Scroll to the end and try every interaction → earn a Resonator Pass. 5 Passes = free attendance at an invitation-only Interference evening.
Interference topic · Why every trust score fails
What if trust is not a quantity that can be compressed into a number — but a state that context alone can carry?
A credit score decides whether you get an apartment. An ESG rating moves billions in capital. A platform reputation point determines whether anyone ever sees your work. A machine-learning model assigns you a risk label you will never see. All of these share one hidden assumption: that trust can be compressed into a single comparable number.
This paper makes a narrow but stubborn claim: that assumption breaks — not because the scores are badly built, but because trust is not the kind of thing a single number can hold.
Trust is not a quantity. It is a state — like "trusted → untrusted" or "verified → unverifiable." States change all at once, and their meaning depends on context. A number cannot carry context, time, and perspective all at the same time.
Across very different systems, the same five pressures recur. None is removed by adding more data or a smarter model — they follow from using one universal number for everything.
"Trust scales easily.
Proof does not.
And that asymmetry is dangerous."
Each of these had excellent trust proxies — top ratings, clean audits, index membership — right up until it didn't. The proxies measured reputation and reporting, not verified events.
| Case | Trusted because… | What the proxy missed |
|---|---|---|
| Enron 2001 |
Investment-grade ratings, Big Four audits, strong reputation | Audits validated narratives, not whether the transactions ever happened. |
| Wirecard 2020 |
Index inclusion, analyst consensus, compliance certificates | Years of evidence existed; institutional confidence suppressed it. |
| VW Dieselgate 2015 |
Emissions-compliance scores, lab results, eco labels | Cars were tuned to pass the test, not to be clean on the road. |
| 2008 crisis | AAA ratings, risk models, structured finance | Local mortgage events were never verified, only aggregated upward. |
| ESG ratings 2018–23 |
High composite scores across industries | Governance metrics diluted real environmental harm into the average. |
The usual fixes — more data, better AI, explainable models, transparency dashboards, "decentralised" governance — all address the implementation. But the five cracks in section 02 don't come from bad implementation. They come from asking a single comparable number to carry context, time, perspective, and resistance-to-gaming all at once.
The paper argues — informally, in the spirit of Arrow's famous impossibility theorem about voting — that these four wishes cannot all be satisfied at once. Comparability costs context. Stability costs accuracy. Universality costs neutrality. Public use costs resistance to gaming. You can privilege some, never all four.
(The paper is careful here: this is a structural argument, not a mathematical proof. Local, well-bounded scores often work fine — the target is the universal, cross-domain, high-stakes score.)
Trust exists because checking is expensive. Where checking becomes cheap and local, trust is no longer required. Instead of asking "how much should I trust this party?", a proof-based system asks "can this specific claim be checked?"
Belief-dependent · one scalar number · central authority · gameable · reputation-driven · broad reach, weak guarantees.
Belief-independent · a yes/no on one claim · local verification · constraint-bound · event-driven · narrow reach, strong guarantees.
This already works in practice: Bitcoin verifies transactions without a trusted bank; zero-knowledge proofs let you prove you are solvent without revealing your balance; supply-chain ledgers record each handoff so you check the shipment, not the supplier's reputation.
Proof is narrow. It can certify that a record was not altered, but not that it was true when entered ("garbage in → immutable garbage"). It cannot settle "is this a good manager?" Outside the bounded class of objective, checkable claims, trust remains — and the honest goal is to keep its scores local, contestable, and honest about their limits.
The whole paper is an argument against measuring trust. But there is a mirror image. The same insight — that trust is a local, lived, relational state, not a number — is exactly why trust can be built in a room, in an evening, between people. This is the basis of NeverNormal's TRUST arc, a facilitation method used in organisational settings.
Where a proof-based system "closes a claim locally," the TRUST arc closes a trust event locally — phase by phase, each one lived rather than scored:
So the two halves of the Equora work meet here: the white paper shows why trust collapses when you try to scale and score it; the TRUST arc shows what trust can do when you keep it local and lived. The same nature of trust — denied to the machine, returned to the room.
The contribution of the paper is not "abolish trust scores." It is a distinction: where a claim is objective and checkable, verify it. Where it is genuinely a matter of judgement, keep the human, lived version of trust — and keep its scores honest. The full argument, references, and the formal version live in the preprint.
◇ Read the full white paper on Zenodo · DOI: 10.5281/zenodo.19959442The research is free — and will stay free. If it matters to you, there's a way to say so.
Support the work →