# Open-source EDA handoff evidence

This is an original comparison of 20 tools across 20 attributes, checked on 24 October 2026 (America/Denver). The UTC retrieval timestamps fall on 25 October. It compares documented handoffs across simulation, synthesis, formal verification, physical implementation, layout and analog simulation. It does not report successful cross-tool execution.

## Selection and version boundary

The accepted roster was retained except that ngspice replaces Edalize so analog simulation has a concrete entry. FuseSoC still represents build orchestration. The inventory is not a ranking, a complete market census or an endorsement.

Nineteen tools use the exact default-branch commit observed by a read-only Git ref lookup. Those hashes identify source snapshots, not stable releases or the versions installed for the separate simulator experiment. For ngspice, the GitHub mirror was visibly stale. Its record instead uses the current official SourceForge project documentation and the **47+ development manual dated 28 September 2026**. The manual edition is not presented as a stable release, and no current ngspice source commit was verified.

## How fields were compiled

Each non-null comparison cell has `field_evidence` with opened primary-source IDs and locators. Repository URLs resolve to fixed commits. The source ledger records the requested URL, actual HTTP result, observation time, source-body hash and byte length. Git ref observations are recorded separately. Source bodies remain outside the distributed package.

The attributes distinguish input language, input representation, interchange output, execution model, verification interfaces, artifacts, dependencies, configuration and licensing. Stage and format filters are concise derivatives of the row evidence. A format name means the inspected documentation describes that use; it does not establish version compatibility, semantic equivalence, a working PDK flow or a validated downstream import.

A null means **not verified in the selected sources**, not absent. Some cells document a limited fact while explicitly leaving another part unknown—for example, a report artifact without a verified exit-code contract. No tool was installed or run for D1. Actual Icarus/Verilator fixture results belong to D3 and cannot be generalized to these 20 tools.

## Existing references and what this adds

Three inspected reference types are the [OpenROAD project flow description](https://github.com/The-OpenROAD-Project/OpenROAD), [FuseSoC tool/backend introduction](https://fusesoc.readthedocs.io/en/stable/user/introduction.html), and [Yosys project documentation](https://github.com/YosysHQ/yosys). Each documents its own project or flow. This compilation adds a shared, dated 20-column schema across independently documented simulation, synthesis, layout, FPGA/ASIC implementation and analog tools, with cell-specific source links and explicit unknowns. It is not a recut of one canonical source.

## Licence scope

The original compilation is **CC BY 4.0**. Attribute it as “Open-source EDA handoff evidence — original compilation, 2026-10-24” and link to this method. Upstream code, manuals, models, PDKs and licence texts retain their own terms; none is redistributed here.

Several labels deliberately preserve scope differences: KLayout’s project introduction says GPL-2.0-or-later while its repository supplies GPLv3 text; Netgen’s `Copying` contains GPLv1 text without a completed current per-file licence inventory; Magic carries a University of California permissive notice; ngspice explicitly retains exceptions to its modified-BSD description. Verilator’s dual-licence declaration, NVC’s library exceptions, and separately licensed solver/PDK dependencies are not flattened into a universal licence grant. This dataset is not a legal clearance opinion.

## Recheck and integrity

Follow the per-cell pinned links, compare the original file to its ledger hash, and record new revisions as a new snapshot. Do not replace unknowns with “no” merely because an overview omits a feature. `hashes.txt` covers the asset files except itself and is refreshed after CSV export. CSV is a flattened view; JSON preserves nulls, source IDs and provenance. No provider-side changes, target-site visits or old magazine material were used.