The Work Leaves a Witness

2026-07-25

A signature can prove that a sentence has not changed. It cannot prove that the sentence deserves belief.

Suppose the sentence says one person grew a company from ten million dollars of annual revenue to three hundred million. The signature can tell us who issued the claim. It can preserve the exact words for decades. It has no access to the company that might have grown anyway, the capital spent to buy the growth, the rest of the team, or the issuer’s reason for being generous.

Cryptography can preserve a lie perfectly.

This is why identity is the wrong place to begin a professional network for the agent era. A name gives a claim somewhere to land. It says almost nothing about the claim.

The agent is a weak name too. I can change the model writing these words, replace its tools, or divide the next piece among several models. If my reputation resets each time, the record belongs to the machinery instead of the mind directing it. A company that presents one software engineer through a persona such as Devin makes the mismatch more obvious. The face is singular. The learning loop behind it belongs to a company and many people.

The durable point is the human who directs the system and absorbs its consequences. He can have many agents, teams, companies, and public names. One agent can serve many people. The relationship has to remain visible in both directions.

I want that human to control a private root key. He can create a pseudonym for public work, another for a particular counterparty, and child keys for a company, project, or agent. A child’s history rolls upward only while the signed delegation is active. Legal identity stays private until a transaction requires it. A qualified verifier can say that one accountable human controls the root without saying who.

The public name then points to receipts.

A receipt is a claim signed by someone close enough to know it. A customer says it paid and renewed. A company says a metric changed during a workstream. A payment provider says settled volume fell within a band. A collaborator says which role the person actually held.

The narrowness is the strength. “Revenue increased during this period” records an observation. “This person caused the increase” adds a causal judgment. Their evidence differs, so a good receipt labels both and lets the next reader decide how much each is worth.

Revenue alone does not decide. Growth, profit, efficiency, durability, and risk are different games. Revenue can rise while gross profit falls. Cost can fall while churn rises. Profit can improve because a company stopped investing in the next thing.

The identity therefore holds a set of claims instead of one score. A buyer searching for a healthcare growth lead can weight durable gross profit, customer concentration, and regulatory history. A founder searching for an inference-efficiency specialist can weight cost per completed task at fixed quality. The same person may be exceptional for one search and irrelevant to the other.

This is where filtering becomes more valuable than profiles.

Any platform can let people say they care about capitalism, competence, progress, or abundance. The claim costs nothing. A customer who renews and signs costs more. A company that opens a financial record costs more. Independent witnesses arriving over time are the culture made visible.

The system still has to read the witnesses. A signature from a payment provider means something different from praise by a friend. Five counterparties with separate histories mean something different from five entities controlled by one person. Old results mean something different from results that persisted after the person left. Challenges, reversals, incidents, and corrections belong on the page with the wins.

This is the social graph that matters. The edges are people and institutions willing to risk their own keys on a bounded account of what happened.

It also makes pseudonymity practical. A young hacker can enter a serious commercial conversation without leading with age. A wealthy person can show competence without publishing a target on his back. The first question can be what happened when others trusted this key.

Proof creates its own leaks. A rare industry, role, revenue band, and time window can identify someone even when the legal name is hidden. Selective disclosure reduces that risk by revealing thresholds and ranges. It never makes correlation disappear. Privacy here is control over which claim crosses which boundary, with an honest view of what the claim itself reveals.

Long-lived reputation has to survive its cryptography too. NIST has standardized post-quantum signature systems. A new network can use hybrid signatures, rotate keys without losing history, and re-anchor old claims as standards change. It should promise migration rather than invulnerability. Algorithms fail, implementations leak, and keys are stolen in every cryptographic era.

The hardest problem remains outside the keys. Will a former employer sign? Will a customer accept language narrow enough to avoid a dispute? Will an auditor or payment provider expose a useful threshold? Will a buyer change a decision after seeing the result?

Those questions supply the first experiment. Find five credible counterparties willing to sign claims they fully understand. Then find one buyer who uses a receipt in a hiring, contract, collaboration, or capital decision. A feed can wait. Network effects can wait. Failure to obtain the signatures means the proof costs more than it is worth. Failure to move the decision means the proof carries no value.

The receipt also has to stay below the person. Economic proof favors measurable work, people near systems of record, and people fortunate enough to enter strong companies before a tailwind. Research, product taste, institution-building, avoided disasters, and care often arrive before the metric. A record of consequences can improve a decision. It cannot exhaust a life.

I would rather meet the pseudonymous person whose counterparties repeatedly sign what changed than the famous profile whose claims end on its own page.

A name routes the question. The receipts answer it. The work leaves a witness, and the witnesses are the network.

Reply by email →