jco leg is a toolchain generation behind the siblings (registry jco-transpile ^0.5.2, transpile-time --map wiring, floating shim pins)

[lann]

The conformance jco leg predates the websocket/webrtc pattern and has drifted from it on three axes (conformance/driver-ct/jco/):

1. Transpiler: registry @bytecodealliance/jco-transpile: ^0.5.2, while websocket and webrtc pin the lann/jco fork release tarball (jco-transpile-0.6.2-fa6990d) that carries the p3/JSPI fixes the family keeps hitting (-I async: future.transfer trampoline const is emitted after the init generator that references it (TDZ ReferenceError at instantiation) lann/jco#34/dx: development-friction debt — shared core crate, generated jco flags, guest wrappers, harness cleanups #35 reproduce on 0.6.3 upstream too).
2. Wiring style: all interface wiring happens at transpile time via a wall of --map flags in package.json scripts (SUT host, test-context, per-interface preview2-shim paths), vs the siblings' -I async instantiation + runtime bindImports — one transpile, imports bound at instantiate time. Two styles is one more thing to know per repo, and the --map copies are the least reviewable text in the tree.
3. Pinning philosophy: ^0.5.2 / preview2-shim ^0.17.4 / playwright-core ^1.61.1 float on semver ranges in the same file whose component-test pin is rev-exact by policy.

Migrate to the sibling shape: fork tarball, -I async, runtime import binding (via the shared helpers once polymorph-components/polymorph-test#58 lands), exact pins. Risk is concentrated in the browser leg (worker pool + import-map plumbing) — gate on the full matrix including jco-browser, not just jco-node.

[lann]

PR #347 carries the modernization with one deliberate residual: js/jco's types generation stays on exact jco-transpile@0.5.2. The fork (0.6.2-fa6990d) types async-import stream parameters as AsyncIterable<number> where webcrypto.js is written against ReadableStream — just jco::typecheck fails against fork-generated definitions. Runtime paths all carry the fork (the full conformance matrix incl. jco-browser proves parity); retyping the host's stream handling is its own change and should ride whichever of the two shapes upstream jco settles on. Re-pin the types tree when that lands.