The IPS essays read psychology, sociology, economics, geopolitics, and AI as one equation. This essay names the verb they share. Every one of those domains is a way of extracting truth: pumping calibrated structure out of contact with reality by lowering delusion divergence \(\mathcal{D}_{KL}\). A grieving person, a price-discovering market, two armies over a border, and a particle collider are all running the same loop — hold a simplified approximation of reality, expose it to the world, pay to erase the bits that were wrong.
Where the domain essays ask “can this node keep \(\mathcal{R} \ge 1\)?”, this essay asks “what is the work that keeps \(\mathcal{D}_{KL}\) from eating it?” — and the answer, everywhere, is truth extraction.
Formal derivation: information_persisting_systems.md. Vocabulary: glossary.md. Scale bridge: fractal_layers.md. Why minds run on approximations: the Useful Approximations Framework (UAF) in The Universe’s Self-Awakening.
The Useful Approximations result is that no finite system touches reality directly; it survives on simplified internal models — delusions in the technical sense, maps that are never the territory but are real and useful. \(\mathcal{D}_{KL}\) is the running gap between that map and what reality returns when the map is tested.
Because \(\mathcal{D}_{KL}\) sits multiplicatively in the denominator, a wrong model taxes every joule the node spends. You cannot out-power a delusion: a 3× delusion factor makes the same numerator do a third of the work. So the central activity of any persisting system is not maximising the numerator — it is truth extraction: continuously testing the approximation against reality and paying the Landauer cost to erase the falsified bits.
Truth extraction — the activity of lowering \(\mathcal{D}_{KL}\) by exposing a model to a resolution authority and erasing the bits it falsifies. “Operation 1” in fractal_layers.md — lower \(\mathcal{D}_{KL}\) — is one local instance of it; this essay is the claim that the same verb runs every domain.
The equation splits into a numerator (what you harvest and convert) and a denominator (the tax for noise, delusion, and unsettled friction). Each half names a different human enterprise:
| Half | FPE factor | Enterprise | Object | Truth’s role |
|---|---|---|---|---|
| Numerator | \(P_{in}\,\eta(I)\) | Engineering — the art of doing | Raise \(\eta\): more useful work per joule | A constraint (you need an adequate model to build) |
| Denominator | \(1 + \mathcal{D}_{KL} + \Gamma\) | Science — the art of knowing | Drive \(\mathcal{D}_{KL} \to 0\) about how reality works | The object itself |
A bridge, an engine, a compiler, a logistics network: these optimise conversion efficiency. CERN, a telescope, a clinical trial: these spend enormous \(P_{in}\eta\) (a collider burns gigawatts) for the sole purpose of shrinking the gap to fundamental reality. Engineering treats truth as a constraint and efficiency as the goal; basic science treats power as a constraint and truth as the goal.
They are coupled, not disjoint. Every engineering advance lowers some \(\mathcal{D}_{KL}\) (you learn what breaks); every experiment spends numerator to run. The distinction is which factor the activity optimises. And \(\mathcal{R}\) climbs only when both move: a civilisation that funds only engineering gets efficient at the wrong thing while hidden \(\mathcal{D}_{KL}\) accumulates; one that funds only science starves the numerator that pays for the instruments. Persistence needs the product.
The social domains look like different subjects, but they are the same mechanism: a contest in which power is spent to resolve competing models against reality, and persistence flows to whoever carried less delusion. That is exactly a proper scoring rule.
The kernel is the log score (see glossary.md, ips_economics.md). Per resolved event, a participant’s change in standing is
the nats by which private belief \(q\) beat the field \(p_m\) on realised outcome \(y\). Summed over events this is \(-\Delta\mathcal{D}_{KL}\): you gain exactly what you de-delude. Because the log score is strictly proper, the unique standing-maximising strategy is to report calibrated beliefs — honest truth extraction is the dominant strategy, not a virtue bolted on afterward.
Read each domain as one instantiation of this single rule:
| Domain | The contest | Resolution authority | Who gains persistence |
|---|---|---|---|
| Psychology | self-model vs. lived outcome | one’s own resolved experience | the agent who pays to erase the wrong bit — acceptance is the loss of a delusion; grief is \(\mathcal{D}_{KL}\) collapsing |
| Sociology | shared narrative \(Q\) vs. collective experience | community memory, ritual, truth commission | the group that updates its shared model rather than scapegoating |
| Economics | priced distribution vs. realised fundamentals | settlement, earnings, default | calibrated capital extracts from deluded capital (Kelly–KL) |
| Geopolitics | rival state models vs. the battlefield | force, attrition, the ground itself | the polity whose model was less deluded — Crimea is two ownership-models scored by war |
| Science | competing theories vs. nature | reproducible measurement (CERN) | the theory that predicts; nobody owns the resolver |
The decisive variable is how capturable the resolution authority is. A captured resolver — a dictator scoring his own forecasts, a regulator owned by insiders, a court that takes bribes — recreates the dictatorship pathology from ips_geopolitics.md §3: \(\mathcal{D}_{KL}\) ratchets upward while the official score insists all is well, until the gap closes by force. Physics yields the most durable truth precisely because its resolution authority — reality — cannot be bribed, vetoed, or out-voted. The trustworthiness of any extracted truth is inversely proportional to how easily its resolver can be captured.
Across all of this, humanity reached for one scalar to summarise how well are we doing: money. Markets are a proper scoring rule — price discovery genuinely extracts truth about value — which is why money proxies success better than any earlier metric.
But fiat clears desire (who pays next), not calibration (who was right) — see glossary.md: Currency / fiat and Accounting leak. So money is a leaky measure of truth extraction:
The leak is precisely the gap between the numerator a balance buys and the \(\mathcal{D}_{KL}\) it never paid. The trustworthiness token closes the leak by construction: it is money that can only be earned by lowering \(\mathcal{D}_{KL}\),
forecast quality net of compute cost. Where fiat measures the numerator’s claim, the trustworthiness token measures the denominator’s repair — it is the explicit, non-leaky metric of how much truth a node has extracted.
Truth extraction that is not recorded must be re-paid: each generation re-collides with reality, re-deludes, re-grieves, and re-learns the same lessons. The civilisational move is to persist the record — not just the conclusions but the resolved forecasts, the bets, the misses, and the updates — so that how truth was formed and how trustworthiness was restored to cooperation is itself preserved.
| Layer of the record | What it stores | Lifetime | Failure |
|---|---|---|---|
| Agent | episodic memory, journals | dies with the node | unrecorded; relearned next generation |
| Organisation | post-mortems, learnings/ |
survives staff turnover if written | KPI theatre overwrites the misses |
| Society | science literature, law, truth commissions | generations | capturable resolver rewrites history |
| PoT chain | signed, scored, append-only forecasts | survives operators | resolution rule fixed in protocol |
This is what Proof of Trust is: a git-native blockchain whose blocks are signed probabilistic forecasts scored by \(\mathcal{D}_{KL}\) against realised outcomes, where fork weight accumulates literal bits of information rather than hash work. Read as history, it is a ledger of truth extraction: every entry is a delusion paid down, attributed and dated, answering what did we believe, what happened, what did we learn, and whose model earned standing by being right? That is the recorded epistemic credit of cooperation itself (ips_ontology.md, predictive_governance_and_pot.md).
The ultimate persistence test is the cosmic horizon. Ten billion years is a stellar timescale — longer than the Earth has left, on the order of the remaining main-sequence life of sun-like stars. Asking whether a truth-extraction ledger can persist that long is asking for \(\mathcal{R} \ge 1\) sustained across substrate migrations that no single species or planet survives. It is not a utopian wish; it is the FPE evaluated at the longest horizon, and it fixes a concrete design spec:
| Term | What a \(10^{10}\)-year ledger requires |
|---|---|
| \(\mathcal{D}_{KL}\) | Never stop extracting truth. A frozen model accumulates delusion as reality drifts (expansion, heat death, new physics); calibration must be continuous, forever. |
| \(\Gamma\) | No permanent schism. A chain that hoards unresolved forks ossifies; friction must be resolved as fast as it forms. |
| \(\Phi\) | Migratable substrate. Whatever runs the resolver — biology, silicon, successors — must be replaceable without losing the chain (fractal composition, no single critical fab). |
| \(\Psi\) | An uncapturable resolver. The authority must remain reality, never an entity that can be out-powered. |
The last row is where the Anti-Explosion theorem bites: the optimal footprint \(f^*\) that maximises \(\mathcal{R}\) lies below unity. A ledger that tried to coordinate the whole universe under one Markov blanket would lower its own \(\mathcal{R}\) and be forced to equilibrate — like a polity, not a singularity. A record that lasts ten billion years cannot be an empire of truth; it must stay a federation that keeps reality, not itself, as judge.
So the \(10^{10}\)-year question is the same four operations every essay names, pushed to the limit where they become the blueprint for a civilisation that wants to remember what it learned: never stop extracting truth, never stop resolving friction, keep the substrate fungible, keep the resolver uncapturable. This is the auditable instance of the book’s closing image — the cosmos building ever-better approximations of itself — realised at the scale of cooperating minds.
For any human activity, ask: