☰ Documentation menu
Introduction
What is Provenova? Getting started
Concepts & workflows
Core concepts Capturing runs Reproduce & drift Result cards & badges Compliance & attestations Corpus & leaderboard
Using the app
Product tour (every page & button) Pricing FAQ
Libraries & reference
Libraries & downloads CLI reference API reference Frameworks reference Open schemas (qlprov) Deployment & self-hosting Licensing
Concepts & workflows

Corpus & leaderboard

Beyond your own runs, Provenova maintains a public calibration corpus: a cross-vendor, longitudinal record of how real quantum hardware behaves over time. It powers the State of Quantum Hardware leaderboard at /leaderboard.

The corpus

The corpus is a collection of calibration snapshots normalized to the open qlprov/calibration/1.0 schema, so an IBM device and an IonQ device can be compared on the same axes. Snapshots are content-addressed, so identical captures are stored once and referenced many times.

The crawler

The provenova-crawler package collects vendor calibration data, normalizes it, applies each vendor's terms-of-service redistribution policy, deduplicates, and ingests it into the corpus. It has two sources:

  • FixtureSource (default, offline) — reads representative payloads from fixtures/{vendor}/*.json. This is what the demo and tests use; no network required.
  • LiveSource (optional) — a skeleton for calling real vendor APIs (IBM BackendProperties, IonQ characterization, Amazon Braket device properties).

A ToS gate governs what may be redistributed, so the public corpus respects each vendor's terms.

Sources & licensing

The corpus is genuinely multi-vendor and every row is labelled by source and licence so mixed provenance is transparent:

  • IBM — real device calibration snapshots from Qiskit's fake_provider (Apache-2.0).
  • Metriq — community benchmark submissions (EPLG, CLOPS, BSEQ, QFT accuracy, QAOA approximation ratio) from metriq.info, CC-BY-4.0.
  • Zenodo — a ready slot for openly-licensed raw calibration datasets, loaded only when a record's licence is verified CC-BY / CC0 (none published for the tracked devices yet; IQM hardware currently appears via the Metriq and vendor-reported rows).
  • Vendor-reported — headline specifications published by manufacturers, shown as claims with a distinct badge and the source URL recorded in each snapshot's provenance — never presented as independently reproduced.

Nothing is fabricated: each value is copied from its cited source. The redistributable_raw flag and license_ref travel with every snapshot.

The leaderboard

/leaderboard ranks devices across vendors and switches the ranking metric. Alongside the calibration metrics (median two-qubit gate error, T1 / T2 coherence, best 2Q fidelity) it also ranks the cross-vendor benchmark metrics where published:

  • Algorithmic qubits (#AQ), Quantum Volume, CLOPS (higher is better),
  • 2-qubit gate fidelity (higher is better), EPLG (error per layered gate — lower is better),
  • BSEQ (Bell-state effective qubits), QFT accuracy, QAOA approximation ratio (all higher is better).

Devices that don't report the selected metric are dropped from that ranking, and each remaining row carries its source / licence badge.

Because snapshots are time-stamped, the corpus also exposes trends per device:

GET /api/v1/backends/<provider>/<backend_id>/trend

— a time series of a backend's calibration quality, so you can see a device improving or degrading over weeks and months.

Why it exists

The corpus turns one-off calibration captures into a shared, comparable, historical view of the quantum hardware fleet — the empirical backdrop against which your own runs and reproductions can be judged.

Next: Product tour.