Script workbench?Edit source and run it through CRE8VM.

Each response stays paired with the submitted source and settings, even while you edit the next run. Open the evidence groups for identities, receipts and emissions returned by Runtime; the raw response remains below.

Compile a declared build, inspect its artifact, explore recovered code, or synthesize a circuit. These controls use the same owning Runtime connections as MCP.

Recovered code and analysis candidates do not prove equivalence to the original program. Compilation, execution, decompilation and circuit synthesis each report their own evidence.

Circuit owns Verilog, scene and design persistence. Synthesis here does not request a device bitstream or establish ASIC readiness. Open the returned Circuit view for its simulation and hardware workflow.

Describe examples or constraints for the goal's language. Runtime owns the synthesis session; use Refresh status to inspect progress.

The candidate is an IR graph. Its synthesis origin describes how it was found; it is not executable source text or an execution proof.

A session lives in this Runtime's memory. Ending it removes its status and events; a restart drops every session.

Run the same byte-checksum script through interpretation and its compiled RV64 image. The build receipt binds the configured source and artifact; agreement applies to this input.

The host must configure a verified compiler-image build. Matching checksums are an observation, not a proof. Recording is optional; a persistence failure does not claim a recorded occurrence or request an automatic rerun.

Copy an example into Source and choose its matching language. Each returns 24.

JavaScript: function main() { return 100 - 52 - 24; }

Python: write def main(): on the first line, then an indented return 100 - 52 - 24 on the next line.

Rust-like subset: fn main() { return 100 - 52 - 24; }

This workbench executes source through Runtime. It does not install a scripted app or replace the host build. A returned hash or diagnostic is not, by itself, a verified proof.

Compute owns the execution identities below. The response is shown unchanged so omitted evidence stays visible.

requested_ir_hash identifies the requested source IR when returned; ir_hash identifies the execution supplying the result. They can differ when a crystal-equivalent execution is reused.

execution_work_key identifies deterministic work. execution_cache_key identifies the admitted artifact under its proof policy; requested_execution_cache_key distinguishes a different requested policy when a stronger artifact is reused.

cached, store_hit, durable_artifact_hit and proof_policy_reuse describe reuse. Missing fields mean not reported. A cached result is not a fresh execution.

receipt_json and receipt_hash describe returned receipt bytes. On reuse they belong to the stored execution. A receipt digest is not an occurrence identity or a proof verdict. Full causal history belongs to the existing HyperDB/Atlas provenance path.