Integrating an existing functional benchmark

SV-Gap can preserve a benchmark's existing functional verdict instead of rerunning its testbench. The imported result is bound to the exact candidate source set by a canonical digest and is then carried into the layered SV-Gap report.

1. Declare the candidate and production intent

Create a normal candidate manifest. Replace functional.commands with a relative import:

[functional]
import = "functional-result.json"

Clock, reset, and crossing declarations describe the production property being evaluated. They are not inferred from the imported verdict.

2. Bind the verdict to the candidate

Compute the digest after the candidate sources are final:

svgap digest path/to/manifest.toml

Create functional-result.json using schemas/functional-result-v1.json:

{
  "schema_version": "1.0",
  "status": "pass",
  "producer": "benchmark-name/revision",
  "candidate_digest": "sha256:...",
  "tool_versions": {"simulator": "version"},
  "evidence": {"tests_passed": 100, "tests_total": 100}
}

producer should identify the benchmark adapter and revision, not make an unsupported claim about who generated the RTL. evidence may include a public result URL, run identifier, formal proof digest, or normalized score fields.

3. Add the structural layer

svgap check path/to/manifest.toml

SV-Gap verifies the candidate digest before accepting the imported verdict. A modified source file turns the functional layer into tool_error until the upstream result is regenerated or explicitly rebound.

4. Export CI evidence

svgap export path/to/build/report.json \
  --sarif build/svgap.sarif \
  --html build/svgap.html

The working example under examples/imported_result uses this flow without executing a functional command.

Adapter policy

An adapter should preserve the upstream result's semantics. It must not turn a timeout, missing test, compilation failure, or unavailable result into pass. When the upstream status cannot be mapped safely, use unknown or tool_error and retain the original evidence.