Operating Scope, in the "One call under pressure" slot you already built. A specific review, the bar you held, what changed. Or the sign-off process you designed with the Creative Directors and why version one of it failed.
6. The 4.8 star rating — mapped, asserted, not evidenced. It appears four times. Two instances carry "at launch." The two hero placements, your most-read surface, do not. No rating count, no date, no store region, no link, no screenshot. The claim that instructor discovery, brand-specific class quality, and physical-digital continuity drove the rating arrives with no quoted review, no coding method, no frequency. It lands neatly on your three strongest decisions, which is either true or exactly what a skeptical reviewer expects a portfolio to write.
Fix, paste-ready. Both hero instances become "4.8 stars, App Store at launch." Under the launch section: "Rating as of [date], [n] ratings, US App Store. Review themes coded from [n] reviews in the first [period] post-launch." If you never coded them: "Themes are my reading of the top reviews at launch, not a coded analysis." Then quote two real reviews and set them beside the decisions that produced them. One member sentence next to the choice that caused it does more work than the number.
Altitude check: zero
Nothing on this page clears the bar. Not one element.
The bar: multiple states of a single element, shown together, with written rationale for why each version was killed or advanced. Not a before-and-after redesign. Not two consecutive screens in a flow.
The page carries 23 images across 13 unique assets and zero figure or figcaption elements anywhere in the markup. There is no captioning apparatus at all, which means annotation has to be built as structure rather than dropped into alt text. Decision 03 comes closest, pairing class detail with booked confirmation, but those are two moments in one flow. Decision 02 pairs an instructor profile with the home feed it produces: system consequence, genuinely valuable, not iteration history. No version labels appear anywhere on the page.
Fix. Build one set. Use the class card. Every multi-brand tension converges on it: it has to read as SoulCycle and as Pure Yoga, carry instructor identity, and survive a brand-neutral Explore surface. Three states left to right, two sentences under each. What this version attempted, why it failed, what the next one inherited. Half a day if the Figma history survived. If it did not, rebuild from memory and label it a reconstruction. Labeled that way it is honest and it still clears the bar; a polished final screen with nothing behind it never does.
One set is sufficient. The bar is existence, not volume.
Fix order, cheapest first
Tonight. Under an hour, nothing new to produce.
- Qualify both hero ratings "at launch."
- Kill the teaser eyebrow still reading "Case Study 03" while the hero and sidebar say 04. Open eight days, thirty seconds to close. A reviewer who catches a numbering error reads everything downstream more skeptically.
- Ship the teaser metadata from July 24, still untouched. The full case has no meta description and no Open Graph tags either, so it previews as a bare URL in every message you send it in.
- Settle the CTA. The teaser says "Request the full case study" and points at a password wall with no request mechanism behind it. Wire a real path or rewrite the copy to describe what actually happens.
This week, from artifacts that already exist. Every item here is recovery of a decision you already made.
- Five constraint sentences, one per decision block.
- One contested brand request, and how it resolved.
- One reasoned cut.
- The retention-claim qualifier.
- The two management sentences.
This weekend, new production.
- The annotated component set. The only item that requires you to make something, and the only item that answers the question the panel is actually running.
On the gate: right now anyone with the URL has the complete document. Pick a side. If it is private, gate it server-side. If the password is a signal rather than a control, take it off, because the friction is filtering out exactly the readers most likely to go deep.
The architecture story is there and it is good. No volume of well-written system-level prose will substitute for one element shown three ways.
-
Suno is live and consumer-adjacent: Equinox+ is your closest published consumer precedent, and the Head of Product Design search at Suno is active with a five-day onsite requirement in LA or New York — but the posting sits over an AI-native creative product, and nothing in your portfolio currently addresses authorship or creator rights, so treat that as a real gap rather than a reframing exercise.
-
Staff craft is the recurring ask: Stripe's Senior Staff Product Designer, Agentic Commerce posting wants 15-plus years and inspects at component level, which is the same bar the Equinox+ altitude check just failed — one annotated set serves both audits.
-
Speed is now an explicit quality risk in postings: Amplitude's Head of Product Design role states outright that AI makes interface production cheap while letting quality erode, which turns "I shipped five brands in three months" from an achievement into a question you have to answer with component evidence.
-
The gate problem is portfolio-wide, not Equinox-specific: All five full-case routes ship their HTML before the client-side access check, including the Alibaba case — fixing one page's gating without fixing the pattern leaves the contradiction intact for any reviewer who checks a second URL.