Compiler & ISA

The NWO-ASM compiler lowers a single source program into the Process-Matrix IR, then routes each block through the MetaState optimizer to its best substrate target. Multi-stage by design: parse, lower, optimize, schedule, emit. Targets are pluggable — and writing the language itself is free, MIT-licensed, with no platform lock-in.

Compilation as a routed pipeline

Each stage produces a typed artifact handed to the next. The optimizer chooses substrate placement after IR lowering, using free-energy minimisation across substrate cost models.

Six pieces of the compiler

From source to QASM3 + USDC settlement

A minimal NWO-ASM program that scores an anomaly signal with the symbolic regressor, then lifts the residual to a 2-qubit process matrix the optimizer can route to a QPU when present, with a classical fallback. Per-call settlement is automatic.

# anomaly score on a 1-D signal, fit symbolic residual, route best path
module signal {
  input  s : f32[8192]
  output r : pmx[2]            # 2-qubit process matrix

  let v = anomaly.score(s)        # free-energy bound
  let e = eml.regress(s, v)       # closed-form residual
  r = pmx.lift(e)                  # causal-coherence channel

  route r via metastate {
    prefer: qpu                  # IBM · Origin Wukong · or Osaka ion-trap if available
    fallback: gpu               # otherwise nwo-agi distributed GPU
    settle: usdc(base)          # per-call, via splitter — no new contract
  }
}

# compile and inspect IR
nwo-asm build signal.nwo --emit pmx,qasm3
  → signal.pmx
  → signal.qasm3

# sign and dispatch (settles in USDC on Base)
nwo-asm dispatch signal.qasm3 --backend qpu --wallet $WALLET
  → dispatched to backend = qpu.osaka(iontrap_yb171) · proof = 0xPOI…
     (optimizer weighed qpu.ibm · qpu.origin · qpu.osaka; picked lowest free-energy)
  → settled $0.0008 USDC · split 35/35/30 · affiliate (if set) +15%

Core instruction set (stable)

What it's built on