Reproducibility and release references

HyperIso ships a release-level reference suite under reproducibility/. The suite is command-line based so that its inputs, commands and normalized outputs can be archived and compared without depending on a notebook or graphical session.

Reference cases

ID	Workflow
R1	SM B-sector Wilson coefficients at matching and low scales.
R2	Scalar B-sector Wilson coefficients.
R3	Inclusive and leptonic observables.
R4	A binned angular observable in a fixed interval.
R5	Fixed-seed Monte-Carlo uncertainty summary and accepted-sample CSV.
R6	THDM Wilson coefficients from an archived 2HDMC spectrum.
R7	SUSY Wilson coefficients from an archived SOFTSUSY spectrum.

The exact commands, tolerances and output filenames are defined in reproducibility/manifest.json.

Run the suite

After building hyperiso-ui in Release mode:

./reproducibility/scripts/run_cli_suite.sh

Use a non-standard executable with:

HYPERISO_BIN=/absolute/path/to/hyperiso-ui \
  ./reproducibility/scripts/run_cli_suite.sh

The checker normalizes startup banners and absolute paths where they are not part of the scientific result.

Statistical determinism

The R5 case uses:

• seed 123456;
• one Monte-Carlo thread;
• 200 accepted draws;
• a frozen set of output columns and numerical tolerances.

This is a release-to-release regression test, not a convergence claim for a physics analysis. Production studies should choose a draw count based on their own precision requirements.

Frozen provenance

Reference outputs are regenerated only for an intentional numerical change:

./reproducibility/scripts/run_cli_suite.sh --update-expected

The freeze records:

• the source commit used to build the executable;
• the executable SHA-256 hash;
• compiler, CMake, GSL, Eigen and Python versions;
• build type and platform information;
• per-case timings;
• hashes of the expected outputs.

Review every reference diff together with the physics or numerical change that caused it. The provenance commit should contain only the frozen expected-output files.

Reproducing external-model cases

The THDM and SUSY release checks consume archived spectrum files. They do not rerun 2HDMC or SOFTSUSY. This keeps the mandatory release gate independent of external-generator availability while retaining the numerical model input.

MARTY validation is documented separately and is not part of the mandatory frozen suite because it requires an external symbolic installation.

Release identity

For a published result, retain:

• the immutable Git tag and commit;
• the source archive and checksums;
• the exact input and override files;
• the software environment or wheel filename;
• the reference manifest and provenance metadata;
• the random seed and thread configuration.