From gc-arena to Stock CLIF: A Two-Year Honesty Arc

Every language project has a runtime story. Most of them are lies — “designed from first principles, implemented in six months, stable ever since.” Beskid’s runtime story is not a straight line. It is an honesty arc: a series of decisions where we chose the less wrong path, documented what we learned, and kept moving.

This post is the index. The full story lives in four phase posts:

• Phase 1: gc-arena (February 2026) — The research spike that asked whether Rust arena allocators could become a GC. The answer was “probably, but not yet” — and the pattern of honest retreat became the project’s default stance.
• Phase 2: Rust Handlers (March–June 2026) — 34 frozen kernel exports, 77 soft dispatch ops, hand-written Rust handlers threaded through a registration chain across every crate. It worked. For a while.
• Phase 3: The Silent Gap (June 2026) — A corelib test passed against a handler that was correct by the old spec but wrong by the new one. The CI gate was green. The behavior was silently wrong. The Rust-first runtime died that day.
• Phase 4: ISLE Shift (July 2026) — The boundary change: ISLE rules emit stock Cranelift CLIF, verified by Cranelift’s own verifier. From trust to verify. A correctness contract, not a runtime.

The arc continues. Cohort 1 is in progress. The migration can stop at any boundary — Rust fallbacks remain. Not “we know the answer.” Not “the runtime is done.” Just: here is what we tried, here is what we learned, here is where we are going, and here is the verifier that will tell us if we’re wrong.

Provenance

Approved ISLE runtime-port design · ABI builtins authority · runtime registration authority