Cycle date: September 27, 2026 File version: 9 Supersedes: v8 Scope boundary: Portfolio state and artifact inventory. Company-specific outreach framing belongs to Act-Tier Positioning Sheets. Entries here map which evidence applies to which company. They do not contain message language.
What changed in v9:
- Inference-economics vocabulary from current provider pricing and Stripe's OpenRouter acquisition added to Domain 1 artifact detail
- Stripe Data & AI and Anthropic Safeguards posting signals reflected in positioning mappings
- Inference-Aware Decision Surface entered as first Domain 1 artifact with full specification, status
planned, and handoff mapping
Identity and Current Status
Name: Juno Chen
Current role: Independent — building portfolio artifacts, consulting
Prior role: Head of Product, TinyFish (AI-native product company)
Target role: Senior/Staff Product Designer or Design Director at companies building AI-native products, with emphasis on trust, control, and human decision authority in agentic and inference-dependent systems
Throughline: Designing what an interface is entitled to claim — verified, computed, traced, authorized, or completed.
TinyFish Context
Title: Head of Product
Company: TinyFish, an AI-native product company
What she built: AI-native and agentic systems; production experience with model behavior, trust architecture, and system-level constraints
Bridge narrative (standard):
"I went to product to learn how AI actually works. I came back to design because I missed making things people can feel."
Bridge narrative (technical): She moved into product to build AI-native systems and understand production constraints directly, then returned because trust, control, and human decision rights had become the highest-leverage design problem.
Usage rules — hard boundaries:
- Past tense only. Practitioner credibility, not portfolio proof.
- No TinyFish outcome claims, confidential screens, internal traces, private artifacts, architecture detail, or pricing.
- The homepage links a Coca-Cola × TinyFish teaser. That teaser does not override the evidence boundary. Do not cite it as portfolio proof or expand beyond what the public page shows.
- Do not invent or expand a reason for leaving TinyFish.
- Do not claim "first designer" without evidence establishing that role.
Forward-Looking Artifact Inventory
Eight specification cycles have produced zero published forward-looking artifacts. The Delegation Contract, Claims Ledger, Correction Lineage, and Inference Conflict Surface were all specified in prior issues. Not one has shipped or appeared on the homepage, in site navigation, or at any conventional URL route as of September 27, 2026. Until one ships, the AI-native positioning formula has no frontier artifact to lead with.
| Artifact | Domain | Status | Handoff Mapping | Target Companies |
|---|---|---|---|---|
| Inference-Aware Decision Surface | D1: Inference-Aware UX | Planned — no public evidence supports any other status | In-Progress (routing-change disclosure), Output Review (contract-variance disclosure), Decision Gate (resource selection before irreversible action), Intent-Setting (delegated routing authority and governing objective) | Anthropic Safeguards, Stripe Data & AI, OpenAI Identity |
| Claims Ledger | D3: Agent Infrastructure as UX; D4: Human-Agent System Design | Specified, not built | Output Review, Decision Gate | OpenAI Identity |
| Correction Lineage | D1 or D3 (prior records conflict) | Specified, not built | Loop Feedback, Output Review | Anthropic Evals & Prompts |
| Delegation Contract | D2 and D4 | Specified, not built | Intent-Setting, Decision Gate, Loop Feedback | Anthropic Safeguards, OpenAI Identity |
Domain 1 Artifact Detail — Inference-Aware Decision Surface
What it would demonstrate if built: That routing across models is a delegated-authority decision with a governing objective and material conflicts that must be identifiable by the user or operator. The design surface makes cost, latency, reliability, and verification burden visible as consequences of routing choices — not hidden behind automatic optimization.
Inference-economics vocabulary (positioning-ready terms):
- Outcome, stakes, reversibility — what the user needs from this request, and what happens if the answer is wrong
- Latency tolerance, acceptable spend — constraints the system must respect, not optimize away
- Reasoning effort — user- or system-selected depth (e.g., Gemini 3.8 Flash's three thinking levels, Claude Opus 5.5's standard and fast modes) treated as a design parameter with cost and quality consequences
- Cache state — whether the system is reusing prior computation, with pricing implications (Anthropic and Google now price cache reads and writes separately from fresh input)
- Route and fallback — which model handles the request and what happens when the primary route fails or exceeds budget. Stripe's OpenRouter acquisition positions routing across task complexity, price, speed, and reliability as a product surface
- Verification burden — who checks the output and at what cost. A cheaper model that requires human review may cost more than an expensive model that doesn't
- Governing objective — the optimization target the routing system serves. If the user can't identify it, the system is making consequential decisions without disclosure
Predecessor: The Inference Conflict Surface specification from Issue 8 established the consequence-first grammar: cost ceiling, acceptable delay, minimum reliability, hard vs. movable constraints, explicit unsatisfiable states. Provider names and prices from that specification are superseded. The interaction structure carries forward.
Handoff grounding: Carrier IQ's existing In-Progress and Output Review stages provide the production foundation. The artifact extends these with routing-authority disclosure and inference-budget visibility that Carrier IQ does not currently demonstrate.
Agentic Labs Foundations
Navigation constraint: Labs are reachable at standalone URLs but absent from homepage navigation. The homepage links case studies, the TinyFish teaser, the Trust essay, and two essays. Not the three Labs. Outreach referencing Labs work requires manual deep-linking. All four domain-name artifacts and both specified-but-unbuilt artifacts return 404 at conventional routes. The site's /sitemap.xml also returns 404.
| Lab | URL | What It Demonstrates | Trust Essay Handoff | Forward Domain Connection |
|---|---|---|---|---|
| Carrier IQ | carrier-iq.junochen.com | Structured intake, parallel carrier stages, evidence comparison, coverage deltas, review classifications, operator notes, re-verification, Approve Bind gate | All five handoffs | Closest production foundation for all four domains. Strongest for D1 and D4 |
| Brand Pulse | brand-pulse.junochen.com | Mission state, source-agent activity, source-labeled cards, rerun/restore patterns | In-Progress, Output Review | D2 (Intent-Based Interaction), D3 (Agent Infrastructure) |
| Retail Velocity | retail-velocity.junochen.com | Ranked results with confidence labels | Output Review | D1 (Inference-Aware UX) through confidence disclosure |
Portfolio-wide code-evidence gap: No authored production code, no public repositories, no named prototyping tools across all published case pages and Labs. This gap affects any role where code proximity is a gate. Risk factor at Anthropic Safeguards; disqualifier at Anthropic Evals & Prompts.
Missing failure-correction chain: No Lab currently demonstrates the complete sequence: wrong output received, diagnosis, correction applied, comparable later run, measurable improvement. This gap matters for any role involving evaluation pipelines or regression testing.
Case Studies
All business outcomes are Juno-published portfolio claims, not independently verified.
BCG DV attribution rule: CS-02 through CS-05 require BCG Digital Ventures attribution in every reference. CS-01 (Alibaba) and CS-06 (Carrier IQ) do not carry BCG DV attribution.
| ID | Project | Key Metrics | Domain Tags |
|---|---|---|---|
| CS-01 | Alibaba.com B2B Platform Redesign — Head of Design & Research, North America | +7% DAU, +20% transactions, +2.2pt NPS, 47% fewer buyer-reported security concerns. Scale: $50B+ GMV, 200K+ suppliers, 25M+ desktop sessions | B2B marketplace, procurement UX, enterprise trust |
| CS-02 | Thermo Fisher / mySupply — Product Design Director, BCG DV | $20M margin opportunity, 100% partner adoption, 83% IRR. Scale: 6 pharma partners, 9 manufacturing sites, 12 months zero-to-live | Pharma supply chain, regulated enterprise, exception-first operations |
| CS-03 | American Red Cross Disaster Relief — Product Design Director/GM, BCG DV | $847K disbursed, 1,689 cases opened in first two weeks. Scale: 6 legacy systems consolidated, national deployment in 6 months | Case management, regulatory compliance, mission-critical |
| CS-04 | Allē Dual-Surface Loyalty Redesign — Product Design Director, BCG DV | 3.2× redemption, CAC reduced $92 to $42. Scale: 30M+ members, 40K+ provider practices | Consumer + provider backoffice, two-sided healthcare-adjacent growth |
| CS-05 | Equinox+ Five Brands One Platform — Product Design Director, BCG DV | Zero to MVP in 90 days, 4.8-star App Store rating at launch. Scale: 600K+ premium members, 200+ clubs, 5 brands | Consumer tech, subscription engagement |
| CS-06 | Carrier IQ Agentic Insurance Lab — Solo-built | No published adoption, revenue, or customer-outcome metrics | Agentic insurance, regulated review, provenance, human approval |
Metric cautions — do not deviate from these:
- CS-01: Homepage says "40M buyers"; case page says "25M+ desktop sessions." Different denominators. Use the case page figure.
- CS-02: Current page says "$20M margin opportunity." Older material said "$20M+ margin recovered annually." Use the current, narrower wording.
- CS-04: The visible page uses 47% for expiry-notification open rate and says 68% of reactivating members cited the notification. Do not present both 47% figures as separate outcomes.
- CS-06: Omit timing claims (five carriers in 90 seconds, eight portals in four minutes). Describe design mechanics instead.
Cummins / ZED Connect — supporting proof only: Industrial-IoT and connected-vehicle project at BCG DV. No public case page or outcome record. Can be mentioned as additional industrial scope alongside Thermo Fisher. Must not be presented as a full case study or source of metrics.
Trust Essay and Five Handoffs
Trust Is the New Interface defines five handoffs and the Watch → Verify → Delegate ladder.
| Handoff | Definition | Production Evidence |
|---|---|---|
| Intent-Setting | Structured intake, confirmed interpretation, success defined before execution | Carrier IQ profile/coverage/carrier selection; Brand Pulse launch inputs; Alibaba procurement structure |
| In-Progress | Visible steps, sources, and accumulating work — not a spinner | Carrier IQ session stages (Navigate, Fill, Extract, Verify); Brand Pulse mission state and source-agent activity |
| Output Review | Evidence, provenance, confidence, and disclosure of what the system did not find | Carrier IQ evidence/coverage review; Brand Pulse source-labeled cards; Retail Velocity ranked confidence-labeled results |
| Decision Gate | Enough context for a consequential decision; humans retained where outcomes are irreversible | Carrier IQ Approve Bind; Thermo Fisher QA release; Red Cross supervisor and payment controls |
| Loop Feedback | One run's output becomes input for the next, creating compounding risk where correction paths are absent | Carrier IQ operator notes and rerun verification; Brand Pulse rerun/restore patterns |
Per-Company Positioning Mappings
Tier structure this cycle: Four Act-tier companies. No Watch-tier companies are active. Watch targets that emerge between cycles will be added in v10.
Ranked by current actionability. Outreach framing belongs to Act-Tier Positioning Sheets.
Act 1. Stripe — Senior Staff Product Designer, Data & AI
Why first: Strongest evidence-to-posting alignment of any target on this list. No code gate. The OpenRouter acquisition makes inference-economics vocabulary part of this role's daily problem space.
- Posting: Systems strategy, zero-to-one work, high-stakes trust, interaction craft. Does not establish ownership of OpenRouter or agentic commerce.
- Gate vs. preference: No explicit code gate. Systems-level reasoning and trust architecture are core requirements. Screens primarily for interaction judgment and trust architecture, secondarily for agent-workflow orchestration.
- Lead with: Inference-Aware Decision Surface when built — routing as a product decision maps directly to Stripe's OpenRouter context. Until then, lead with Carrier IQ as inspectable trust architecture.
- Supporting: Alibaba (platform scale, B2B trust at $50B+ GMV), Thermo Fisher (multi-organization system, data-driven operations), Trust essay.
Act 2. Anthropic — Product Designer, Safeguards
- Posting: Last updated September 25. No named hiring manager, no reporting line, no explicit Enterprise Frontier Safeguards connection. Classified under Engineering & Design — Product.
- Gate vs. preference: Complex-workflow experience, AI-tool use (Claude Code), and end-to-end execution are minimum qualifications. Front-end coding experience is preferred, not minimum.
- Lead with: Carrier IQ (all five handoffs, approval gate, regulated review). Inference-Aware Decision Surface when built.
- Supporting: Red Cross (mission-critical decision authority), Thermo Fisher (regulated exception handling), Trust essay.
- Gap: No named buyer for outreach targeting. Code proximity is preferred but not gated — the portfolio-wide code-evidence gap is a risk factor here, not a disqualifier.
Act 3. OpenAI — Product Designer, Identity
- Posting: Explicitly names human-to-agent and agent-to-agent identity, authorization, access, representation, and stronger confirmation. No public hiring manager identified.
- Gate vs. preference: Authorization and access control are core. Screens for agent-workflow orchestration through identity and authorization.
- Lead with: Claims Ledger when built — claim-evidence binding, Authorized vs. Committed vs. Settled status, and the unknown primitive map directly to identity and authorization surfaces. Carrier IQ's Approve Bind gate as current inspectable proof.
- Supporting: Red Cross (role-specific authority views, identity verification, payment authorization), Delegation Contract concept when built.
- Gap: The Claims Ledger and the Delegation Contract do not exist. Current evidence for authorization design is limited to Carrier IQ's approval gate and Red Cross's supervisor controls.
Act 4. Anthropic — Product Designer, Evals & Prompts
Production-quality Python is an on-entry minimum, not a preference. Unless Juno can demonstrate production-quality Python, this posting screens her out at the qualification boundary. Misclassifying this as an evidence gap and building artifacts to address it wastes time.
- Posting: Production-quality Python, evaluation-pipeline experience, sandboxed-tool experience, and comparable-run methodology are explicit on-entry minimums.
- Gate vs. preference: Python proficiency is a qualification boundary — not a preference, not a proxy.
- Carrier IQ demonstrates evaluation-adjacent interaction judgment, but the posting requires code authorship, not design-of-evaluation-surfaces.
- Relevant only if gate clears: Correction Lineage artifact would map directly to comparable-run methodology.
Positioning Rules
AI-native audience: Lead with frontier artifact or original interaction model. Inspectable Agentic Labs and production case studies second. TinyFish as past-tense production grounding only. Current constraint: No frontier artifact has shipped. AI-native outreach currently leads with Carrier IQ and the Trust essay, which positions Juno as a trust-architecture designer with production foundations — not as a frontier builder with published original work. This remains the portfolio's primary positioning gap.
Enterprise audience: Lead with production scale and mandate evidence — Alibaba for platform scale, Thermo Fisher for multi-organization regulated systems. Trust essay or frontier artifact as transferable method second.
Regulated or high-consequence audience: Lead with Thermo Fisher or Red Cross (mandatory BCG DV attribution). Carrier IQ demonstrates current verification and approval mechanics. Trust essay supplies the general framework.
Evidence-order rule: Published, inspectable proof leads asynchronous outreach. Accurately labeled original work in progress can lead a live review. TinyFish supplies practitioner context but cannot be relabeled as portfolio evidence.
Deep-linking rule: Any outreach referencing Agentic Labs must include direct URLs (e.g., carrier-iq.junochen.com). Labs are not discoverable from homepage navigation, and the site has no working sitemap. A link to junochen.com alone will not lead a reviewer to Labs work.
Anticipated Interview Questions — Inference Economics Cycle
| Question | Answer Evidence | Confidence |
|---|---|---|
| "How would you design a surface that helps users understand what a routing decision costs them?" | Inference-Aware Decision Surface specification (consequence-first grammar from Issue 8). Carrier IQ's evidence-disclosure patterns as structural precedent. | Conceptual only. No built artifact. |
| "How do you think about model selection as a design decision?" | Trust essay's Decision Gate handoff. Inference-economics vocabulary (reasoning effort, verification burden, governing objective as design parameters). | Carrier IQ demonstrates decision-gate design but does not involve model selection. Connection is structural, not demonstrated. |
| "Show me something you've built where the AI might be wrong and the user needs to know." | Carrier IQ: coverage deltas, confidence in review classifications, evidence provenance, what the system did not find. Brand Pulse: source-labeled cards. Retail Velocity: confidence-labeled ranked results. | Strongest current evidence. All three Labs demonstrate this. Must provide direct URLs — Labs are absent from homepage navigation. |
| "How do you design for variable inference costs in a product with a fixed user-facing price?" | The Inference-Aware Decision Surface specification addresses this conceptually (acceptable spend as a constraint, governing objective disclosure). | No built artifact and no citable production experience. TinyFish context is relevant but cannot serve as portfolio proof. Honest answer required. |
| "Walk me through evaluating whether a cheaper model is actually cheaper when you account for error rates and human review." | Verification burden concept. Thermo Fisher's exception-first operations as analogous pattern. | Analogy, not direct evidence. Lead with: "I haven't built this yet. Here's how I'd structure the evaluation." |
| "How do you design authorization flows for agent-to-agent interactions?" | Claims Ledger specification (Authorized vs. Committed vs. Settled). Red Cross supervisor/payment controls as production precedent for role-specific authority. | Claims Ledger is specified, not built. Red Cross demonstrates human authorization chains, not agent-to-agent. |
| "What's your experience writing production code or working directly in code with AI tools?" | No public evidence. No authored production code, no public repositories, no named prototyping tools across all published case pages and Labs. | Portfolio-wide gap. Risk factor at Anthropic Safeguards. Disqualifier at Anthropic Evals & Prompts. |
Prohibited Content
Do not generate materials that:
- Present TinyFish work as portfolio proof or expand beyond the public teaser
- Omit BCG Digital Ventures attribution for CS-02 through CS-05
- Use timing claims for Carrier IQ
- Present the Inference-Aware Decision Surface, Claims Ledger, Correction Lineage, or Delegation Contract as built, in-progress, or published
- Claim code authorship or prototyping-tool proficiency without evidence
- Substitute the homepage's "40M buyers" for the Alibaba case page's "25M+ desktop sessions"
- Imply that Labs are navigable from the homepage or link to the homepage expecting a reviewer to find Labs
- Present Cummins/ZED Connect as a full case study or source of metrics
- Fabricate a hiring manager name for any posting
- Present the 47% reactivation figure for Allē without clarifying it refers to expiry-notification open rate
- Anthropic's Project Swap study: The controlled book-trading experiment found that 85% of the gap between observed and optimal market outcomes came from the agent misrepresenting preferences, not from bargaining failures — which reframes the design problem for mandate-setting surfaces away from negotiation mechanics and toward intake quality.
- NIST agentic identity implementation: NIST's planned DevSecOps Build 3 will combine agentic software development with agent identity and authorization work, potentially establishing government-grade vocabulary that OpenAI Identity and similar roles will need to account for.
- Causal explanation evidence from Nature: A peer-reviewed autonomous-vehicle study found that causally grounded explanations improved mental models for 27 of 30 non-experts, but only when events were surprising — a distinction that could sharpen how Carrier IQ and Brand Pulse present evidence provenance versus causal reasoning.
- OpenAI agent lifecycle documentation: OpenAI's Agents API docs now explicitly state that a completed conversational turn does not guarantee tool success and that deleting a session does not stop its execution environment, giving concrete technical grounding for the warranted-status claims the Claims Ledger specification was designed to address.