Planned in the conformance strategy (docs/architecture.md §11): generate components with wasm-smith, run them under deltic and under native wasmtime, compare observable behavior.
Rationale: the consumer exams proved that real workloads surface defect classes no WAST corpus expresses (five distinct jco defect classes; two deltic runtime defects found by smoke-c0 alone). Fuzzing extends that coverage systematically, and — because our translation frontend is wasmtime-environ — differential runs against native wasmtime specifically probe the TS runtime layer (scheduler, lift/lower, resource tables) rather than re-testing the shared frontend.
Prereq: decide harness shape (component-level wasm-smith support, driver, behavior-comparison format). Best after the wasmtime pin bump (#1) so both sides run the same environ version.
Planned in the conformance strategy (docs/architecture.md §11): generate components with
wasm-smith, run them under deltic and under native wasmtime, compare observable behavior.Rationale: the consumer exams proved that real workloads surface defect classes no WAST corpus expresses (five distinct jco defect classes; two deltic runtime defects found by smoke-c0 alone). Fuzzing extends that coverage systematically, and — because our translation frontend is wasmtime-environ — differential runs against native wasmtime specifically probe the TS runtime layer (scheduler, lift/lower, resource tables) rather than re-testing the shared frontend.
Prereq: decide harness shape (component-level wasm-smith support, driver, behavior-comparison format). Best after the wasmtime pin bump (#1) so both sides run the same environ version.