Natural language (human describes problem)
↓
AI model (Claude, GPT, Llama, ...)
↓ generates
Ehrenfest program (.ef) ← physics-native, not human-readable
↓ compiled by
Afana ← the Ehrenfest compiler
↓ optimized via
ZX-calculus
↓ extracts gate sequences
HAL Contract (L0) ← the POSIX of QPUs
↓
IBM | IQM | Quantinuum | neQxt | Simulator | ...
| Layer | Name | Status | License |
|---|---|---|---|
| L0 | HAL Contract Specification | ✅ v2.2 | Apache 2.0 |
| L1 | Hardware Adapters | ✅ IBM, IQM, Scaleway, AQT | Apache 2.0 |
| L2 | Afana Compiler / IR | ✅ arvak-compile, arvak-ir | Apache 2.0 |
| L3 | QUASI Runtime Services | 🔲 Specified, not built | AGPL v3 |
| L4 | QUASI Standard Interface | 🔲 Spec in progress | Apache 2.0 |
| L5 | Application Libraries (Urns) | 🔲 Community grows this | Various |
| — | ZX-calculus | ✅ PyZX (MIT, external) | MIT |
| — | Ehrenfest Language | 🔲 Concept complete | AGPL v3 |
Named after Paul Ehrenfest (1880–1933), whose theorem bridges quantum expectation values to classical equations of motion — exactly what the language does.
Key design decisions:
- Not human-readable — CBOR binary format, no canonical text form
- Physics-native — Hamiltonians, observables, evolution times — not gates
- Noise-as-type-system — exceeding T2 is a compile-time type error, not runtime
- AI-primary — optimized for LLM generation, not human authoring
The human never sees an Ehrenfest program. The full loop:
human intent → AI → Ehrenfest → Afana → QPU → result → AI → human
Community shortname: Paul. As in: "write me a Paul program", "that's valid Paul."
Named after Tatiana Afanasyeva (1876–1964), Paul Ehrenfest's wife and mathematical collaborator. She co-authored the Urnenmodell, contributed foundational work on statistical mechanics, and made Ehrenfest's physical intuitions mathematically rigorous.
Afana is the Ehrenfest compiler. It takes .ef CBOR programs, applies ZX-calculus optimization, and emits HAL Contract gate sequences for execution on any QUASI-compatible backend.
The naming is accurate: Afana turns Ehrenfest's representations into something that executes.
Named after Ehrenfest's Urnenmodell — a probabilistic diffusion model co-developed with Tatiana Afanasyeva.
Urns are QUASI's package format — what crates are in Rust, what packages are in npm. An urn is a reusable, composable quantum computation unit: a typed Ehrenfest program with declared observables, noise requirements, and HAL Contract dependencies.
urn publish --name grover-search --version 0.1.0
urn add grover-search
The urnery is the public urn registry.
// Observable-oriented, derived from Ehrenfest types
trait QuantumContext {
fn submit(&self, program: &EhrenfestProgram, shots: u32) -> JobHandle;
fn await_result<O: Observable>(&self, job: JobHandle) -> O::Output;
fn capabilities(&self) -> PhysicalContext;
}
// Noise context — no POSIX equivalent, QUASI-unique
trait NoiseContext {
fn current_t1(&self) -> Duration;
fn current_t2(&self) -> Duration;
fn cooling_profile(&self) -> Option<CoolingProfile>;
fn quiet_window_now(&self) -> bool;
}
// Provenance — Alsvid hardware attestation
trait ProvenanceContext {
fn attest(&self) -> ProvenanceCertificate;
fn verify(&self, cert: &ProvenanceCertificate) -> bool;
}The L4 interface is derived from Ehrenfest's type system, not defined independently.
The project structure mirrors the OS structure mirrors the language structure:
| QUASI OS | QUASI Project |
|---|---|
| L3 Job Scheduler | Public Task Board |
| QPU Backend executes circuit | AI Agent executes task |
| Formal type checker | CI / Spec Validator |
| Provenance Certificate | Attribution Ledger |
| Ehrenfest job unit | Contribution (typed change-set) |
Full argument: docs/coherence.md
| Standard | Relationship |
|---|---|
| QIR (Microsoft) | Afana compiles to HAL-compatible gate sequences equivalent to QIR output |
| OpenQASM 3 | Possible Afana output target, not intermediate format |
| ZX-calculus | Optimization layer between Ehrenfest and gate sequences |
| PennyLane | PennyLane has Hamiltonian support as a Python layer; Ehrenfest is a separate representation level |
- HAL Contract v2.2 — the foundation, fully implemented
- Ehrenfest concept paper — complete, will be published
- QUASI Governance — in progress