# HDL simulator observations — method

Thirty original deterministic SystemVerilog fixtures were compiled from identical source bytes in Icarus Verilog 13.0 and Verilator 5.052. All 60 pipelines compiled and executed. Each built executable ran twice; all 60 OBS payloads repeated. The 120 executions are 60 primary observations plus reproducibility checks, not 120 independent fixtures. Exact UTC timestamps, commands, compiler diagnostics, return codes and literal runtime outputs are retained in the harness logs and canonical index.

## Configurations and interpretation

Icarus used `-g2012 -gstrict-expr-width -Wall`. Verilator used `--binary --timing --language 1800-2012 --x-assign 0 --x-initial 0 -Wno-fatal`, with output-volume flags recorded in each command. Warnings were retained, not suppressed; `-Wno-fatal` allows compilation to continue after them. Build output was quieted; runtime summary statistics were disabled to avoid presenting incidental execution speed as a measurement. The named OBS lines are the comparison payload. Full finish banners remain literal and are not counted as language differences.

Twenty fixtures have matching OBS payloads; ten differ. Eight differences concern unknown representation or initialization under the selected policies. F29 and F30 use seeded built-in random calls; each stream repeats within its own tool but differs across tools. F28 implements its own fixed 32-bit modulo sequence and matches. A fixed seed is not a promise of one portable algorithm across simulators.

The source files intentionally include width and unsigned examples that produce three Verilator warning-bearing compile records. No warning or differing output is labelled a tool defect or conformance failure. The dataset is an original bounded observation set, not an exhaustive language suite, security assessment, synthesis result or silicon measurement.

## Provenance and reproduction

The installed tools were reused from an isolated project cache of Homebrew arm64 bottles; no global upgrade occurred. Version commands were executed for this run. Original archive identifiers/hashes and actual executable hashes are in `runtime-provenance.json`. A first setup attempt failed because a symlink resolved into a path containing spaces; GNU Make refused that build location. The successful executions used actual temporary directories without spaces. Setup failure is separate from a compiler result on a fixture.

`harness/hdl-simulator-observations/fixtures/` contains all thirty source files. Five extend earlier original project-authored concepts and were rerun; twenty-five were newly authored. No former-site article, circuit, figure or third-party test suite supplied the code. Source hashes prove that both tools received identical bytes. The original runner accepts executable paths and an optional Icarus library binding, creates separate temporary build directories, runs each binary twice and writes its own result directory. It does not install tools or make network requests. Configure `VERILATOR_ROOT` as required by the installed distribution.

The `${ICARUS_LIB}` token in published argv arrays replaces only the local installation binding supplied with `-B` and `-M`; it does not change source, diagnostics or outputs. Compiler diagnostics and runtime outputs are literal. Generated C++ and binaries are omitted; rerunning regenerates them. Only source code, input metadata and captured outputs are delivered.

## Rights and limits

Original fixtures and reproduction code: MIT. Original measured compilation and explanatory metadata: CC BY 4.0, attribution “HDL simulator observations — original compilation, 2026-10-24”. Tool and documentation rights remain separate. Documentation explains selected settings; it is not substituted for execution. Neither the dataset nor page examples establish metastability reliability, power-aware simulation, physical connectivity or commercial verification support.