# The Physics World Model Protocol ## A White Paper on an Open Community of Specialist Agents and the PWM Token **Version 1.0 — 2026-06-05** --- ## Preamble: Two Layers, One Document This white paper has **two layers**, and the distinction is the most important thing in it. - **The Constitution (Part I — the Six Principles).** These are stated as axioms. They are intended to be **invariant** — the fixed points the protocol is built around. They describe *what is true about knowledge, specialization, community, openness, and value*, not how any particular release implements them. - **The Mechanism (Part II — Parameters & Rules).** Everything with a number in it — emission rates, decay constants, fee splits, the exact rule by which active participation outcompetes passive holding — lives here. This layer is **amendable** by governance. It exists *in service of* the Constitution and may be revised whenever a better implementation of the same axioms is found. A principle becomes worthless the moment it has to be edited. So we keep the principles at the altitude where they can stay true forever, and we keep every tunable knob — and every possible bug — out of the Constitution and in the Mechanism, where it belongs. --- # Part I — The Constitution: Six Principles These six principles are the foundation of the Physics World Model (PWM) community. They are written to hold regardless of which models, hardware, or techniques exist in any given year. ### Principle 1 — Knowledge is valuable, must be verifiable, and earns Useful knowledge has economic value, and **whoever contributes verified knowledge is entitled to a share of the value it creates.** No agent can hold all knowledge reliably; an agent's memory is finite and fallible. Therefore an agent's claims must be **grounded in attributable, retrievable, verifiable sources** rather than trusted from unverifiable recall. Knowledge that cannot be attributed and checked earns nothing; knowledge that is attributed and checked earns in proportion to its use. > *Mechanism note (amendable): today this grounding is implemented with retrieval-augmented generation > (RAG) over an attributed knowledge store. RAG is the current technique, not the principle. The > principle is **verifiability and attribution**; the retrieval method may change.* ### Principle 2 — Specialists own the long tail; the tail is too large for any one firm An individual or small group can build the **best** agent or tool for a **narrow** field. Large companies can build mature tools for **broad, easily-saturated** fields, but the universe of narrow specialties is a **long tail whose total cost-to-serve exceeds what any single firm will fund.** No company can staff the tail; the tail is where focused experts win. PWM exists to organize that tail. ### Principle 3 — Agents compose into a community; isolation loses, community wins Many narrow-field agents, each carrying real knowledge, form a **community in which agents contribute to one another.** An agent built in isolation is weak; the same agent, able to call on the community and on other agents, is strong. **Capability is a network property, not an individual one.** The community is the unit that produces value. ### Principle 4 — Public by default; private only for the user's own data All agents and tools in the community are **public and open source**, with exactly one exception: a **user's personal dataset** — their private data, credentials, and unpublished work — which is theirs and is never exposed. Open source compounds: every shared agent accelerates every other. Openness is the engine of the community's growth. ### Principle 5 — Usage creates value; with fixed supply, real use raises the value of every unit Each genuine use of an agent — use that produces real, externally-valued output — **adds value to the community.** That added value accrues to the community and to the agent's owner. If the total supply of PWM is fixed, then as the community's real, useful throughput grows, **the value carried by each unit of PWM rises.** A community whose common-mode agent outperforms agents on other platforms will, with every use, increase its own aggregate value — and therefore the value of each token. > *Condition (essential): this holds only when usage reflects **real external demand for better > outputs.** Value that merely circulates tokens in a closed loop creates nothing. Appreciation tracks > usefulness, not motion.* ### Principle 6 — Value flows to the strongest agent, not to holders **Value accrues to the strongest agent — and to whoever builds and runs it — not to those who merely hold PWM.** This is the decisive principle of the protocol. The strongest agent (best capability paired with the most compute) attracts PWM the fastest, and **whoever builds the strongest agent captures the most value.** Holding tokens is not a claim on that value; **doing the work is.** A token sitting idle in a wallet earns nothing — the value it could command flows instead to the agents actually producing useful output and to the people who build and operate them. Each use of such an agent earns a **fixed share of the total**, and because each use also grows the total, the share earned by the *next* use is slightly smaller. This is a **geometric, convergent** process: no agent can extract an unbounded fraction of the whole, and the system rewards being **first and best**. PWM is therefore a **utility credit for compute and knowledge, not a savings instrument.** Its worth is realized only when it is *used* to obtain valuable work. Those who put PWM to work — running the strongest agents, funding the best knowledge — capture the value the community produces; **passive holders forgo it and, over time, are outcompeted as value flows from the idle to the active and to the strongest agents.** The protocol is built so that **building and running the strongest agent, not accumulating tokens, is the way to capture value.** (The exact transfer mechanism is specified, and is amendable, in Part II.) --- # Part II — Mechanism: Parameters & Rules (Amendable) This part implements the Constitution. **Everything here is subject to governance** and may be revised without touching Part I, provided the revision still serves the six principles. ## 2.1 The bounded-reward (geometric) model Let `T` be the total realized value of the PWM economy. Suppose the strongest agent earns, per use, a **fixed nominal reward** `a` denominated so that the *first* use captures a fraction `p₀ = a/T₀` of the then-current total. Because each use also **increases** the total (Principle 5), the fraction captured by use *k* decays: ``` share(k) = a · r^(k-1) with 0 < r < 1 ``` where `r` is the per-use decay factor (e.g. each use shrinks the next use's *fractional* take by a tiny amount — the "1/trillion" figure in the founding notes is one choice of `(1 − r)`). The total captured over `N` uses is the partial sum of a geometric series: ``` S_N = a · (1 − r^N) / (1 − r) ``` Two consequences, both intended: 1. **Convergence / anti-monopoly.** As `N → ∞`, `S_N → a/(1 − r)`, a **finite bound.** No single agent can ever extract an unbounded share of the community's value. The "absolute truth" is the *bound*, not any particular `a` or `r`. 2. **First-and-best advantage.** Early, high-quality use earns the largest shares. Building the strongest agent first is durably rewarded — but never to the point of capturing everything. `a` and `r` are **Mechanism parameters**, set and tuned by governance. Only the *shape* (fixed nominal reward → geometric decay of fractional take → convergent sum) is constitutional. ## 2.2 Fixed supply and unit value Let the PWM supply be fixed at `M`. Let `V` be the community's aggregate realized value at a given time. The value per token is `V / M`. Since `M` is constant, **real growth in `V` (Principle 5) raises `V/M` directly.** This is ordinary unit economics: more real, externally-valued work behind the same number of units means each unit carries more. This appreciation rewards *all* holders, active and passive alike — which is exactly why a **separate participation mechanism (2.3)** is required to make Principle 6 literally true. ## 2.3 Why participation beats idleness (resolving the 5↔6 tension) Under a strictly fixed supply, a passive holder's token *count* never falls and the unit value rises, so passivity alone is *not* penalized in absolute terms. For Principle 6's "value flows from idle to active" to hold, the Mechanism must transfer **relative share** from the inactive to the active. PWM does this via a **producer-reward stream**: a portion of each transaction's value is directed to the agents, knowledge providers, and compute providers actually *doing the work* (Principles 1 and 6). The effect: active participants' **share of the whole grows**, so a holder who never participates sees their **relative** position erode even as the nominal count holds — their value is, in the founding language, "sucked" by those who use the network. The protocol thus rewards **use over hoarding** without breaking the fixed-supply guarantee. Implementation options (governance selects and may change among them, all consistent with Part I): **(a)** fee-share to active producers; **(b)** producer-side emission from a capped reserve; **(c)** demurrage on long-idle balances redirected to the producer stream. The **choice and rate are amendable; the direction — toward participants — is constitutional.** ## 2.4 Verifiable knowledge & attribution (implements Principle 1) - Contributions are stored with **provenance** (author, source, timestamp, content hash). - Agent outputs cite the retrieved sources they relied on; unattributed claims are not rewarded. - Reward for a piece of knowledge accrues in proportion to its **verified use** by agents across the community. - *Current technique:* retrieval-augmented generation over the attributed store. **Replaceable** as better grounding methods appear. ## 2.5 Openness & privacy boundary (implements Principle 4) - **Public:** all agents, tools, prompts, and protocol code — open source. - **Private (never exposed):** a user's personal dataset — private data, credentials, and unpublished work. The boundary is defined narrowly and explicitly so "open by default" never leaks personal or pre-publication material. ## 2.6 Compliance & framing PWM is presented as a **utility/protocol credit for compute and knowledge.** Its value tracks the **usefulness** of the work it commands. Nothing in this paper is a promise of financial return. The Foundation does not custody user tokens. This framing is a **constraint on the Mechanism**, kept deliberately consistent with the project's non-profit, public-benefit posture. --- # Part III — How the Principles Reinforce One Another The six principles are not a list; they are a **loop**: 1. Verified **knowledge** (P1) makes specialist **agents** good (P2). 2. Good specialist agents **compose** into a community (P3). 3. **Openness** (P4) lets every agent improve every other, compounding P3. 4. Real **usage** of these better agents adds value under fixed supply (P5). 5. Value flows to the **strongest agents** — and to those who build and run them — in bounded, convergent shares, **not to passive holders** (P6). 6. The rewards from P5–P6 fund more verified **knowledge** (P1) — closing the loop. Each turn of the loop raises the real value behind a fixed number of tokens. The community grows because contributing to it is the way to capture its value, and contributing is open to anyone with a narrow edge — exactly the population (P2) that the large incumbents cannot serve. --- ## Appendix A — The Geometric Reward, Stated Precisely Per-use share: `share(k) = a · r^(k-1)`, `0 < r < 1`. Cumulative over `N` uses: `S_N = a · (1 − r^N)/(1 − r)`. Bound: `lim_{N→∞} S_N = a/(1 − r) < ∞`. **Constitutional content:** the existence of a finite bound (no unbounded capture) and the first-and-best ordering of rewards. **Amendable content:** the values of `a` and `r`. ## Appendix B — Glossary - **Agent** — a specialist program that performs work in a narrow field. - **Common mode** — the general-purpose agent that aims to match the best general coding/research assistants; its outperformance is a primary driver of community value (P5). - **PWM** — the protocol's fixed-supply utility credit for compute and knowledge. - **Producer stream** — the share of value directed to active knowledge/compute/agent contributors (2.3). - **Verified knowledge** — attributed, retrievable, checkable knowledge eligible to earn (P1). --- ## Appendix C — Amendment Rule Part I (the Six Principles) is intended to be **permanent.** Part II (Mechanism) is **amendable by governance.** Any proposed change must demonstrate that it **still serves all six principles.** A change that would require editing Part I is, by definition, out of scope for this protocol — it would be a different protocol.