Type inference for plain JavaScript to enable static tier compilation, and broader JS semantics compatibility.

Hi, author of [JS2](github.com/loopdive/js2 here. Same architecture as scriptc as far as I can tell: tsc for parsing and type checking, a typed IR in the middle, swappable backends (we lower to WasmGC and linear-memory wasm). Fun to see how many of the same decisions you ended up making.

A couple of things we do differently. We also take plain JavaScript as input, with type inference to recover static types where there are no annotations. And we aim at full JS semantics rather than rejecting what can't be typed — in standalone mode (pure wasm, no JS host, so roughly comparable to your static tier) we pass about 22.6k of test262, tracked publicly.

The IR itself is work in progress — we're migrating from direct AST-to-wasm codegen onto a typed SSA IR node kind by node kind, with backend-agnosticism as the design goal.

The direction might be relevant to scriptc: untyped code currently goes to your quickjs tier, and with inference some of it could land in the static tier instead. It wouldn't touch the "nothing is silently miscompiled" rule — inferred types could go through the same runtime validation you already do for `any` at the boundary.

Would you be open to a small experiment at some point, say compiling one inferred-JS function through your IR? The main thing that needs is the IR/serializer being documented enough to target. Either way, happy to compare notes.