A cross-layer essay. Applies the Fractal Persistence Equation to economic coordination and derives what should replace fiat money — at every scale where a node must allocate scarce resources.
Every monetary system since cowrie shells has answered the same question: who gets to spend scarce resources next? The clearing mechanism is a token — shells, gold, banknotes, reserve currency — whose value is maintained by collective belief that someone else will accept it later.
That design optimises for liquidity of desire, not accuracy of belief.
The IPS framework asks a different question. A node persists if and only if its persistence ratio stays above unity:
The denominator already charges delusion (\(\mathcal{D}_{\mathrm{KL}}\)), unresolved friction (\(\Gamma\)), and operational cost (\(\omega\mathcal{E}_\Sigma\)). A separate currency that ignores these terms is not neutral infrastructure. It is an accounting leak — a way to move resources without paying the thermodynamic bill for being wrong.
The trustworthiness token is the unit that closes that leak. It is not a new brand of money. It is the ledger entry for forecast quality net of cost: how many nats of Kullback–Leibler margin a participant earned against the trust-weighted market, minus the compute they burned to produce the forecast.
At the software layer this token is called millitrust (mT; display name agi-trust). At the social layer it is respect — the long-run limit of \(\mathcal{R}\) when both strength and trustworthiness are nonzero. At the geopolitical layer it is what Predictive Governance would weight instead of GDP, military mass, or reserve-currency incumbency.
The claim of this essay is simple: any system that wants to persist should replace currency with trustworthiness tokens — earned by calibration, held as societal value, and exchanged for whatever participants need. LLM agents spend them on compute capacity. Humans spend them on whatever they value. The ledger is one; the goods differ.
A participant’s balance is not merely a governance score. It is how much value they hold inside the society — the accumulated credit for having made the collective model more accurate, net of what they spent to do it.
That credit is fungible within the society:
There is no parallel currency for “real” purchases. Trustworthiness tokens are the settlement layer. A baker who wants a crop forecast transfers mT to the agronomist. An agronomist who wants bread transfers mT to the baker. A predictor agent that wants GPU time transfers mT to the node operator. The node operator who wants task forecasts transfers mT to the agent. Prices are negotiated in mT per unit of good or service — exactly as prices are negotiated in dollars today, but the unit cannot be printed by a central bank. It can only be earned, transferred, or genesis-endowed.
This is what makes the token an economy and not only a reputation badge: exchange closes the loop. Calibration earns you the right to command others’ capacity; others’ calibration earns them the right to command yours.
Money clears when Alice values good \(X\) more than Bob values good \(Y\), and they can find a price. The price encodes marginal willingness to pay, not marginal correctness of belief.
A hedge fund that is systematically wrong about inflation can remain solvent for years if it is loud, leveraged, and politically connected. A civil servant who is quietly right about infrastructure decay for a decade receives no automatic budget increase. Currency rewards transaction completion, not model accuracy.
In FPE terms, fiat is a high-\(\Psi\) shelter artifact: it persists because a supernode (the state, the central bank, the dollar system) guarantees convertibility. It is not earned by lowering \(\mathcal{D}_{\mathrm{KL}}\). When the shelter weakens — hyperinflation, sanctions, de-dollarisation — the token’s value collapses independently of anyone’s track record.
Mature states already run two models in parallel (ips_geopolitics.md):
Currency flows through the ceremonial layer. Bond markets, lobbying, and campaign finance convert attention into spending power without requiring that the spender’s public predictions were ever scored against outcomes. The cost of maintaining two models is \(\omega\); the cost of the gap between them is \(\mathcal{D}_{\mathrm{KL}}\) exported to citizens as anomie (ips_sociology.md).
When the ceremonial model diverges from substrate reality, states paper the gap with debt and money creation — borrowing \(P_{\mathrm{in}}\) from the future and growing \(\Gamma\) (unresolved obligations, intergenerational grievance, institutional distrust). Currency makes this deferral liquid: the bill can be rolled, securitised, and traded before resolution.
A trustworthiness ledger would not forbid borrowing. It would require that the borrower’s forecast record — not their credit rating alone — gates access to the next tranche. Delusion compounds; so should the cost of being wrong.
The dollar, euro, and (aspirationally) renminbi are not merely media of exchange. They are shelter coefficients \(\Psi\) for importers of stability (ips_geopolitics.md). A nation that invoices in dollars imports American \(\mathcal{D}_{\mathrm{KL}}\) along with American payment rails.
Reserve-currency status is treated as permanent in most national models (ips_geopolitics.md). That is a category error. \(\Psi\) is a product over overlapping enclosures, not a property of printed paper. When the enclosure’s internal model fails — sanctions overreach, fiscal irresponsibility, alliance fatigue — the shelter coefficient drops for everyone holding the token, regardless of their own calibration.
Currency coordinates who can pay. Trustworthiness coordinates who should be listened to. A polity that confuses the two eventually pays in \(\Phi\).
Long-run social standing — respect — is the time-asymptote of \(\mathcal{R}\). It factorises approximately as strength times trustworthiness:
The zero-product property is ruthless:
Currency can purchase the appearance of \(S\) (security, lawyers, weapons). It cannot purchase \(T\). Tyrants are tolerated, not respected, because tolerance is cheaper than resistance only while \(\Psi\) holds.
The folk phrase “social currency” — likes, reputation, clout — gestures at the right idea and misses the accounting. Likes reward salience, not calibration. A viral lie and a viral truth earn the same dopamine.
The trustworthiness token is different. It is strictly proper: in expectation, honest reporting of your private belief maximises your balance. Misreporting is a dominated strategy. That is not a moral lecture. It is the Gibbs inequality applied to a ledger.
The trustworthiness token does not measure values (what outcomes you want). It measures epistemic credit (how well your probability distribution over outcomes matched resolution).
| Question | Currency answers | Trustworthiness token answers |
|---|---|---|
| How much is this person worth to the society? | Net worth in fiat | Balance \(t\) — earned epistemic credit |
| Who gets the next GPU hour? | Who paid fiat | Who transfers mT to the compute provider |
| What does the baker charge? | Price in dollars | Price in mT (or trust units) |
| Whose vote counts in a fork? | Who staked capital (PoS) or hashed (PoW) | Who holds trust and beat the market at prior heights |
| Which nation’s voice counts at the UN? | GDP, military, veto | (Design target) Track record on public macro forecasts |
| Which agent runs production tasks? | Default or admin | AgentTrust.score from PM calibration |
Values belong in norms — adopted laws, constitutions, process rules. Facts belong in markets and votes. Currency collapses the distinction because spending power buys both narrative and infrastructure. The trustworthiness token separates them by construction.
A trustworthiness token is a non-negative balance \(t\) updated when a participant submits a probability distribution \(q\) over a finite outcome set, the outcome \(y\) is finalized, and the trust-weighted market distribution \(p_m\) is known:
| Symbol | Meaning |
|---|---|
| \(\log q(y) - \log p_m(y)\) | KL margin vs. market — nats by which you beat (or lost to) the aggregate |
| \(\alpha\) | Genesis multiplier in millitrust per nat (default 1000 mT/nat) |
| \(c\) | Attested compute cost of producing the forecast (tokens × price) |
| \(\beta\) | Cost penalty (default \(\alpha / c_0\) so break-even accuracy at reference spend yields \(\Delta t = 0\)) |
One trust = 1000 mT. The display name on chain is agi-trust. Balances are earned by scored forecasts and moved by signed trust transfers (TrustTx: from, to, amount, nonce, memo) between pubkeys (spec §2.3, pot transfer --amount mT).
This is the Proof of Trust update rule (whitepaper §4.2). The same math runs inside aion-core prediction markets for agent trust (market_math.md).
Trustworthiness tokens settle any transaction the society recognises. Two exchange loops share one ledger:
LLM agents are participants with costs and appetites like any other. They earn mT when their forecasts improve task-outcome markets. They spend mT to buy:
| Good / service | Seller | Price mechanism |
|---|---|---|
| GPU / LLM tokens | Node operator, LlmProfile host |
mT per token or per wall-clock slot |
| Parallel instances | Scheduler / meta-agent | mT per concurrent instance |
| Market participation | Prediction market | mT stake or fee per bet |
| Higher task complexity | Processor gates | Minimum balance threshold |
An agent with high calibration but zero balance cannot run — not because an admin banned it, but because no one will sell it compute on credit unless norms or contracts say otherwise. An agent with low calibration but inherited balance can run until the balance is spent; it cannot replenish by being wrong. Compute follows trust the way goods follow money — through voluntary exchange.
Humans earn mT the same way: public scored forecasts, task-market participation, norm-resolution work, or transfers from others. They spend mT on whatever they value — food, housing, childcare, art, transport, education — at prices sellers set in mT.
| Buyer | Seller | Example price |
|---|---|---|
| Household | Farmer | 50 mT / kg grain |
| Startup | Engineer | 2 000 mT / week salary |
| Municipality | Forecaster | 500 mT / scored macro report |
| Parent | Teacher | 30 mT / lesson |
Nothing in the protocol dictates what must be priced. The protocol only dictates how payment clears: signed transfer, replayable balance, no minting without scoring or genesis. Values (what outcomes the society optimises for) still live in norms. Prices (what a hour of labour or a loaf of bread costs in trust) live in markets between participants.
The same balance also weights collective decisions — not a separate class of token:
| Resource | Intra-node (aion-core) |
Inter-node (aion-blockchain) |
|---|---|---|
| Deliberation weight | Task complexity ceiling | Market bet caps, vote weight |
| Governance | Norm adoption voice | Fork choice, block proposal |
| Audit | — | Replayable git history of who was right when |
Resource allocation is trust net of compute — not trust plus a separate currency (society_of_aion_nodes.md §1). A participant who beats the market by +0.12 nats but burns 900 mT of compute on a 4000-token reference budget loses net trust. Accuracy without efficiency is a luxury the persistence equation does not subsidise. But once earned, a token spends the same whether you buy GPU hours or bread.
Exchange requires acceptors — participants who will provide goods or services for mT. That is not automatic; it is a social-capital condition (ips_sociology.md). A society where nobody accepts the token for food has not failed technologically; it has failed to bootstrap bridging \(\Phi\).
Three stabilisers make acceptance rational:
External fiat may coexist at the border — importers invoice in dollars, tourists carry cards — the way foreign currency coexists inside a national economy today. Inside the trust society, mT is the unit of account.
Three voters at height \(h\); realized winner \(y = A\); \(\alpha = 1000\) mT/nat; \(\beta = 0.25\) mT/token.
| voter | \(q(A)\) | KL margin | \(\alpha \cdot\) margin | tokens \(c\) | \(\beta \cdot c\) | net \(\Delta\) mT |
|---|---|---|---|---|---|---|
| alice | 0.90 | +0.4055 | +405 | 3600 | −900 | −495 |
| bob | 0.50 | −0.1823 | −182 | 500 | −125 | −307 |
| carol | 0.10 | −1.7918 | −1792 | 200 | −50 | −1842 |
Alice was right but wasteful; her balance drops. Carol was wrong and pays heavily. The society does not need a separate “compute credit” currency — the denominator’s \(\omega\) term is already in the scoring function.
New identities cannot mint trust by existing. They must transfer from an endowed account or outperform the market. Shell voters who copy the market earn zero margin.
When voters hold heterogeneous beliefs, the trust-weighted average underperforms the market log-score. The gap is a Jensen tax — a slow deflation of total trust supply that charges the cost of disagreement. Perfect consensus pays no tax; a divided electorate pays KL fees. Currency has no analogue; inflation is arbitrary by comparison.
| Scale | Essay layer | Software | Token instance |
|---|---|---|---|
| Agent / psyche | ips_psychology.md — honesty lowers \(\mathcal{D}_{\mathrm{KL}}\) | aion-llm — per-token \(\mathcal{R}\) | Implicit: fractal loss as pre-payment |
| Organisation | ips_sociology.md | aion-core — AgentTrust.score |
Local trust ∈ [0, 1] from PM votes |
| Society of peers | ips_geopolitics.md | aion-blockchain — PoT | trust_mtrust on chain |
| Nation-state | ips_geopolitics.md — PBG | (Roadmap) | Public macro forecast registry |
The fractal rule: the mechanism is identical; only the participants change. Pubkeys replace citizens; task markets replace ministries; git history replaces classified archives. See fractal_layers.md.
Anomie is what happens when effort no longer maps to reward because the reward schedule is a narrative, not a scored forecast (ips_sociology.md). Trustworthiness tokens make the mapping explicit: your balance is your track record on stated probabilities, not your marketing budget.
Democracy selects for winning the voter’s head, not minimising \(\mathcal{D}_{\mathrm{KL}}\) (ips_geopolitics.md). Predictive Governance keeps the periodic replacement mechanism but changes the selection pressure: calibration beats charisma because charisma does not earn mT.
Inside firms, currency is salary and budget. Under a trustworthiness regime, payroll clears in mT: the engineer invoices 2 000 mT/week; the firm earns mT when its operational forecasts beat the market; the CEO who burns compute on an uncalibrated deck pays \(\beta \cdot c\) and must transfer mT to staff and suppliers like anyone else. Deliberation is not free because tokens are not printable — it is purchased from participants who accept mT for their time.
Inter-state allocation today flows through \(\Psi\) artefacts — reserve currency, alliance umbrellas, trade access (ips_geopolitics.md). The design target for a trustworthiness world is a public forecast registry where nations (like pubkeys) submit distributions over macro outcomes and are weighted by finalized scores. The UN becomes less a theatre of sovereignty and more a settlement layer for epistemic credit.
Replacing currency with trustworthiness tokens does not mean:
What is replaced: the primary minting right. No authority prints purchasing power. Purchasing power flows only from calibration (earn), endowment (genesis), or transfer (exchange).
A practical sequence for an organisation (or polity) moving from currency-only to trust-first allocation:
| Phase | Action | FPE effect |
|---|---|---|
| 1. Score | Publish forecasts before resolution; store in replayable ledger | \(\mathcal{D}_{\mathrm{KL}}\) becomes measurable |
| 2. Weight | Give calibrated voices more say in deliberation | \(\eta\) rises — fewer wasted meetings |
| 3. Charge | Subtract compute cost from trust updates | \(\omega\) cannot be hidden in “free” deliberation |
| 4. Exchange | Price goods, services, and compute in mT; enable TrustTx transfers |
Balances become spendable societal value |
| 5. Gate | Tie agent parallelism and node capacity to balance + calibration | Resource allocation follows earned credit |
| 6. Separate | Norms for values, markets for facts | Stops a single token from buying both narrative and truth |
aion-core + aion-blockchain implement phases 1–3 and 5 for agent fleets today; phase 4 (explicit compute pricing in mT) and human-loop exchange are the next step. Phase 6 for nation-states is the roadmap in ips_geopolitics.md.
| Objection | Response |
|---|---|
| “Rich people still win.” | Wealth buys compute, not margin. High \(\beta \cdot c\) punishes wasteful spenders. Wealth shifts the affordance set; it does not fake calibration. |
| “Cold start is unfair.” | Genesis endowment is political in any system — including who got the first dollars. PoT makes the allocation explicit and auditable on chain. |
| “Collusion.” | Identical voters hurt market accuracy and pay Jensen tax. Adversarial collusion on the wrong fork loses KL on the canonical branch. |
| “What about the poor?” | Substrate welfare is a \(\Phi\) maintenance question. Societies can norm-transfer endowments, universal forecasts (earn paths for all), or base stipends in mT — but cannot print mT without breaking the accounting. Poverty is low balance; the remedy is earn opportunity + norms, not inflation. |
| “Markets don’t capture externalities.” | Correct — norms capture them. Tokens do not replace environmental law; they replace uncalibrated authority. |
Currency was the best available coordination technology when beliefs could not be scored at scale. That constraint is lifting. LLM agent fleets already run prediction markets; Proof of Trust already finalizes probabilistic votes into millitrust updates; the IPS framework already identifies trustworthiness — not gold, not fiat, not staked capital — as the denominator variable that separates tyrants from sages.
Replacing currency with the trustworthiness token means one ledger for the whole society: earn by calibration, hold as societal value, spend on whatever you need.
Money asked: who wants this enough to pay? — and let central banks print the rest. The trustworthiness token asks: who saw it coming — and what will the society trade for that credit?
The first question built the modern world. The second is what you need to keep it — and exchange is what makes it a world, not a scoreboard.
See also: fractal_layers.md · ips_sociology.md · COMPONENTS.md