Skip to content

PQID-Bench Documentation

Distribution status: PQID-Bench v1.0.0 frozen evidence release with pqid-bench v1.2.1 software tooling. The versioned benchmark archive is identified by 10.5281/zenodo.21649753.

PQID-Bench is an acquisition, reproduction, and replication toolkit plus a frozen evidence bundle for validation-aware quantum-program generation research. The package keeps eight activities separate:

  1. acquiring the authenticated compact benchmark distribution;
  2. checking release-file integrity;
  3. reproducing published metrics from archived evaluations;
  4. summarizing compatible evaluation records;
  5. comparing a candidate on an aligned prompt denominator;
  6. rendering an interactive report from frozen local evidence;
  7. collecting a new traceable model-response panel; and
  8. evaluating generated programs inside an isolated Docker worker.

Offline reproduction never contacts a provider. run-model does so only after explicit prompt-export acknowledgement. Only replay executes generated Python, and it requires a separate code-execution acknowledgement.

Choose A Starting Point

Goal Start here
Understand how GitHub, PyPI, Hugging Face, Zenodo, Docker, and OSF fit together Ecosystem And Study Roadmap
Download the ready-to-run benchmark Installation
Install and run the package for the first time Installation
Reproduce the paper quickly Reviewer Quickstart
Understand what each command can do Capability Guide
Test a new external model Live Model Testing
Look up a command or option CLI Reference
Print a readable report or export to R/pandas Reporting And Numerical Exports
Explore the frozen results interactively Interactive Explorer
Use the importable Python interface Python API
Run small end-to-end Python examples Worked Examples
Prepare or inspect JSON/JSONL records Data Dictionary
Understand ES-Gap, AS-Gap, and related metrics Metrics And Invariants
Follow an end-to-end task Workflows
Diagnose an error Troubleshooting
Check a term or abbreviation Glossary

Documentation Sets

The User Manual is the operational source of truth for installation, commands, records, metrics, workflows, and error handling.

The following documents have narrower purposes:

Ecosystem Boundaries

The complete ecosystem contains several related but non-interchangeable objects:

object responsibility
archived PQID dataset immutable upstream source dataset, DOI 10.5281/zenodo.20674853
PQID-Bench v1.0.0 archive immutable benchmark software and evidence release, DOI 10.5281/zenodo.21649753
benchmark authoring workspace canonical manuscript sources and complete research evidence
this PQID-Bench tree standalone package source, manual, schemas, tests, and selected frozen evidence
compact core ZIP ready-to-run records, splits, prompts, schemas, and isolated evaluator
frozen ZIP complete downloadable scientific evidence bundle
wheel installable command and Python interface, without the evidence bundle
source distribution package source and documentation, without large scientific artifacts
container archive optional isolated executable-replay worker
Hugging Face dataset adoption-focused records, splits, prompts, schemas, and compact core download
OSF PQID-Bench 2 registration immutable prospective protocol for a separate future semantic-validity study; contains no current benchmark result

The upstream dataset and benchmark release require separate citations and version identifiers. Generated distribution files are rebuilt from source and must not be hand-edited.

For a reader-facing explanation of these roles, including what OSF preregistration means and why PQID-Bench 2 is not a newer current-data release, see the Ecosystem And Study Roadmap.

Rendered manuscript outputs remain outside this public release. Their scientific inputs and regeneration scripts remain available; see Regenerating Publication Outputs. The package-facing Plotly explorer is generated only during the Pages build and is not inserted into the frozen evidence archive.

Version Dimensions

Dimension Identifier
Python package 1.2.1
benchmark release 1.0.0
evaluator pqid-bench-evaluator-1.1.0-safe-builtins
structural predicate pqid-bench-reference-signature-1.0.0-count-map
schema 1.0.0
artifact manifest 1.0.0
container artifact 1.0.0

These identifiers describe different contracts. A change to one must not be reported as though every other dimension changed with it. The replay container currently has local image ID sha256:849bf53e449fd618633199c0b622abeca270591dff248cd0bf3a0fd461abf2e2. It is published separately from the Python toolkit as ghcr.io/elias-abebe-gasparini/pqid-bench-evaluator:1.0.0. Its immutable OCI registry manifest digest is sha256:39825f5635cd6273e9e23c2848f2c88a2ff9d461e16a263fd89f22c6e664ac8f; it is distinct from the local image ID.

Documentation Site

The public manual is available at https://elias-abebe-gasparini.github.io/PQID-Bench/. Every source page is also readable as ordinary Markdown. Install the documentation extra to build the navigable site locally:

python -m pip install ".[docs]"
python scripts/build_pqid_bench_ecosystem_site.py
mkdocs serve

Use mkdocs build --strict when validating a documentation release.