Design Context File v3 — Claims Architecture

Juno Chen | August 9, 2026

Supersedes v2 (Four Altitudes)

v2 sorted evidence by craft, system, and leadership altitude. This one sorts by five falsifiable claims. If both files are loaded in a project, use this one and delete v2.

Disclosure Key

Tag	Class	What AI may do with it
`PUB`	public	Cite in any generated text. Source: junochen.com.
`CTL`	controlled-interview	Share in live conversation with context. Never in cover letters, outreach, or written artifacts.
`VRB`	verbal-only	Spoken reference only. No written artifact, no screen share, no generated text.
`UNV`	unavailable	Do not reference. Do not confirm existence.

Standing rules:

All TinyFish production artifacts (screenshots, traces, architecture, pricing): VRB. No clearance obtained.
UAT Sentinel: UNV. Visible on site due to inventory drift; never approved.
Private essay: UNV. Reachable via broken client-side gate; not authorized.
Allē scale figure: 30M+ members only. The 18M figure is retired.

Identity Block

Field	Value
Name	Juno Chen
Current role	Directing Design, TinyFish (AI productivity platform)
TinyFish constraint	Context and credibility signal for production-AI proximity. Never portfolio proof. `VRB` ceiling on all product detail.
Target role	Staff–Director product design at AI-native or regulated-transactional companies
Published pillars	(1) Trust essay — five handoffs for human-agent trust `PUB` (2) Case studies CS-01–CS-06 `PUB` (3) Agentic Labs: Brand Pulse, Retail Velocity, Carrier IQ `PUB`

Five Claims

Each claim states something Juno can do. Under it sits the evidence chain with source and disclosure class, then a boundary note marking where the proof is thin, contested, or not yet built.

Claim 1 — Bounded Authority for Consequential Agent Actions

Juno designs the controls that determine what an autonomous agent may do without human approval, and where the gate closes.

Evidence	Source	Class
Carrier IQ confidence thresholds: system acts autonomously above threshold, requests human review below	Agentic Labs	`PUB`
Trust essay handoff 1 (scope): the first design decision is what the agent is allowed to touch	junochen.com	`PUB`
Confidence rule: confidence rendered from observed input, never from model self-report	Published analysis	`PUB`
TinyFish: production agent governance, scope boundaries, permission architecture	Current role	`VRB`

Current revision. The Trust essay's original ladder implied a single trust axis. Current position: two independent dials. Autonomy scope is how much the system may do without asking. Intervention sensitivity is how sharply the human reacts when something looks wrong. The revision is live and not settled; the calibration test has not been run against Carrier IQ's confidence distribution. If pressed, say the framework is evolving and name the production observation that forced the change.

Boundary. Carrier IQ's confidence threshold is implemented. The longitudinal control record — whether the threshold holds over time — has not shipped. The claim rests on design intent and initial implementation, not production validation.

Claim 2 — Controlled Recovery from Wrong Agent Output

When the agent is wrong, Juno's interfaces let the human detect the error and reverse it before the consequence reaches downstream systems.

Evidence	Source	Class
Carrier IQ correction control: designed, demo script not yet built	Agentic Labs	`PUB` (design) / `UNV` (demo)
Trust essay handoff 4 (recovery): recovery is a designed state, not an error state	junochen.com	`PUB`
Allē: error handling in dual-surface loyalty transactions across 30M+ members	CS-06	`PUB`
TinyFish: trace-based error detection, session replay for agent behavior audit	Current role	`VRB`

Boundary. Carrier IQ's correction control is the strongest single piece of evidence here, and it cannot be shown running. Juno can describe the design in a walkthrough. She cannot demo it. The build queue identifies this demo script as the spine artifact. Until it ships, route recovery questions to Allē error states (PUB) first, then describe the Carrier IQ correction control verbally.

Claim 3 — Staff-Level Craft Within Organizational Scope

Juno produces component-level design work while running the design function. Strategy and making occupy the same week.

Evidence	Source	Class
Thermo Fisher: design system, component architecture for regulated scientific workflow	CS-02	`PUB`
Red Cross: complex multi-step form architecture	CS-03	`PUB`
Allē: dual-surface loyalty interface (consumer + provider)	CS-06	`PUB`
Alibaba: transactional interface components across 4 buyer segments	CS-01	`PUB`
Equinox+: multi-brand design system across five brands, 90-day zero-to-one build	CS-04	`PUB`

Boundary. The traditional cases are interrogable at the component level. Agentic Labs projects prove velocity, not making depth at that resolution. If the panel's real test is technical builder, the weight sits on Thermo Fisher, Red Cross, Allē, and Equinox+. Labs supplements.

Claim 4 — Trustworthy Complex Transactional Systems at Scale

Juno has designed systems where financial transactions, health data, or compliance obligations move through the interface, at verifiable scale.

Evidence	Source	Class	Metric
Alibaba	CS-01	`PUB`	$292B GMV; 20% higher daily transactions; 2.2-pt NPS increase; 47% fewer security concerns
Allē	CS-06	`PUB`	30M+ members, 40K+ practices (portfolio-published, not independently verified)
Thermo Fisher	CS-02	`PUB`	Regulated scientific procurement; $20M+ margin opportunity; 100% adoption; 83% IRR
Equinox+	CS-04	`PUB`	Five brands, 600K members, 4.8-star rating
Carrier IQ	Agentic Labs	`PUB`	Carrier billing reconciliation

Scale-claims custody rule. One claim, one evidence carrier. State the mechanism Juno designed, attribute the metric to the program, same sentence. Do not claim personal ownership of the GMV or the member count.

Claim 5 — Framework Evolution Under Production Evidence

Juno's design frameworks change when production data contradicts them. The Trust essay is a working document.

Evidence	Source	Class
Trust ladder to two-dial revision: production observation that autonomy scope and intervention sensitivity operate independently	Published analysis	`PUB`
Confidence rule: emerged from Carrier IQ implementation, not from theory	Published analysis	`PUB`
TinyFish: production agent behavior changed the framework's assumptions	Current role	`VRB`

Boundary. The revision is written but not drawn. Trust Pattern Library screens exist as prose only. The argument holds in conversation and will not survive a silent portfolio scan.

Strongest answer in the file

When a panel asks whether your thinking has changed, the ladder-to-two-dials story shows intellectual honesty and production grounding in the same move. Use it.

Case Study Reference

ID	Title	Metrics	Trust handoff	Cover letter proof block	Routing note
CS-01	Alibaba	$292B GMV; 20% higher daily transactions; 2.2-pt NPS increase; 47% fewer security concerns; 4 buyer segments	H1 (scope)	"Designed transactional interfaces across four buyer segments within a $292B GMV platform."	Active
CS-02	Thermo Fisher	Regulated scientific procurement; design system; $20M+ margin opportunity; 100% adoption; 83% IRR	H3 (transparency)	"Designed the component architecture for a regulated procurement system where every interface state carried compliance obligations."	Active
CS-03	Red Cross	$847K disbursed; 1,689 cases in first two weeks; 6 systems consolidated	H4 (recovery)	"Built the form architecture that moved disaster-response data through a multi-step flow where abandonment had human cost."	Active
CS-04	Equinox+	Five brands; 600K members; 90-day zero-to-one; 4.8-star rating	H2 (confidence)	"Led a 90-day zero-to-one build across five fitness brands and 600K members, shipping a multi-brand design system that launched at 4.8 stars."	Reserve
CS-05	—	See junochen.com	H5 (calibration)	Available on junochen.com if needed.	Not in active rotation
CS-06	Allē	30M+ members; 40K+ practices; dual-surface loyalty; 3.2x redemption; 47% reactivation; $42 CAC	H2 (confidence)	"Led the dual-surface loyalty redesign serving 30M+ members and 40K+ provider practices, where transaction errors had direct financial consequence."	Active

Trust Essay + Five Handoffs

The essay names five moments where a human-agent system must pass control, information, or consequence between the person and the machine. The two-dial model (autonomy scope × intervention sensitivity) is the current interpretive frame. When an interviewer raises the Trust essay, enter through one specific handoff and its implementation. Bring in the two-dial revision second, as evidence that the framework moved under production pressure.

Handoff	Name	Labs implementation	TinyFish production behavior	Act/Watch company relevance	Claim
H1	Scope	Carrier IQ: confidence threshold defines what the agent may act on without approval	Agent permission architecture, task-boundary governance `VRB`	OpenAI Payments (transaction ceiling), Stripe Link (payment-flow authority)	1
H2	Confidence	Retail Velocity: confidence-scored outputs; Carrier IQ: threshold rendering from observed input	Trace-level confidence surfacing `VRB`	Amplitude (data-informed decision confidence), Fieldguide (audit-record certainty)	1, 4
H3	Transparency	Brand Pulse: human review gates on agent-generated intelligence	Session replay, attribution tracing `VRB`	Vanta (compliance auditability), Fieldguide (defensible audit records)	3, 4
H4	Recovery	Carrier IQ: correction control (designed, demo not shipped)	Trace-based error detection, reversal workflow `VRB`	OpenAI Payments (transaction reversal), Gusto (payroll error recovery)	2
H5	Calibration	Carrier IQ: longitudinal threshold behavior (control record not shipped)	Production observation that forced ladder-to-two-dials revision `VRB`	Amplitude (system learning from usage data), OpenAI EA (tool adaptation to expert workflows)	5

Agentic Labs — Approved Evidence Set

Project	Demonstrates	Claims
Brand Pulse	Agent-generated competitive intelligence with human review gates	1, 2
Retail Velocity	Automated retail analytics with confidence-scored outputs	1, 4
Carrier IQ	Carrier billing reconciliation with confidence thresholds and correction control	1, 2, 4, 5

UAT Sentinel and Creator Scout are excluded. Do not reference either in generated output.

Company Routing

Act This Week

OpenAI Payments | Posted Aug 6

Design problem: Agent-mediated financial transactions where a wrong action costs dollars.
Lead with: Claim 1 (bounded authority) into Claim 4 (transactional scale).
Portfolio moment: Carrier IQ confidence thresholds as spending-limit analogs. Alibaba transaction-failure recovery at $292B GMV.
Differentiation: The two-dial model translates into payments without adaptation. Autonomy scope is the spending ceiling; intervention sensitivity is the fraud-detection threshold. Both are already denominated in dollars, which is the unit this panel reasons in.
Do not lead with: TinyFish title. Labs as portfolio proof without traditional case depth. Allē loyalty framing.

Gusto | Head of Design, Unified Service Platform | Fresh window per KB

Design problem: Unifying payroll, benefits, and HR workflows under one design function with organizational authority.
Lead with: Claim 3 (craft) into Claim 4 (transactional).
Portfolio moment: Thermo Fisher component system as making depth inside leadership scope — she designed the regulated architecture while running the function. Red Cross multi-step form architecture where completion carried operational consequence.
Differentiation: This panel is buying someone who can run the org and still make things. Prove both inside the first five minutes.
Do not lead with: AI-native framing before the organizational story. Agentic Labs before traditional cases.

Prepare — Verify Window, Then Act

OpenAI Engineering Acceleration | Posted Aug 6

Design problem: Internal tooling that accelerates engineering workflows. Expert users, not consumers.
Lead with: Claim 5 (framework evolution) into Claim 3 (craft).
Portfolio moment: Thermo Fisher tooling for domain experts, where the interface served the scientist's expertise instead of simplifying it away. Trust essay as systematic thinking applied to tool design.
Differentiation: Expert users in regulated contexts, where the interface has to respect domain knowledge it cannot replace.
Do not lead with: Consumer loyalty (Allē). Agent autonomy framing unless the role involves AI-assisted engineering tools.

Fieldguide | Staff Product Designer | High fit, 22+ days old at last check

Design problem: Audit and compliance workflows where the interface must produce defensible records.
Lead with: Claim 4 (trustworthy systems) into Claim 2 (recovery).
Portfolio moment: Thermo Fisher audit-trail architecture. Alibaba cross-segment compliance at scale.
Differentiation: Consequence literacy — she has designed systems where an interface error creates compliance exposure for someone else.
Do not lead with: AI-agent framing as primary identity. Labs velocity without regulated-domain grounding.

Stripe Link | Staff-level | Live at Aug 1, age unresolved

Design problem: Payments infrastructure where the interface mediates between consumer trust and financial-system complexity.
Lead with: Claim 4 (transactional scale) into Claim 1 (bounded authority).
Portfolio moment: Alibaba payment-flow architecture across buyer segments. Carrier IQ billing reconciliation.
Differentiation: Financial transactions and system trust, where the interface has to communicate what happened and how confident the system is about it.
Do not lead with: Organizational leadership narrative ahead of the transactional-design story.

Monitor

Amplitude — Watch window, mandate ambiguity unresolved. Lead claims: 5 into 1. Anchor with Alibaba's data-informed decision architecture. Open on how production evidence changes design decisions (Claim 5), then route to confidence-scoring mechanics (Claim 1) if the role involves AI-augmented analytics. Do not lead with agent-autonomy framing. Wait for mandate clarity before outreach.

Vanta — Watch window, mandate ambiguity unresolved. Lead claims: 4 into 2. Trust essay, Alibaba auditability, Thermo Fisher regulated workflow. Do not lead with AI-agent identity ahead of the compliance story. If Vanta surfaces before the mandate resolves, route to Claim 4 and hold on specifics until scope is confirmed.

Giga — Needs rubric scoring on the IC/leadership split and geography before any positioning is actionable. If Giga comes up in conversation before that scoring exists, default to Claim 3 (craft within organizational scope) and hold role-specific framing. Do not lead with Labs or AI-native identity until the level is confirmed.

Positioning Rules

TinyFish is context, never proof. It explains current production-AI proximity. It is never cited as a portfolio case.
Pillar routing by audience. AI-native panels: Trust essay, then Labs, then traditional cases. Regulated and transactional panels: traditional cases, then Trust essay, then Labs.
Trust essay entry protocol. Always enter through one handoff and its implementation. The two-dial revision is what proves the framework isn't theory.
One candidate, one sentence. Every piece of outreach has to leave the reader able to describe Juno to a colleague in one line that matches what the other interviewers wrote down. Positioning that needs a paragraph produces two different candidates in the debrief, and the room tends to resolve that discomfort by advancing whoever was easier to summarize.

Gap Artifacts and Workarounds

Artifact	Status	What it costs you without it
Carrier IQ demo script	Not started	Claim 2's strongest evidence stays verbal. Route to Allē error states (`PUB`) when asked to show recovery design.
Career narrative visual	Not started	"Why this role now?" has no visual support. The verbal answer has to stand alone.
Trust Pattern Library screens	Partially drafted (prose, no visual)	Claim 5 holds in conversation and disappears in a silent portfolio scan.
Trace-evidence annotation	Not started	No way to show a production loop where agent behavior changed the design. Describe TinyFish verbally (`VRB`).
Execution artifacts (design-system proof, intent-to-trace, constraint before/after, Figma-to-code)	Partially implemented (Carrier IQ evidence-run exists)	"How do you work at the component level?" routes to Thermo Fisher and Allē until the remaining artifacts ship.

Anticipated Interview Questions

Question	What it's testing	Route
"How do you decide what the agent should do on its own?"	Claim 1	Carrier IQ confidence thresholds, then the two-dial model, then TinyFish governance `VRB`.
"What happens when the AI is wrong?"	Claim 2	Trust handoff 4 (recovery), then Carrier IQ correction control (describe, cannot demo), then TinyFish trace detection `VRB`.
"Show me something you designed at the component level."	Claim 3 — making-doubt diagnostic	Thermo Fisher design system, then Allē dual-surface components. Do not open with Labs. Do not open with strategy.
"How do you handle scale?"	Claim 4	Alibaba $292B GMV: state the mechanism, attribute the metric to the program, same breath. Then Allē 30M+ members.
"Has your framework changed?"	Claim 5	Ladder to two dials. Name the production observation that forced the revision. Strongest answer in the file.
"Who did design report to in your last role?"	Enterprise-doubt diagnostic	Lead with organizational scope, then prove craft depth.
"What happened when it went wrong?"	Consequence-doubt diagnostic	Name the consequence, the recovery mechanism, and what changed afterward. Use regulated grammar regardless of who is asking.

Prohibited Output

Do not generate any text that:

References UAT Sentinel or Creator Scout by name
Quotes or paraphrases the private essay
Includes TinyFish screenshots, architecture details, pricing, or product specifics
Uses the 18M Allē figure
Presents 30M+ members or 40K+ practices as independently verified (they are portfolio-published)
Frames TinyFish work as a portfolio case study
Claims the Carrier IQ demo script or correction control is viewable
Presents the two-dial model as settled rather than current working position
Uses the phrase "design-led" to describe any target company

Things to follow up on...

CHI transparency tension unresolved: A peer-reviewed CHI 2026 study found that eight of twelve participants preferred progressive or on-demand transparency over maximal process visibility, which directly challenges the Trust essay's argument that confidence develops through visible work — prepare a spoken position on where you draw the line.
Anthropic's permission model: Anthropic's agent research on trustworthy agents frames permission as always-allow, require-approval, or block per action rather than per trust level, which maps to the two-dial revision and gives you external vocabulary if a panel asks about capability boundaries.
Vercel's codified judgment approach: Vercel now stores accepted product decisions as reviewed, agent-accessible code because coding agents can see existing code but cannot infer why decisions were made — a concrete industry example to reference when Gusto, Amplitude, or Vanta ask how you preserve design reasoning for agents.
Stripe Agentic Commerce gone: The former Senior Staff Product Designer, Agentic Commerce URL now returns a 404, so any cached search results should not be treated as evidence of an open role — verify Link's window freshness independently before outreach.