Nine days since the July 10 audit. The teaser was modified July 18. The diff is invisible in the rendered HTML. No structural changes, no copy changes, no asset changes detectable from the public source. Every issue flagged on July 10 persists in the July 19 fetch.
Still broken:
og-cs-02.jpgreturns 404. Every Slack share, every LinkedIn forward, every recruiter paste renders without a preview image.- Zero
<img>elements. The teaser is imageless HTML. 42% overhead reductionabsent. The full case carries it. The teaser does not.100% partner adoptionstill reads as a percentage, not a count. Still not6/6.- Artifact grid still 13 items. Same count, same labels, same over-reveal risk.
The browser title remains strong: Thermo Fisher / mySupply — 0→1 Pharma Supply Chain, $20M+ Margin — Juno Chen. That was the one thing the July 10 audit didn't need to touch.
Now the six points. Each one evaluated against the question that should have anchored the first audit: does your teaser surface, at scan speed, that you designed a human-agent decision boundary with regulatory stakes? That is the single most differentiating element in this case for the buyers who matter right now. If it's buried, every other fix is cosmetic.
1. Page Title
Your browser title communicates 0→1, pharma, and margin scale. The meta description adds Six partners, nine sites, twelve months. $20M margin opportunity, 100% partner adoption, 83% IRR. Dense, directional.
What it does not communicate: that this case contains designed human-agent decision architecture. A CPO at an AI-native company reads "pharma supply chain" as a domain label. It tells them where you worked. "Designed the gate where human judgment overrides five continuous agents in a regulated release workflow" tells them you've already solved the problem they're still trying to frame. One is a credential. The other is a hiring signal.
The title can't carry all of that. It should carry a trace.
Recommended browser title: Thermo Fisher / mySupply — 0→1 Pharma Supply Chain, Human-Agent Decision Architecture, $20M+ Margin — Juno Chen
That runs long. Google truncates around 55–60 characters, browser tabs around 70, LinkedIn shares around 65. The recommended title exceeds all three. If you need to cut, drop $20M+ Margin from the title and let the meta description carry it. The margin metric already appears in the hero Outcome row. The human-agent frame appears nowhere on the public surface. [Moderate confidence on the truncation thresholds; they shift by platform and character width.]
2. First Screen
Your H1 is The supply chain six partners never had. Clean, but it's a tagline. The structural insight that makes this case worth requesting lives in the hero intro: exceptions surfaced at the delivery gate, costing 3–5× more than catching them a week earlier. That diagnosis is present. It lands.
What doesn't land at first-scroll speed: the line between that diagnosis and your design response. The hero intro says "one platform ended that" and "all six companies adopted it." Outcome language, not architecture language. A hiring manager scanning this registers: problem existed, platform solved it, partners adopted it. They do not register: the designer diagnosed where human judgment needed to intervene in a continuous monitoring system and designed the boundary between automated exception routing and irreplaceable regulatory review.
One sentence fixes this. Add it after "one platform ended that."
Recommended bridge sentence:
"The design moved exception detection upstream by a week, automated what could be verified, and preserved the one human gate that regulation makes irreplaceable: batch QA release."
The sentence quantifies the timing shift, signals agent-human architecture thinking, names the regulatory constraint. It does not over-reveal the five-agent model or the full Section 03 vision. It gives the AI-native buyer a reason to request the full case.
3. Outcome Proof
$20M margin opportunity · 100% partner adoption · 83% IRR appears in the hero right rail Outcome row. On desktop, the .hero-split layout places this in the right column. On mobile, it collapses below the intro paragraph.
Three issues. All unchanged from July 10.
42% overhead reduction is still missing. The full case carries it in the hero stats row. Your teaser should too. Four metrics are stronger than three. Overhead reduction speaks directly to operational buyers who live in cost structures, not IRR calculations.
100% partner adoption should be 6/6 partner adoption. The fraction makes the denominator visible. "100% adoption" could mean two out of two. "6/6" means you got every single one, and the number is large enough to matter.
The Outcome row sits after the hero intro and CTA in source order. On a fast scan, a hiring manager's eye may hit the CTA link before registering the metrics. Metrics should appear before or visually adjacent to the CTA. Not after it.
Recommended Outcome row: $20M+ margin recovered · 42% overhead reduction · 6/6 partner adoption · 83% IRR
4. Craft Signal
The zero-<img> problem compounds here. Your teaser's only visual craft evidence is the embedded iframe dashboard (desktop browser shell at mysupply.patheon.com/supply-chain/batches and a mobile batch-detail mockup) and the 13-item CSS-rendered artifact grid.
The dashboard iframe loads. The mobile mockup shows a real batch record with real data: batch AB1234, At Risk state, $174,000 order total, sterility test pending. Credible product design work rendered in HTML/CSS rather than images.
But scan-speed legibility is the problem. Fifteen seconds on this page, a hiring manager sees text-heavy HTML with one embedded app frame and a grid of small labeled boxes. They do not see the high-fidelity design comps that signal senior craft at a glance.
Your artifact grid contains the teaser's only hint of the agentic architecture: item 05, the KPI Card Component, carries labels reading AI Gate Reviews, 24 this wk, 87% avg confidence. Those labels are doing real work. They're buried in a 13-item grid where the eye has no hierarchy. A hiring manager scanning the grid sees "Color Tokens" and "Buttons" first because those are items 01 and 03. The AI Gate Reviews card is item 05. The regulatory constraints section is item 13.
The grid order is actively working against your strongest argument. Lead with the KPI card. The agentic signal should be the first artifact a scanning eye registers. Not the fifth.
5. Density
The July 10 audit recommended cutting the grid from 13 items to 4–5. That recommendation stands. The reasoning has sharpened.
Your teaser currently reveals design-system primitives (color tokens, buttons, type scale), multiple analytics panels, data visualizations across six categories, and regulatory-constraints copy. The full case reserves the five-problem-to-five-module map, the detailed module walkthroughs, and the entire Section 03 agentic architecture. Right split in principle. Wrong execution in practice: the teaser gives away the artifacts without giving away the argument they serve.
A CPO who sees color tokens, status badges, and six data-viz panels thinks: design-system case study. A CPO who sees an AI Gate Reviews KPI card, an Exception Summary panel, and a regulatory-constraints frame thinks: decision architecture. Same artifacts, different selection, different signal entirely.
Recommended grid (5 items): [High confidence. This is the single highest-leverage change on the page.]
- KPI Card — AI Gate Reviews (relabel:
AI Gate Reviews · Agent Confidence Scoring) - Exception Summary
- Regulatory Constraints — GDP & cGMP
- One analytics panel (whichever is most visually striking)
- Batch Pipeline
Cut color tokens, buttons, type scale, status badges, and the remaining analytics/data-viz panels. Those belong in the full case. They are craft evidence, not positioning evidence. Your teaser's job is positioning.
6. CTA
Request the full case study appears in the hero and closing block, linking to cs-02-supply-chain.html. Functional copy. It tells the reader nothing about what they'll get that they haven't already seen.
Recommended CTA:
"Request the full case — five modules, one human gate, the agentic architecture behind $20M in recovered margin."
That names what's behind the gate, gives the AI-native buyer a specific reason to click, and introduces the human-gate concept at the exact moment the reader is deciding whether to invest further.
The client-side gating issue persists: the full case still ships complete HTML before the localStorage redirect fires. For a case study about trust architecture and regulatory decision gates, this remains the structural contradiction flagged on July 10.
What This Adds Up To
Your full case Section 03 contains the strongest positioning evidence in the portfolio. Five modules become five agents running continuously. Exceptions classified before a manager logs in. One human gate preserved where a regulatory signature is irreplaceable. Your Trust essay carries the same frame, published and public, naming the five agents and the QA release gate explicitly.
Your teaser carries none of it. No agents, no human gate, no architecture. Scattered breadcrumbs in artifact labels nobody reads at scan speed. The content exists. It's publicly available on another page. The one page that's supposed to make someone want more doesn't surface it.
Add the bridge sentence to your hero intro. Reorder and cut the artifact grid to lead with AI Gate Reviews. Rewrite the CTA to name the agentic architecture. These take content that already exists and make it legible at teaser scan speed, without revealing the architecture itself.
Your teaser's job is to make a CPO at an AI-native company read the first screen and think: this person has already designed the thing we're trying to figure out. Then they request the full case. Then they see it.
Right now they see a supply chain dashboard. The argument is already written. Fix the surface.
- Amplitude's human-agent language: Their live Head of Product Design posting explicitly asks for someone who can define how agents and humans work together in the same workflows, which is the exact buyer anxiety the Thermo Fisher teaser's bridge sentence would answer.
- Vanta's AI-native design org: Their Head of Design role asks for agentic UX patterns, quality systems, and definitions of done for AI-native product experiences, making it another buyer where the human-gate frame would land before the supply-chain domain label does.
- Trust essay as public proof: The Decision Gate section of Trust Is the New Interface already publishes the five-agent and QA-release-gate language explicitly, which means the teaser is withholding a signal that's freely available on another page of the same site.
- Client-side gate pattern persists: The full CS-02 case still ships complete HTML before the localStorage redirect fires, matching the same structural contradiction flagged across all five full cases in the July 10 site-gate audit.