Juno, this is a repair dossier. Your homepage is generating evaluator questions that your positioning rules prevent you from answering with design proof. The objection forms during portfolio inspection, before any conversation starts. A verbal reframe cannot intercept something that happens while someone is alone with your website at 9 PM.
This is distinct from a perception gap (fixable by reframing) and from an evidence gap (fixable by building a new artifact). The portfolio's structure itself is manufacturing the objection. The fix is changing what the homepage shows and how it routes attention.
I've audited the homepage as of September 26, 2026, walking the path a hiring evaluator would take. Below: every element that creates a claim-control risk, the evaluator question each triggers, and a prioritized repair list.
What the evaluator encounters
The homepage has no Work, Labs, or Essays navigation. Your name, LinkedIn/GitHub/email icons, a theme toggle. The only way into your work is scrolling.
The hero copy names five inline case links (Alibaba, Thermo Fisher, Red Cross, Allē, Equinox+), all cleanly attributable as BCG Digital Ventures or Alibaba. Then the second paragraph begins "At TinyFish" and links directly to Selected Work. The phrase "designing trust into agentic systems" links to the Trust essay. TinyFish becomes the narrative bridge between your established track record and your AI credibility. That bridge is the source of the problem.
Selected Work, card one: TinyFish. Labeled "Shipped 2025–2026," tagged "NDA Protected," described as "Design patterns for agents that act, not just suggest." This is the first thing an evaluator clicks. The remaining five cards (Alibaba, Thermo Fisher, Allē, Red Cross, Equinox+) follow.
Field Studies contains three articles: the Trust essay, the pricing essay, and a leadership reflection.
Brand Pulse, Retail Velocity, and Carrier IQ — your approved AI evidence — do not appear anywhere on this page. An evaluator who visits only the homepage will never discover them.
Five questions the homepage invites that you can't answer
The bolded line in each section is the likely evaluator question. The explanation covers why the standard TinyFish verbal bridge (past-tense role context, technical currency, what-I-learned narrative) fails at portfolio-inspection stage — the evaluator has already formed the question before they talk to you.
1. The TinyFish case page is fully delivered by the server
The TinyFish card links to a 57,799-byte HTML document. Six chapters, 21 images, two live iframes, Coca-Cola named in the title and meta description, implementation file references (claims.ts, AdjudicationView.tsx), before-and-after design comparisons, rejected alternatives, specific metrics (5,505 restaurants scanned, 163 design decisions, three enforced stop conditions). Client-side JavaScript checks for a local-storage key and redirects unauthenticated browsers to a login page, but the server has already sent the complete page content. No noindex directive.
"Walk me through the trust-architecture decisions in the Coca-Cola pilot." The evaluator has seen screenshots, system-state maps, and before-and-after comparisons. They will ask about specific design choices visible on the page. Your positioning rules classify TinyFish work as past-role context, not design portfolio proof. You cannot answer with design specifics. The page also contains no role or attribution block — nothing stating what you personally designed versus what was produced by TinyFish, Coca-Cola, or engineering. The case reads as single-author, which makes the attribution question harder to manage.
I flagged the painted-on gate problem in Issue #6. That piece covered the pricing essay. The situation has worsened: a full design case with named client, artifacts, and metrics now occupies the first card position.
2. TinyFish occupies the first Selected Work position
Position one is where evaluator attention concentrates. The card description — "Design patterns for agents that act, not just suggest" — uses design-portfolio language. The tags (Agentic Platforms, Human-in-the-Loop, Trust & Oversight, AI-Native Product) are the exact vocabulary an AI-design hiring manager searches for. The card is doing the work of a portfolio centerpiece for a role you held as Head of Product.
"Your most recent design work is NDA-protected — what can you show me from the last two years?" The card has already framed TinyFish as design work. The NDA label sharpens the question rather than deflecting it. And the verbal bridge ("I led product, not design, at TinyFish") directly contradicts what the card just told them.
3. The pricing essay contains commercially specific TinyFish decisions
The pricing essay is 35,103 bytes, fully delivered by the server behind a password overlay that checks a browser hash. It names TinyFish, places you with the CEO and CPO making pricing decisions, describes per-agent-step billing, a failed-run-costs-zero policy, tier renaming after user-behavior analysis, and an enterprise buyer requiring 50 simultaneous agents. No permission signal or publication-authorization statement.
"How did you decide on step-based pricing, and what were the retention effects?" The essay invites this question. You wrote it. But under your positioning rules, you cannot cite TinyFish engagement details as portfolio proof, and the essay contains no numerical outcomes. The illustrations are metaphorical (origami, fish bowls), not product screenshots, so the piece doesn't create inspectable-UI exposure. But the first-person operational detail is specific enough that an evaluator will treat it as a product case study and want numbers you cannot provide.
4. Approved AI evidence is unreachable from the homepage
Brand Pulse, Retail Velocity, and Carrier IQ exist at standalone URLs. They are working applications. Issue #11 positioned Carrier IQ as your strongest approval-and-control artifact and Brand Pulse/Retail Velocity as prototyping evidence. These artifacts answer the AI-credibility question without triggering a positioning conflict.
They are invisible from the homepage. An evaluator who spends five minutes on your site — generous — will encounter TinyFish, five BCG DV/Alibaba cases, the Trust essay, the pricing essay, and a leadership reflection. They will not encounter a single piece of approved AI evidence beyond the Trust essay unless you send a direct link in a separate message.
"Where's the AI work? I see one NDA case and some essays." The homepage creates this question by omission.
5. LinkedIn and search-index lag
LinkedIn still reads "Juno Chen - TinyFish" with present-tense language about building human-AI collaboration infrastructure. Third-party directories (The Org, SignalHire) list you as Head of Product at TinyFish, current. The search index may still serve a cached version of the site from roughly two months ago with different structure and language.
"Are you still at TinyFish? Your LinkedIn says current, but you're describing it in past tense." Secondary surfaces, but recruiters check LinkedIn before or instead of the portfolio. If LinkedIn says TinyFish/present-tense and the homepage says TinyFish/first-card, the evaluator has no reason to question the framing. The verbal bridge has to fight both surfaces at once.
Where the approved proof sits
Brand Pulse, Retail Velocity, and Carrier IQ run at standalone URLs. Carrier IQ in particular answers the approval-and-control question that AI-design hiring managers are screening for. The Trust essay provides the conceptual frame and is linked twice from the homepage (hero and Field Studies). The three applications are not named, linked, or discoverable from any page the homepage connects to.
Repair checklist
Ordered by which fix eliminates the highest-severity risk first.
-
1. Server-side gate the TinyFish case page. Move the access check from client-side JavaScript to the server. Return a 401 or minimal stub for unauthenticated requests. Remove the full HTML payload from the public response. This eliminates the most damaging exposure: a complete design case with named client, artifacts, and metrics that your positioning rules say you cannot use as proof. Until this ships, every other repair works around it.
-
2. Move TinyFish out of position one. Reorder Selected Work so a BCG DV case (Thermo Fisher or Red Cross — both have inspectable design proof and clean attribution) leads. If TinyFish stays in the grid, move it to position five or six and revise the card description to remove design-portfolio language. Replace with product-leadership framing that matches your actual role. This changes what the evaluator clicks first: work you can discuss freely.
-
3. Add a navigation route to approved AI evidence. Create a Labs or AI Work section in the nav bar — not buried in the scroll — linking directly to Brand Pulse, Retail Velocity, and Carrier IQ. These artifacts answer the AI-credibility question without triggering a positioning conflict. They need to be reachable in one click from the homepage, ideally before the evaluator scrolls to Selected Work.
-
4. Rewrite the hero's TinyFish paragraph. The current version makes TinyFish the narrative bridge to your AI story. Rewrite so the bridge runs through approved evidence instead. Link "designing trust into agentic systems" to Carrier IQ or the Trust essay paired with Carrier IQ. Use TinyFish as a one-line role mention (Head of Product, past tense), not the anchor of the AI narrative.
-
5. Server-side gate the pricing essay. Same fix as the TinyFish case: move the access check to the server. If you want it available to specific readers, use a token-based URL you can share selectively.
-
6. Add a role-attribution block to the TinyFish case. If the page remains accessible to authenticated readers, it needs a clear statement: your role was Head of Product. What you personally drove versus what was produced by design, engineering, or the client. Without this, the case reads as single-author design narrative, and every artifact on the page becomes a claim you'll be asked to defend as design work. This makes authenticated sharing safe — you can send the case to specific evaluators without the attribution ambiguity generating a new objection.
-
7. Update LinkedIn. Shift the headline and TinyFish experience entry to past tense if you've left or are leaving. If you're still there, ensure the framing matches the homepage's positioning: product leadership role, not design case material. This aligns the surface recruiters check most often with the story the portfolio tells.
-
8. Force a re-crawl. Once homepage repairs ship, use Google Search Console's URL Inspection to request re-indexing. The cached version in the search index uses older structure and language.
Interim verbal bridge (stopgap only)
If a conversation happens before repairs ship, use these. But be clear: this manages one conversation while the homepage continues generating the objection for every evaluator who visits without talking to you.
When asked about the TinyFish case: "I led product at TinyFish — the trust-architecture thinking you see there is real, but my role was Head of Product, not design lead. The design work I can walk you through in detail is [Carrier IQ / Thermo Fisher / Red Cross]. Want me to pull that up?"
When asked about the pricing essay: "That's a reflection on decisions I was part of as Head of Product. I can talk about the reasoning, but the design proof I'd point you to for AI work is Carrier IQ, where I own every decision end to end."
When asked where the AI work is: "I have three working applications — Brand Pulse, Retail Velocity, and Carrier IQ — that demonstrate trust architecture and agent-oversight patterns. They're live but not linked from the homepage yet. Let me send you direct links."
That last line — "not linked from the homepage yet" — is an admission that the portfolio's own structure is failing you. Ship repair #3 so you can stop saying it.
- Client-side vs. server-side access: Google's own guidance identifies URL Inspection in Search Console as the authoritative way to compare what's indexed against what's live, which matters for confirming whether the TinyFish case content is appearing in search results.
- Role constructs behind "AI-native design": OpenAI's Engineering Acceleration posting screens for observability, uncertainty, and instrument-to-decision loops — the closest current match to Carrier IQ's approval-and-control pattern, and worth tracking as a proof-alignment target.
- Where code becomes a gate: Anthropic's Evals & Prompts role places production Python and maintained evaluation pipelines under minimum qualifications, a mandate boundary that no portfolio reorder or verbal bridge can address.
- What work samples can prove: OPM's assessment guidance says work samples should mirror actual job tasks and are appropriate only when candidates are expected to possess the measured competencies on entry — a useful boundary for deciding which artifacts to build versus which gaps to acknowledge.