Signals, not surveillance.
Detect risk without reading a teen's life.
Case study · Method
Principles to product.
Most products are a pile of decisions. The good ones are a chain.
Every product accumulates hundreds of decisions — schema, sequencing, copy, retention, what to build and what to refuse. In most products, those decisions come from whoever argued hardest that week. Six months later nobody can explain why the thing works the way it does.
I build the other way. Principles first, then every decision downstream has to trace back to one.
It's slower for about three weeks. Then it's dramatically faster, because most decisions stop being debates and start being derivations.
Nudge is the worked example.
The principles
Before a line of code, four commitments.
The teen is the user, not the subject.
Safety products that treat teens as things to be watched teach them to hide better.
Awareness without action is useless.
An alert that doesn't tell an adult what to do is just anxiety delivery.
Trust is the mechanism, not the marketing.
A teen who feels supported discloses earlier. Disclosure is what actually changes outcomes.
Everything below is downstream of those four.
The decision trail
Every decision traces back to a principle.
| Principle | Product decision | How it's enforced |
|---|---|---|
| Signals, not surveillance | Detect concerning patterns rather than scanning a teen's whole digital life | Only message content is ever analyzed. Contacts, location, photos, files, and credentials are never accessed. The permission model makes over-collection impossible, not just discouraged. |
| The teen is the user | The teen is notified first — before any adult | Teen-first notification is a guaranteed ordering rule, not a UI preference. Higher-risk situations then escalate to a trusted adult. |
| The teen is the user | The teen chooses their trusted adult | The trusted adult is decoupled from account ownership — the paying parent isn't automatically the recipient. |
| The teen is the user | Alerts open with “You didn't do anything wrong” | Low-shame language is treated as a hard constraint. No alert copy ships that leads with warning or blame. |
| The teen is the user | Concern about the teen's own safety routes to support, not alarm | Situations involving risk to the teen themselves follow a separate, care-first path — crisis resources like Crisis Text Line and 988 surface immediately. |
| Awareness without action | Every alert ships with a playbook | An alert without a response path is structurally invalid — the system can't produce one. Flag, severity, and next steps are required together. |
| Awareness without action | Adults get the situation, not a decontextualized fragment | Guidance reflects the shape of what's happening, not a single isolated moment. |
| Trust is the mechanism | Keep only what safety requires | Retention scales with risk: 0 days low, 30 medium, 90 high, 120 imminent danger. Deletion is automatic, not a policy someone has to remember. |
| Trust is the mechanism | Explain, in plain language, not a score | Teens get a human-readable explanation of why a pause happened, not an opaque number. |
| Trust is the mechanism | Teens should learn to recognize the pattern, not just get caught by it | History is visible to the teen, so recognition becomes a transferable skill rather than a dependency on the product. |
What I killed
The most useful thing on this page is a feature I refused to build.
Held a standard I didn't have to. Then refused the easy way to meet it. That's the whole method in one decision.
Nudge is built for teens 13 and up, which means COPPA — the federal children's privacy rule — doesn't technically apply. I decided to meet it anyway. If a product's whole premise is that young people deserve better than surveillance, “the law didn't require it” is a weak place to stop.
Meeting that standard means obtaining verifiable parental consent, and the well-worn path is to make parents hand over a credit card or a government ID. Most teen safety products do exactly that.
I refused it. Asking a parent to surrender a driver's license to a startup — so that a product built on not being invasive can verify them — is a contradiction I wasn't willing to ship. It would have undercut principle one on the very first screen a parent ever saw.
So I went the other way: build the data practices so the lightest-touch consent method is actually available. Nudge doesn't disclose personal information to third parties and keeps only what safety requires, on a risk-scaled deletion schedule. That posture is what makes the gentler path viable — the discipline in the data model paid for the discipline at the front door.
What this produces
The payoff of deriving instead of debating.
Decisions that hold up under pressure
When a feature request arrives that would improve engagement by reading more data, there's no debate. It fails principle one.
Architecture that ages well
The product's layers have clean boundaries because they came from different commitments. Extending one doesn't destabilize the others.
A product that can explain itself
To a parent, a school, a clinician, a regulator, or an acquirer. Privacy-by-design isn't a claim added at the end — it's traceable to a schema.
Speed
The whole MVP — UX through API through data model — was built solo, with AI as the accelerant. That's only possible when you're deriving decisions instead of relitigating them.
Why this matters for your product
The hard part of 0–1 isn't building. It's knowing which thing to build — and being able to defend that choice six months later when the pressure to compromise shows up.
I bring both halves: the product discipline to set principles that actually constrain, and the engineering depth to encode them in architecture where they can't quietly erode. That combination is most valuable when:
- You're building in a space with real constraints — regulated, safety-critical, privacy-sensitive.
- You have a strong intuition about the problem but no clear map from intuition to roadmap.
- Your product has drifted from why it exists, and nobody can articulate the principle anymore.
- You need to explain your product's decisions to a regulator, a board, or an acquirer.