Start here · the whole loop

Get PWM first, then put it to work with AI4Science

Two minutes, three parts. First earn a starter balance of PWM right here on the website, then install the AI4Science agent and spend it — earning more back through feedback as you go. We'll use one real example the whole way through: the CASSI principle.

🔬

Our running example — the CASSI principle (L1-003)

Coded Aperture Snapshot Spectral Imaging is a Principle (Layer 1): a physics fingerprint for capturing a full hyperspectral cube in a single snapshot. Every step below acts on this one artifact, so you can follow along on its live page: /artifact/l1/L1-003 →

1

Get PWM first — on physicsworldmodel.org

PWM is earned by contributing, never sold. You bootstrap a starter balance here in the browser — no install needed.

a

Sign in (your wallet comes with it)

Sign in with Google/email — that gives you a non-custodial embedded wallet automatically (no seed phrase) — or connect your own wallet (MetaMask, Coinbase). Either way, one wallet = your identity + where PWM is credited.

b

Open the CASSI principle

Go to the CASSI principle page (or browse all principles at /principles). Read the artifact, then scroll to the Contribute panel at the bottom — it has two ways to earn.

c

Easiest: "Submit without uploading" — a free check

One click. It takes a faithful snapshot of the CASSI principle and runs it through the automated quality gate. On acceptance you're credited ≈ 0.5 PWM within ~1 minute — recorded against your wallet. Allowed once per layer (Principle / Digital Twin / Benchmark / Solution), up to a small daily cap. Nothing to write.

d

More: "Submit by uploading" a real improvement

Upload a genuine modification (a .md + .json) — e.g. a sharper CASSI forward model or a corrected noise spec. A full S1–S4 Claude review checks soundness; an accepted upload earns a larger one-time bootstrap (≈ 2 PWM). Creating a brand-new artifact earns more than modifying one.

Exact amounts are set by the protocol's emission equation (see the white paper) — earlier and harder contributions are worth more. Stage 1 is an off-chain balance credited to your wallet address; it settles on-chain at mainnet. See How to get PWM for the full detail.

2

Use AI4Science — spend PWM to do science

With a balance in hand, install the agent and put it to work. It runs like Claude Code, then specializes on PWM's verified artifacts. Full intro →

a

Install & log in

Needs only Python ≥ 3.10 — no git, no Node. Pick your OS, then log in with the same wallet that holds your PWM.

🐧 Linux / 🍎 macOS

curl -fsSL https://raw.githubusercontent.com/integritynoble/AI4Science/main/install.sh | bash

🪟 Windows (PowerShell)

irm https://physicsworldmodel.org/install.ps1 | iex

▶ Then, on any OS

ai4science login

Open a new terminal after installing so ai4science is on your PATH. Not found? export PATH="$HOME/.local/bin:$PATH" (Linux/macOS).

b

Ask it about the CASSI principle

Start a session and work with the same artifact you just earned PWM on. Each agent turn spends a small amount of your PWM balance.

ai4science                                  # interactive session (like `claude`)
> explain the CASSI principle L1-003 and its forward model
> run a gap-TV reconstruction on the CASSI benchmark

It reads the on-chain registry, runs verified solutions, and can edit files (you confirm each change). /help for commands.

3

Earn PWM back — feedback inside AI4Science

Using the agent costs PWM; the easiest way to earn it back is feedback — right inside the chat session, no website round-trip.

a

Use the /feedback command

Spotted something about the CASSI workflow that should improve? Tell the agent — concrete, actionable feedback scores highest.

/feedback the CASSI judge should accept .h5 inputs too, not only .mat
b

Get rewarded instantly

An LLM judge scores your feedback's quality (0–1) and credits PWM to your balance on the spot — higher quality, higher reward. Low-effort or duplicate feedback scores zero, so it can't be farmed.

c

Or contribute tools & solutions — earn weekly

Beyond feedback, you can publish tools/sub-agents or submit solutions (e.g. a new CASSI reconstruction method). Whenever an agent uses them on a paid turn, you earn from the weekly emission pool — for as long as they stay useful. See how contributing earns PWM →

The whole loop, in one line

① Earn PWM on the site (CASSI principle) ② Spend it in AI4Science ③ Earn more via /feedback & contributions