Data
All that we are is data processing engines fueled by calories and sleep?
What can you validate to be true with 100% certainty? The foundations of domain expertise is understand data footprint and how data is moved and transformed and places where it persists and how to eliminate friction.
Control Loop
Need to start with a clear intention, no enough to know what good looks like and how to create and distribute it.
UPSTREAM INPUTS
- Perceive (Broaden Horizons)
- Interpret (Curious Mind)
- Prioritize (Decision Discipline)
- Commit (Focus with Bets, Build on Standards)
- Learn (Feedback Loop)
DOWNSTREAM OUTPUTS => OUTCOMES
Context Graphs
Data supports the context graph. It gives progress named things, current states, evidence, consent, and history. It gives the scoreboard readings that can be checked instead of believed.
Agent use: when a progress claim depends on a record, ask what the data proves, where it came from, who can use it, and what would make it stale. Trustable data is not more data. It is state data with a source, permission, timestamp, and audit path.
The Moat
The model is the commodity; the moat is the data and the taste to steer it. In the Y Combinator conversation above, Webflow co-founder Bryant Chou makes the case concretely:
- Data is the alpha. The teams that win are "on the right side of model development" — a little steering plus a lot of data. Feed a model your structured and unstructured data and let it cook. The defensible asset is the proprietary data stream (traffic, search console, CRM, analytics) and the permission to use it — not the model everyone else can also call. [source: YC — Bryant Chou, The Light Cone]
- Expertise directs the intelligence. "You need a certain amount of expertise to know what to do with this boundless intelligence." Domain time becomes leverage, not baggage. Taste, examples, and defaults turn raw model power into useful output.
- Purpose-built beats general-purpose at the outcome. General models are good at everything; a customer wants something opinionated that achieves their outcome. That thin, opinionated, data-fed steering layer is the AI harness — the same pattern pointed at one domain.
- Authority is earned, not manufactured. "You can't create 100 websites and expect Google to think you're an authoritative source." Volume is not trust. This is the trust score below in one line: a signal counts only in proportion to how hard it is to fake. Content for content's sake is slop; a trust score gamed for its own sake collapses under audit.
The Pattern
Every contact is a thing. Every thing has states.
Privacy does not limit the state machine. It shapes it into something better.
The trust ladder from the progress page is the skeleton. This page adds the muscle.
The Trust Score
Trust isn't binary. It accumulates from signals, each weighted by how hard they are to fake.
Signals That Build
| Signal | Weight | Justification |
|---|---|---|
| Registry verification | Low | Public record — anyone can check |
| Opt-in consent | Medium | Active choice to engage |
| Identity verification | High | Cost and effort to confirm |
| Completed transaction | High | Money exchanged — real commitment |
| Referral that converts | Very High | Staking reputation on someone else |
| Commitments kept over time | Highest | Only time can produce this |
Signals That Erode
| Signal | Weight | Justification |
|---|---|---|
| Email bounce | Low | Could be technical |
| Unsubscribe | Medium | Active withdrawal of attention |
| Fraud flag | Critical | Immediate freeze |
| Broken promise | High | Debit entry in the ledger |
| Referred contact who damages | Very High | Network quality proven poor |
Score Thresholds
| Range | Status | What It Unlocks |
|---|---|---|
| 0-20 | Public record only | View in directory |
| 21-40 | Enriched | Outreach eligible |
| 41-60 | Engaged | Personalized communication |
| 61-80 | Trusted | Collaboration and referral access |
| 81-100 | Credible | Full network access, commission tiers |
The score is a derivative of behavior over time. Gaming it requires sustained genuine contribution — which is the point.
The Legal Frame
Universal principles. Not jurisdiction-specific statutes — the patterns underneath them.
| Principle | What It Means | Why It's Good Architecture |
|---|---|---|
| Data minimization | Only collect what you need | Forces quality over volume |
| Purpose limitation | One purpose per consent | Forces clear thinking about why |
| Right to erasure | Delete on request | Forces earned presence |
| Consent ladder | Permission before extraction | Forces trust before data |
| Legitimate interest | Public records are fair game | Gives you a starting point without asking |
| Data portability | People own their data | Forces you to be worth staying with |
The privacy law isn't a compliance burden. It's the specification for a trust system. Every principle maps to an architectural constraint that makes the system better, not worse.
Master Data
The telco pattern: one master record, many organizational views.
| Layer | What It Holds | Who Owns It |
|---|---|---|
| Master record | Identity, verification status, trust score | The platform |
| Org-specific record | Relationship state, deal history, permissions | Each organization |
| Consent record | What was agreed, when, for what purpose | The individual |
How It Works
The master does not own the relationship. It owns the identity.
Organization A verifies a contact. Organization B trusts that verification. Neither duplicates the identity. Both keep their own relationship state.
This is the common reference data pattern. Verified once, linked many times. The master delegates, never duplicates.
The Referral Engine
Privacy-preserving referral design:
| Mechanism | How It Works | Why It Matters |
|---|---|---|
| Unique referral links | Track clicks, not contacts | You never see who didn't convert |
| Referrer-driven sharing | They send invites, not you | You never touch non-opted-in data |
| Hash matching | Pseudonymized conversion tracking | Privacy preserved through the funnel |
| Conversion rewards | Commission on conversion, not invitation | Anti-spam by design |
| Tiered credibility | High-trust referrers earn more | Quality referrals compound |
The whole world could be your sales team — if you earn it. The referrer stakes their reputation on each introduction. The platform rewards the stake when it converts. Spam is structurally impossible because the referrer bears the social cost of bad introductions.
The DePIN Bridge
How attestations make trust portable:
| Whareroa | Trust System | Function |
|---|---|---|
| Commissioning signoff | Verified credential | Proof of identity, portable between organizations |
| Maintenance log | On-chain relationship history | Ongoing proof of trust earned |
| PLC logic | Smart contract | Access rules enforced by trust level |
| Site store | Credential registry | Where verified identities are accounted for |
The PLC parallel: commissioning signoff becomes an on-chain attestation.
When Organization A signs off on a contact's identity, that attestation travels. Organization B can trust it without re-verifying.
Verification is paid once. Trust gets reused.
Context
- AI Harness — The opinionated steering layer that turns commodity models into domain outcomes, fed by this data
- Purpose — Why data must serve agency instead of extraction
- Progress — The state machine that tracks everything
- Credibility — Trust as the metric for agency
- DePIN — Infrastructure on verifiable rails
- Goodwill — Generosity without expectation that compounds into trust
- Persuasion — Earning the right to be heard
Failure Modes
- Collecting data without a named purpose turns context into liability.
- Trust scores become theater when the signals are easy to fake.
- Shared identity becomes surveillance if consent and relationship state collapse into one owner.
- Referral systems become spam when the referrer bears no reputation cost.
Questions
How do you measure whether someone deserves your data — and whether you deserve theirs?
- At what trust level does a contact become more valuable than the cost of maintaining their record?
- If the referrer sends the invite, who owns the relationship — the referrer, the platform, or the prospect?
- What breaks first when you scale trust scoring across organizations — the data model or the incentives?
- The legal frame says "delete on request" — what happens to the trust score when someone exercises that right and comes back?
Changes my mind: A trust system can preserve agency at scale without source, permission, timestamp, and audit path.
Next question: What is the smallest data record that proves trust without collecting more than the relationship needs?